Soares C G Et Al Maritime Engineering and Technology

MARITIME ENGINEERING AND TECHNOLOGY
PROCEEDINGS OF MARTECH 2011, 1st INTERNATIONAL CONFERENCE ON MARITIME

TECHNOLOGY AND ENGINEERING, LISBON, PORTUGAL, 1012 MAY 2011
Maritime Engineering and

Technology
Editors
C. Guedes Soares, Y. Garbatov & S. Sutulo
Centre for Marine Technology and Engineering (CENTEC), Instituto Superior Tcnico,
Technical University of Lisbon, Portugal
T.A. Santos
Ordem dos Engenheiros, Portugal
CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business
2012 Taylor & Francis Group, London, UK
Typeset by MPS Limited, Chennai, India
Printed and bound in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY.
All rights reserved. No part of this publication or the information contained herein may be
reproduced, stored in a retrieval system, or transmitted in any form or by any means,
electronic, mechanical, by photocopying, recording or otherwise, without written prior
permission from the publishers.
Although all care is taken to ensure integrity and the quality of this publication and the
information herein, no responsibility is assumed by the publishers nor the author for any
damage to the property or persons as a result of operation or use of this publication
and/or the information contained herein.
Published by:
CRC Press/Balkema
P.O. Box 11320, 2301 EH, Leiden, The Netherlands
e-mail: Pub.NL@taylorandfrancis.com
www.crcpress.com www.taylorandfrancis.com
ISBN: 978-0-415-62146-5 (Hbk + CD-ROM)

ISBN: 978-0-203-10518-4 (eBook)
Maritime Engineering and Technology Guedes Soares et al.

2012 Taylor & Francis Group, London, ISBN 978-0-415-62146-5
Table of contents
Preface
Organisation
XI
XIII
Ship traffic
Maritime situational awareness indicators based on AIS and MONICAP information
R.P. Gonalves de Deus, R.M. Correia Guerreiro & R.M. Alves Francisco
Intelligent guidance in collision avoidance of maritime transportation

L.P. Perera, J.P. Carvalho & C. Guedes Soares
The importance of satellite-based vessel monitoring system (VMS) for fisheries management:
A case study in the Portuguese trawl fleet
T. Pilar-Fonseca, A. Campos, P. Fonseca, B. Mendes, V. Henriques, J. Parente & M. Afonso-Dias
19
An algorithm for minimum-time ship-routing for the North Indian Ocean region
D. Sen & C.P. Padhy
25
Analysis of maritime traffic off the coast of Portugal

P. Silveira, A.P. Teixeira & C. Guedes Soares
35
Ship design
Experimental study of the performance of a rigid wing sail
J.H. Bettencourt, N. Fonseca & C. Guedes Soares
45
Considering disabled people in sailing yacht design

F. Cerveira, N. Fonseca & L. Sutherland
53
Hull shape design of a compressed natural gas ship

E.B. Malta, M.C. Nogueira, A.A. Ramos & C.M. Sampaio
63
Short sea shipping route and stowage plan optimization

P.T. Martins, V. Lobo & A. Moura
73
Non-linear optimization applied to preliminary ship design

J.C. Ruas & M. Ventura
79
Planning a fleet of containerships for a given set of ports

C.A. Silva & C. Guedes Soares
87
Design development of an oil and garbage collector and boom layer vessel
J.A. Valente, P. Antunes, L. Batista & S. Ribeiro e Silva
97
Generation of an adaptive triangular mesh from a parametric surface

J.M. Varela & M. Ventura
107
Ship propulsion & control

Controllable pitch propeller control with neural network algorithm
F. Alarcin, S. Ekinci & K. Gulez
115
Conversion of large-bore diesel engines for heavy fuel oil and natural gas dual fuel operation
J.M. Gomes Antunes, R. Mikalsen & A.P. Roskilly
121
Conditioned maintenance predictive analysis in maritime propulsion engines

S.S. Lampreia, V. Vairinhos, A.S. Matos, J.G. Requeijo & J.M. Dias
127
A nonlinear control approach for a vessel steering system

L.P. Perera & C. Guedes Soares
133
Onboard systems
Broadband communications with adhoc networks case study with simulated scenarios
F. Correia, J. Santos & P. Mesquita
141
Bridge officer decision support tool

P.T. Martins & V. Lobo
147
Decision support system for the safe operation of fishing vessels in waves
J.M. Rodrigues, L.P. Perera & C. Guedes Soares
153
Ship dynamics and hydrodynamics

Prediction of the motions of fishing vessels using time domain 3D panel method
R. Datta, J.M. Rodrigues & C. Guedes Soares
165
Influence of sea currents on manoeuvring of a surface autonomous model

V. Ferrari, L. Moreira, S. Sutulo & C. Guedes Soares
173
Interaction of gravity waves with moored flexible floating membrane

D. Karmakar & C. Guedes Soares
181
Review of model test techniques of local slamming on ships

H. Luo & C. Guedes Soares
189
A static analytical method for the preliminary design of multiple line mooring systems
A. Monarcha & N. Fonseca
195
Estimation of hull bending moment and shear force from ship motions using neural networks
L. Moreira & C. Guedes Soares
205
Wave frequency dynamics of a tanker moored inside the port

A. Pao, N. Fonseca & C. Guedes Soares
213
Experiment and time domain method comparison for the responses of a containership
induced by the three sisters abnormal waves
S. Rajendran, N. Fonseca, & C. Guedes Soares
223
Dynamic analysis of a purse seiner and a trawler fishing vessels in the Portuguese coastal zone
M. Tello, S. Ribeiro e Silva & C. Guedes Soares
231
Ship hydrodynamic interaction forces in restricted waters

X. Zhou, S. Sutulo & C. Guedes Soares
241
Ship structures
Influence of boundary constraints on the residual stresses and deformations in a
high strength steel plate after welding
M. Adak & C. Guedes Soares
253
Innovation protecting ballast tanks of new ships against corrosion: An IMO PSPC-compliant
Portuguese-born new coating concept
J. Azevedo
261
Analysis of the fluid-structure interaction of a composite motor yacht

H. Bagbanci, N. Baykut, A.S. Oktem & C. Guedes Soares
269
Structural reinforcement study of a floating dock mooring structure using FEM software
T. Carvalho, E. Mota & A. Correia Rodrigues
277
VI
Numerical investigations to study the effect of weld parameters on the temperature-time

history in steel plates
B.Q. Chen, M. Adak & C. Guedes Soares
285
Mechanical properties assessment of specimen subjected to random non-uniform general

corrosion and tensile load
B.Q. Chen, Y. Garbatov & C. Guedes Soares
293
Fatigue damage assessment of stiffener-frame structure

W. Huang, Y. Garbatov & C. Guedes Soares
301
Study on ultimate torsional strength of ship hulls

B. Liu & C. Guedes Soares
309
A new trigonometric shear deformation theory for sandwich and composite laminated plates
J.L. Mantari, A.S. Oktem & C. Guedes Soares
317
Layerwise finite element formulation for the analysis of laminates and sandwich panels
323
FE parameters estimation and analysis of ultimate strength of a box girder

S. Saad-Eldeen, Y. Garbatov & C. Guedes Soares
331
Structural behaviour of a lightweight craft

M.S. Santos, R.F. Martins, P.P. Silva & A.R. Mateus
339
Comparison of numerical and experimental results of the modal analysis of a ship deck panel
G. timac, H. Luo & C. Guedes Soares
349
Ultimate strength assessment accounting for the effect of finite element modelling
M. Tekgoz, Y. Garbatov & C. Guedes Soares
353
Fatigue damage assessment of a tanker structural detail based on the effective notch stress approach
K. Tran Nguyen, Y. Garbatov & C. Guedes Soares
363
Strength of a tanker double bottom in various grounding scenarios

R. Villavicencio, J. Amdahl & C. Guedes Soares
375
Review of material relations for marine structures collision simulations

R. Villavicencio & C. Guedes Soares
381
Response of stiffeners with attached plate subjected to lateral impact

R. Villavicencio, B. Liu & C. Guedes Soares
393
Explicit FE simulation of slamming load on rigid wedges with various deadrise

angles during water entry
S. Wang, H. Luo & C. Guedes Soares
Comparison of numerical results with experiments on ultimate strength of stocky stiffened panels
M. Xu & C. Guedes Soares
Numerical study of the effect of geometry and boundary conditions on the collapse behaviour of
long stiffened panels
M. Xu & C. Guedes Soares
399
407
415
Risk and reliability

Risk assessment to the approach and berth of LNG vessels at FLNG systems
P. Anto & C. Guedes Soares
425
Modeling the risk of product spills in LNG tankers

F.B. Natacci, P. Anto, C. Guedes Soares & M.R. Martins
433
Production regularity assessment using stochastic Petri nets with predicates

F.P. Santos, A.P. Teixeira & C. Guedes Soares
441
VII
Reliability assessment of a plate subjected to randomly non-uniform corrosion and compressive load
J.E. Silva, Y. Garbatov & C. Guedes Soares
451
Vertical breakwater reliability analysis with direct integration method

L. Zhang, S. Dong & C. Guedes Soares
459
Wind & wave Modelling

Assessment of wind and wave simulations for an enclosed sea using satellite data
M. Bernardino, N. Salvao & L. Rusu
467
Comparison of two wind and wave data sets from the North Atlantic
R.M. Campos & C. Guedes Soares
473
Evaluation of the wave models SWAN and STWAVE in shallow water using nested schemes
M. Gonalves, E. Rusu & C. Guedes Soares
481
Assessment of wave height extreme quantiles in the North Atlantic using regional frequency analysis
C. Lucas, G. Muraleedharan & C. Guedes Soares
487
Maximum entropy modelling of extreme significant wave heights on the Portuguese coast
V. Petrov, C. Guedes Soares & H. Gotovac
497
Boussinesq-type numerical model for wave propagation near shore and wave penetration in harbors
L. Pinheiro, C.J. Fortes, J.A. Santos, L. Fernandes & M. Walkley
505
Influence of wave data bases on the long term prediction of wave induced loads on ships
J. Queirs, M. Bernardino & C. Guedes Soares
513
Study of the wave transformation in the central part of the Portuguese nearshore with
high resolution models
E. Rusu, M. Gonalves & C. Guedes Soares
525
Modelling the wave conditions in the Archipelago of Azores

L. Rusu, P. Pilar & C. Guedes Soares
533
The global wave age climate

A. Semedo, A. Rutgersson, A. Sterl & K. Suelj
539
Renewable energy
Effect of the environment on the design loads on monopile offshore wind turbine
H. Bagbanci, D. Karmakar & C. Guedes Soares
547
Review of offshore floating wind turbines concepts

553
Modelling wave energy for the north coast of Spain

A.R. Bento, P. Martinho & C. Guedes Soares
563
Reflection and transmission of gravity waves by a large floating circular elastic plate
J. Bhattacharjee & C. Guedes Soares
571
Computational simulation of a buoy behaviour for automation and power generation in

aquaculture facilities
S. Ferreo, A. Munn, V. Daz-Casas, P. Farias & F. Lpez Pea
Review and classification of wave energy converters
C. Guedes Soares, J. Bhattacharjee, M. Tello & L. Pietra
Analysis of water depth effects on the wave power resource and the energy captured by a
wave energy converter
VIII
579
585
595
Conceptual design of an ocean current turbine for deep waters

L.R. Nuez Rivas, A. Lpez Pieiro, J.A. Somolinos Sanchez & F. Robledo de Miguel
607
Assessment of the wave energy in two enclosed seas

E. Rusu & C. Guedes Soares
613
Multi-purpose Offshore Platform (MpOP)

N. Santana
623
Development of an integrated system for personnel and equipment transfer to offshore wind turbines
T.A. Santos, E. Carichas, N. Fonseca, J. Pessoa, F. Duarte, J. Abreu Valente, L. Baptista,
J. Cruz & M. Leal
631
Modeling wave energy for the Portuguese coast

D. Silva, P. Martinho & C. Guedes Soares
647
Dynamics and hydrodynamics of ship like wave energy converters in roll

M. Tello, J. Bhattacharjee & C. Guedes Soares
655
Diverse
The CARCACE project deepwater platforms Modular designs for in situ experiments
D. Ribeiro & A. Hilrio
669
MAR International Shipping Register of Madeira

M. Sequeira
677
Author index
683
IX

Preface
Since 1987, the Naval Architecture and Marine Engineering branch of the Portuguese Association of Engineers
(Ordem dos Engenheiros) and the Centre for Marine Technology and Engineering (CENTEC) of the Instituto
Superior Tcnico (IST), Technical University of Lisbon, have been organizing national conferences on Naval
Architecture and Marine Engineering. Initially, they were organised annually and later became biannual events.
These meetings had the objective of bringing together Portuguese professionals and giving them an opportunity
to present and discuss the ongoing technical activities. The meetings have been typically attended by 150 to 200
participants and the number of papers presented at each meeting was in the order of 30 in the beginning and 50
at later events.
The papers presented in these meetings have been collected in a series of books written in Portuguese:
Guedes Soares, C. (Ed.) (1987), Design of Ships and Marine Structures, Naval Architecture and Marine
Engineering in Portugal (in Portuguese), Vol. I, Lisbon, (ISBN: 972-8348-00-2).
Guedes Soares, C. (Ed.) (1987), Maritime Transport and Shipyards, Naval Architecture and Marine Engineering
in Portugal (in Portuguese), Vol. II, Lisbon, (ISBN: 972-8348-01-0).
Guedes Soares, C. (Ed.) (1988), The Achievements of Portuguese Engineering, Naval Architecture and Marine
Engineering in Portugal (in Portuguese), Vol. III, Lisbon, (ISBN: 972-8348-02-9).
Guedes Soares, C. (Ed.) (1989), Maritime Safety, Naval Architecture and Marine Engineering in Portugal
(in Portuguese), Vol. IV, Lisbon, (ISBN: 972-8348-03-7).
Guedes Soares, C. (Ed.) (1989), Maritime Safety, Naval Architecture and Marine Engineering in Portugal
(in Portuguese), Vol. V, Lisbon, (ISBN: 972-8348-04-5).
Guedes Soares, C. (Ed.) (1989), The Teaching of Naval Architecture and Marine Engineering, Naval Architecture
and Marine Engineering in Portugal (in Portuguese), Vol. VI, Lisbon, (ISBN: 972-8348-05-3).
Guedes Soares, C. (Ed.) (1990), The Merchant Fleet, Naval Architecture and Marine Engineering in Portugal
(in Portuguese), Vol. VII, Lisbon, (ISBN: 972-8348-06-1).
Guedes Soares, C. (Ed.) (1990), The Fishing Fleet, Naval Architecture and Marine Engineering in Portugal
(in Portuguese), Vol. VIII, Lisbon, (ISBN: 972-8348-07-X).
Guedes Soares, C. (Ed.) (1992), Exploitation of the Portuguese Exclusive Economic Zone, Naval Architecture
and Marine Engineering in Portugal (in Portuguese), Vol. IX, Lisbon, (ISBN: 972-8348-08-8).
Guedes Soares, C. (Ed.) (1995), The Portuguese Maritime Industry in an European Context, Naval Architecture
and Marine Engineering in Portugal (in Portuguese), Vol. X, IST, Lisbon, (ISBN: 972-8348-09-6).
Guedes Soares, C. and Mira Monerris, A. (Eds.) (1997), Safety, Quality and Environment in the Maritime
Industries (in Portuguese), Instituto Superior Tcnico, (ISBN: 972-8348-10-X).
Guedes Soares, C. and Beiro Reis, J. (Eds.) (2000), The Sea and the Challenges of the Future (in Portuguese),
Edies Salamandra, Lda., Lisbon, Vol. XI, (ISBN: 972-689-179-5).
Guedes Soares, C., Beiro Reis, J. and Martins Guerreiro, M.B. (Eds.) (2002), The Sea, Source of Sustainable
Development (in Portuguese), Edies Salamandra, Lda., Lisbon, Vol. XII, (ISBN: 972-689-215-5).
Guedes Soares, C. and Brito, V.G. (Eds.) (2004), Maritime Activities and Engineering (in Portuguese), Edies
Salamandra, Lda., Lisbon, Vol. XIII, (ISBN: 972-68-229-5).
Guedes Soares, C. and Brito, V.G. (Eds.) (2006), Innovation and Development in Maritime Activities,
(in Portuguese), Edies Salamandra, Lda., Lisbon, Vol. XIV, (ISBN: 978-972-689-232-8).
Guedes Soares, C. and Costa Monteiro, C. (Eds.) (2008), The Portuguese Maritime Sector (in Portuguese),
Vol. XV, Edies Salamandra, Lda., Lisbon, (ISBN: 978-972-689-237-3).
At the same time, as the conferences have become more mature, the international contacts have also increased
and the industry became more international in such a way that the fact that the conference was in Portuguese
started to hinder its further development with wider participation. Therefore, for the session that was planned
for 2010, a decision was made to experiment with having also papers in English, mixed with the usual papers
in Portuguese.
The national meetings have normally taken place around the month of November but, given that IST was
completing 100 years of existence in May 2011, another decision was made to postpone this Conference to
1012 May 2011 so as to associate it also with the commemorations of the 100 years of IST.
Thus, the Naval Architecture component of the 100 years commemorations of IST has become the First
International Conference of Maritime Technology and Engineering (MARTECH 2011). In this transitional
XI
situation, the papers and sessions were in English and in Portuguese, so that the MARTECH proceedings are
two books, the present one in English and another one in Portuguese.
As the decision to change the nature of the Conference was made at a relatively late stage, it was also decided
that the papers from the Conference would be published afterwards instead of being published at the time of the
Conference. This is the reason for the publication of this book at this stage.
The Centre for Marine Technology and Engineering of the Instituto Superior Tcnico (IST) completed its 15th
anniversary during 2010 and during that year an anniversary book was prepared with contributions from several
experts that have cooperated with CENTEC during the years. The publication of that book was postponed to
coincide with MARTECH 2011 and some of the authors of that book made presentations of their papers during
the Conference.
The MARTECH Conference is planned to be continued at two-year intervals and it is planned that at future
events papers will be prepared beforehand and the book with the proceedings will be ready for the Conference.
The contents of the present book are organized in the main subject areas corresponding to the main sessions
in the Conference and within each group the papers are listed by the alphabetic order of the authors.
We want to thank all contributors for their efforts and we hope that this Conference will be continued and
improved in the future.
C. Guedes Soares, Y. Garbatov, S. Sutulo & T.A. Santos
XII

Organisation
Conference Chairmen
Prof. Carlos Guedes Soares, IST, Technical University of Lisbon, Portugal
Eng. Nuno Santos, Ordem dos Engenheiros, Portugal
Organizing Committee
Prof. ngelo Teixeira, IST, Technical University of Lisbon, Portugal
Eng. Dina Dimas, Ordem dos Engenheiros, Portugal
Dr. Leigh Sutherland, IST, Technical University of Lisbon, Portugal
Eng. Paulo Viana, Ordem dos Engenheiros, Portugal
Prof. Serge Sutulo, IST, Technical University of Lisbon, Portugal
Prof. Srgio Ribeiro e Silva, IST, Technical University of Lisbon, Portugal
Dr. Tiago A. Santos, Ordem dos Engenheiros, Portugal
Prof. Yordan Garbatov, IST, Technical University of Lisbon, Portugal
Technical Programme & Conference Secretariat
Maria de Ftima Pina, IST, Technical University of Lisbon, Portugal
Sandra Ponce, IST, Technical University of Lisbon, Portugal
In commemoration of the
100 years of Instituto Superior Tcnico
(1911/2011)
Organised by:
XIII
Sponsored by:
Associao das Indstrias Navais

Comunidade Porturia de Aveiro
Det Norske Veritas Portugal
Direco de Navios Marinha
DouroAzul
ENONDAS Energia das Ondas
Estaleiros Navais de Peniche
Estaleiros Navais de Viana do Castelo
Euroshide European Ship Design
Instituto Hidrogrfico
Lisnave, Estaleiros Navais
RINA
Transtejo
XIV
Ship traffic

Maritime situational awareness indicators based on AIS and

MONICAP information
R.P. Gonalves de Deus
CINAV Portuguese Naval Research Center, Portuguese Naval School, Portugal
Direco de Anlise e Gesto de Informao, Portuguese Navy, Portugal
R.M. Correia Guerreiro

Naval Command, Portuguese Navy, Portugal
R.M. Alves Francisco

Direco de Anlise e Gesto de Informao, Portuguese Navy, Portugal
ABSTRACT: In this paper we present a set of statistical indicators built from AIS and MONICAP data, which
we designate as maritime situational awareness indicators. To achieve this purpose a prototype was built in order
to easily manipulate large amounts of data and to rapidly implement the necessary routines to build and visualize
these indicators.
1
1.1
INTRODUCTION
understanding of anything associated with the global

maritime environment that could impact the security,
safety, economy or environment of a country. The
concept of MSA was rapidly disseminated through
NATO (North Atlantic Treaty Organization) and EU
(European Union) nations and had a particular impact
on the monitoring capabilities on their respective
Armed Forces. In this work we intend to provide basic
information elements to achieve this condition. These
basic information elements are statistical indicators,
for which there is a process capable of building them
for any given geographical area and time period, as
long as there is available data ready for use. This
article will focus mainly on the statistical indicators
defined, which are designated as Maritime Situational
Indicators and the prototype developed to build them.
Maritime dimension of Portugal
Portugal has a territory with approximately

91,760 km2 , but has sovereignty or jurisdiction over a
much larger area, of around 1720.560 km2 , including
inland waters, territorial sea and Exclusive Economic
Zone.About 53% of the EUs foreign commerce passes
on waters under Portuguese jurisdiction, and around
60% of all Portuguese foreign trade is performed by
sea. It is very important to know which areas have
more maritime traffic and which type (manufactured
good, energy transport, etc.) prevails along a given
time period (Marinha Portuguesa, 2010). In this work
we aim to estimate the number of distinct vessels that
transit through several areas of interest. One example
is the Economic Exclusive Zone for the Portuguese
mainland. To do this, the Automatic Identification System (AIS) plays a critical role to access information
regarding the position of ships. The Portuguese Navy
has implemented a network ofAIS stations in the mainland territory in order to efficiently cover all the areas
in the Portuguese territorial waters and also the EEZ
of the mainland. The data used in this work is provided
from this network.
INFORMATION SOURCES
2.1 AIS data

AIS, which stands for Automatic Identification System, is at its most basic level an electronic transponder
installed on a boat or vessel that continuously transmits a VHF signal containing information including
its name, call sign, type and position (IALA, 2002a,
2002b). An AIS receiver mounted on another craft or
on the coastline can receive that information and use
the signal to calculate and display the vessels course
and speed.
The Portuguese Navy has a network of AIS stations that cover the mainland territorial waters up to
50 nautical miles from shore, depending on the VHF
propagation conditions (Correia, 2007). The AIS data
1.2 Maritime Situational Awareness and scope of

this work
Maritime Situational Awareness (MSA) is an evolving concept that emerged from the need to improve
safety and security for US citizens after the terrorist attacks of September 2001 (9/11). Maritime
Situational Awareness is defined as the effective
related to the Search and Rescue Service which is

coordinated by the Navy.
3
Figure 1. Image
information.
of
SADAP
showing
The Maritime Situational Indicators presented in this

work are defined for a specific geographical area and
a period of 24 hours each. By geographical area we
mean a polygon. If the area is the Portuguese mainland EEZ then we would like to know, for example,
how many vessels equipped with AIS passed through
the Portuguese mainland EEZ on 20th June 2011.
The time period in the later case is a 24 hour period
starting from 00h00m in 20th June to 00h00m in
21th June 2011. These indicators were validated by
the Portuguese Naval staff as relevant and important
information elements to achieve Maritime Situational
Awareness.
MONICAP
is then sent from the stations to a server in the Navys

private network designated AIS server. A local application was developed to continuously register the AIS
strings received by the AIS server. Two other applications were developed to decode and to store the
decoded AIS information into data structures in order
to easily access the AIS data and perform intensive
calculations.
2.2
3.1 Number of distinct vessels

One critical question we aim to answer is How many
vessels equipped withAIS and MONICAP were present
in a given geographical area in a period of 24 hours?.
This question was the major driver for this work.
In fact, it seemed easy to answer, however before the
development of the current prototype we found that it
was just impossible to get this information. The number of distinct vessels in a geographical area and time
period was designated as Maritime Situational Indicator 1. The answer to this type of question is a number,
but we also want to provide a list with all the dynamic
AIS messages related to the unique list of vessels used
to answer the how many question. To build this indicator, we only need to build the unique list of vessels
that were detected in a specific polygon and day.
MONICAP data
MONICAP is a monitoring system for fishery activities that uses GPS (Global Positioning System) for
positioning and Inmarsat C for satellite communications between vessels and shore control centers.
This system started has a pilot project in 1998 by
INOV in Portugal and its results have transformed
the way technology is used in fishery control worldwide. MONICAP system has been industrialized and
his currently installed in more than 800 fishing vessels
in countries such as Portugal, Spain, France, Ireland
and Angola.
The MONICAP system sends information regarding the position, name, speed and heading of fishing
vessels through satellite to a coordinator center in
Norway. This data is sent to each country according
to its maritime areas for which they have jurisdiction
regarding fishing activities. In Portugal the information is received by the Ministry of Agriculture and
Fisheries, which is then sent to the Portuguese Navy.
MONICAP information is received by internet in
every 6 hour period and stored in a local database.
The storage is performed by an Information System
developed exclusively by the Portuguese Navy called
SADAP (Sistema de Apoio Deciso para a Actividade de Patrulha1 ). SADAP integrates a GIS system
(Geographical Information System) that allows the
user to access statistics regarding fishery inspection
activities and accesses the latest position of fishing
vessels operating in the Portuguese EEZ. The main
purpose of the MONICAP capability in SADAP is
1
MARITIME SITUATIONAL INDICATORS
3.2 Average speed

This indicator, designated Maritime Situational Indicator 2, is related with the previous indicator because
the average speed is computed out of the average speed
of each distinct vessel calculated on Indicator 1. The
calculation of the average speed of a particular vessel is
done by a simple average of the speed values recorded
in the dynamic AIS messages received and registered
in our system. For example, if a vessel has 10 dynamic
type 1 AIS messages inside the polygon on a specific
day, we can average the 10 values of speed and produce a final average for that specific vessel. Doing this
for every distinct vessel and doing the average of those
values, we obtain an average speed for that polygon on
that specific day. This information helps to characterize the navigational behavior of AIS equipped vessels
in a specific area.
3.3 Average transit duration
Having a list of dynamic AIS messages (with the
decoded information) for a specific polygon and day, it
is possible to obtain, for each unique vessel, the earliest
Fishing Patrol Decision Support System
(2010). The choice for this development environment

was due to the availability of high level and also
low level functions that allowed one to produce very
interesting graphics and also quickly generate the
required indicators.
and the latest position and also its time of occurrence.

If we calculate the distance in time between the earliest
and the latest position inside the polygon for a specific day, we know how long a vessel stayed inside a
polygon. We can measure this time in hours or minutes. Usually this period of time corresponds to the
duration of a transit through a specific maritime area.
If we calculate those values for each vessel and do the
respective average we have the Maritime Situational
Indicator 3. Once more, this piece of information helps
to characterize the navigational behavior for a specific
maritime area.
4.1 Data structures

In this work, only dynamic information from AIS and
MONICAP messages are used. This has the advantage
to represent all the data as numbers. Since we are only
working with numbers, one can use the fastest data
structure in MATLAB which is the multidimensional
array. The AIS data is stored in a table with few basic
attributes. The relational model is implicit in the data
stored in the multidimensional arrays (Eckerson, 2003,
Elmasri, 2004). Each line of the data tables corresponds to a dynamic AIS message. The basic attributes
are MMSI (maritime mobile system identification),
latitude, longitude, speed, COG, NGS (navigational
status), DTG (Date-time group) and Heading, not
necessarily by this order. The MONICAP data is
similarly stored the same way, although the main operational data base in SADAP stores this data with
a different structure. All the data used in the calculations of the maritime indicators are stored in
multi-dimensional arrays in MAT-files.
The raw data received from the AIS server contains
only the AIS strings. These are recorded continuously
24 hours daily in a cell array and saved in a new
MAT-file within each 10 minute period. The reason
for the 10 minute period is that if the period were
larger it would produce larger MAT-files that would
be slower to import for decoding purposes. Approximately 144 files are created in a 24 hour period.
These files (with the raw AIS strings) will be decoded
by a second application that runs on a separate local
workstation that produces a mirror MAT-file with the
decoded dynamic AIS messages. In these new files the
content are multidimensional arrays while in the first
were merely cell arrays.
3.4 Average distance to shore

Since the data used in this work is captured by the AIS
receiving stations installed in the Portuguese mainland
territory, we are interested in knowing how far the vessels equipped with AIS navigate from the Portuguese
coast. To calculate this we need a polygon of the
Portuguese coast and perform some complex calculations. These calculations require knowing the distance
from each position of a vessel, inside a polygon on a
specific day, to each point in the coastal polygon inside
the area of interest.A good estimation of the distance to
shore from a given position of a vessel is the minimum
distance from that point to all the points of the coastal
polygon. Doing the respective average of those minima we have a good estimation of the average distance
to shore for a specific vessel that navigated in a specific polygon and day. All the points in our polygons
are in degrees and defined by latitude and longitude,
with positive values for North and East, and negative
values for South and West.
3.5
Dominant course
One of the several parts of information contained in

a dynamic AIS message is the COG (course over
ground). The COG is the angle that the intended path
of the vessel makes with true north. Course is measured in degrees from 0 clockwise to 360 in compass
convention (0 being north, 90 being east). In order to
identify the major routes in a specific area we defined
8 sets of 45 starting from true north clockwise. Then
we count the number of dynamic AIS messages, which
the respective COG, belong to each of the 8 bins. The
first bin corresponds to directions between North and
North-east, the second bin to North-east and East, and
so on. This is a rough estimation of the courses made
by the vessels equipped with AIS in a specific area.
But in areas with good AIS coverage ratios, this indicator reveals the major dominant courses. This is the
Maritime Situational Indicator 5.
4
4.2 Operational database schema

The MONICAP and AIS data come from different sources and have to be stored separately. Here
we describe the process of extracting and decoding
the AIS strings retrieved from the AIS server and the
routines responsible for producing the previous indicators. The processes of extracting and decoding the
AIS strings were written in MATLAB native language
and an executable was produced to connect to the AIS
server, read the strings and produce the MAT-files in
each 10 minute period.A second script was also written
in MATLAB to decode the AIS strings. Only messages
of type 1, 2, 3, 5, 18 and 19 were decoded for this
work. The decoding process saves the AIS information
in multidimensional arrays according to the relational
model and make up the AIS operational database.
The MONICAP data is stored in specific repository native to the SADAP system. This process enable
the storing of large amounts of historical AIS and
INFORMATION ARCHITECURE
Several routines were implemented to build the previous indicators. These routines were embedded on a
prototype developed in MATLAB and a more thorough
description can be found in Melo (2010) and Filipe
Figure 2. AIS operational database schema.
Figure 5. Circle polygon 70 nautical miles west from Cape

Roca.
Table 1. MSI for polygon area west of Lisbon during the
first week of May 2011.
Maritime Situational Indicators
Figure 3. Prototype figure: Lists of AIS messages and
MONICAP vessels in polygon area on 22th July 2011.
2
Kts
3
Hours
4
Nm
5
Course
Sunday (1th)
Monday (2th)
Tuesday (3th)
Wednesday (4th)
Thursday (5th)
Friday (6th)
Saturday (7th)
156
138
135
124
133
152
137
13.2
13.3
13.7
13.4
13.4
12.8
13.1
4.7
5.1
4.6
4.9
4.6
5.1
4.7
29.8
30.7
29.1
30.3
31.8
28
30.6
SE-E
SE-E
N-NE
SE-E
SE-E
NW-N
NW-N
circle, and a button to load a polygon from a MAT-file

or XLS-file.
The upper left list box shows the attributes of the
AIS messages present in the database which are inside
the polygon on the specific date chosen by the user.
The upper right list box shows a list of unique vessels and their minimum speed during their transit on
that day and inside the polygon. The lower list boxes
show the same information but regarding MONICAP
vessels.
Figure 4. Maritime Situational Indicator report.
MONICAP data, that can be easily imported to the

prototype.
4.3
Polygon 1
Development Laboratory prototype
In order to define a geographical area of interest and

time period the user must have GIS capabilities and
these must be linked with the routines to build the
maritime situational indicators. For this purpose we
developed several figures in MATLAB using functions from the Mapping Toolbox that allow the user to
define a polygon in the world map and choose a specific day to analyze.The routines to build the indicators
were also written in the same language.
The figure above is divided in three parts: world
map (upper center), analysis panel (right) and AIS and
MONICAP lists (bottom left). The coast line in the
world map is plotted from vector data containing latitude and longitude values. Latitudes and longitudes are
stored as MATLAB vectors in a MAT-file. The analysis
panel allows the user to select a specific day, month and
year. The toolbar has, from left to right, zoom and pan
options, followed from drawing buttons and statistics
calculation. Next to the pan there are three buttons that
allow drawing a polygon: button to draw a rectangle, a
4.4 Reporting
After choosing a polygon and a day, the user selects
the button to load the AIS and MONICAP data from
the database. This action will draw in the map all
the positions regarding these two types of information
and loads all the list boxes mentioned before. The user
can also export the information contained in the list
boxes to Excel. This option simply copies the content
of the lists to an Excel table.
In order to produce the Maritime Situational
Indicators for a specific polygon and day, the user only
needs to press the ICS button in the toolbar. This
button will produce a small report in a text file with
all the indicators described in this article.
The user can edit the report by filling the polygons
name or add other relevant information. To exemplify
these indicators we have drawn a circle with 70 nautical
miles radius westward from cape Roca, has shown in
Figure 5.
Table 1 shows a weekly variation of the indicators defined in this work. The prototype can be used
to produce this information for a full year. The only
requirement is that the data exists in the database.
5
Eckerson (2003). Four ways to build a Data Warehouse.

Elmasri (2004). Fundamentals of Database Systems
(4th Edition). Boston: Pearson Addison Wesley.
e-Nav (2009). Introduction to the IALA Universal Maritime
Data Model (UMDM)
Filipe (2010). Modelao e Implementao de repositrio
de dados do sistema AIS. Portuguese Naval School,
Portuguese Navy.
Fong (2005). A Simulation Study of Multi-Sensor Maritime
Surveillance Using SIMLAB. DRDC CORA Technical
Report TR 2005-21, Department of National Defense,
Canada.
Gucma (2008). Low Cost AIS System for Safe Navigation.
Jornalof KONBIN, Vol. 3.
IALA (2002a). IALA Guidelines onThe UniversalAutomatic
Identification Systems (AIS) Technical issues (1.1 ed.
Vol. 1, Part II). Saint Germain en Laye: IALA).
IALA (2002b). IALA Guidelines on The Universal Automatic Identification Systems (AIS) Thecnical issues
(1.1 ed. Vol. 1, Part I). Saint Germain en Laye: IALA).
IALA (2005) Technical Clarifications on Recommendation
ITU-R M.1371-1.
IALA (2008). The Comparison of AIS stations, IALA Guideline N 1059 (Ed. 1ed.) Saint Germain en Laye: IALA.
IEC (2001). Maritime Navigation and Radiocommunication Equipment and Systems Automatic Identification
Systems (AIS) (Vol. IEC 61993-2). Geneva, Switzerland:
IEC.
IMO (2002a). Guidelines for the Onboard Operational Use
of Shipborne Automatic Identification Systems (AIS)
(VOl. Resolution A.917 (22)).
IMO (2002b) Introduction to IMO. Retrieved 10th October
2009, from http://www.imo.org/.
Marinha Portuguesa. (2010). Misso da Marinha Portuguesa.
Obtido em 23 de Junho de 2010, de Marinha Portuguesa:
http://www.marinha.pt/PT/amarinha/Pages/missao.aspx
Mason (2006). Design Specifications: Simulation of Surface
Surveillance System-of-Systems Success. DRDC CORA
Technical Report TR 2006-05, Department of National
Defense, Canada.
Melo (2010). Indicadores de Conhecimento Situacional
Martimo com recurso aos dados AIS. Portuguese Naval
School, Portuguese Navy.
Molinelli et al. (1996). Historical Temporal Shipping (HITS)
3.2 Database Description, Update Process, Display
Software. Unpublished draft copy, Planning Systems
Incorporated.
Navy Modeling and Simulation Management Office.
Resorce Description: Historical Temporal Shipping
Database.
CONCLUSIONS
The work presented in this article is the unclassified

part of a major project which is currently being developed in the Portuguese Navy. The primary objective
is to put together a set of indicators that may help the
decision maker to better understand the navigational
behavior on a specific area.
The maritime situational indicators that were
defined are developed from a set of data which is
very well defined in terms of space and time. In this
way, we can define rigorously the statistical population from which we retrieve a statistical sample to build
these indicators. The statistical population, for the AIS
equipped vessels, is the set of AIS messages (type 1, 2
and 3) sent from AIS equipped vessels that were inside
a given polygon on a specific day. Of course we may
not have access to all of these messages due to several
technical reasons, and have a missing data problem.
In future work we aim to do statistical inference
with this data on a new set of objectives regarding the
concept of maritime situational awareness and investigate navigation patterns to improve the efficiency of
maritime surveillance missions.
REFERENCES
Codd (1970). A Relational Model of Data for Large Shared
Data Banks. Communications of the ACM, Vol. 13,
377387.
Comisso Estratgica dos Oceanos (2008). Relatrio da
Comisso Estratgica dos Oceanos Parte 1. Lisboa:
Comisso Estratgica dos Oceanos.
Correia (2007). Controlar Remotamente o Mar. Revista da
Armada, 1012.
Cortes (2005). Sistemas de Suporte Deciso (1a ed.).
Lisboa. FCA.
Dittrich (1986). Object-Oriented Database Systems: The
notion and the issues. Paper presented at the International
Workshop on Object-Oriented Database, New York.

Intelligent guidance in collision avoidance of maritime transportation

Lokukaluge P. Perera
Centre for Marine Technology and Engineering (CENTEC), Instituto Superior Tecnico,
J.P. Carvalho
INESC-ID, Instituto Superior Tecnico, Technical University of Lisbon, Portugal
C. Guedes Soares
Technical University of Lisbon, Portugal.
ABSTRACT: This paper focuses on an overview of an autonomous navigation system in maritime transportation that is supplied with intelligent guidance to avoid collision, while respecting the COLREGs rules and
regulations and using expert knowledge in navigation for close quarters manoeuvres. The proposed collision
avoidance process consists of two modules: a fuzzy logic based parallel decision making module and a Bayesian
network based sequential action formulation module. Successful simulation results of the intelligent guidance in
collision avoidance of maritime transportation that is capable of making multiple collision avoidance decisions
and actions in order to avoid complex multi-vessel collision situations are also illustrated in this study.
1
INTRODUCTION
violation of the COLREGs rules and regulations. The

detailed description of the COLREGs rule and regulations and their interpretation for autonomous ocean
navigation with respect to the collision avoidance are
presented in Perera et al. (2010a).
This paper focuses on an overview of an intelligent maritime transportation system that includes
an autonomous collision avoidance process while
respecting the COLREGs rules and regulations and
expert knowledge in navigation. The autonomous
guidance in collision avoidance described in this work
consists mainly of a fuzzy logic based parallel decision
making module whose decisions are formulated into
sequential actions by a Bayesian network based module. The proposed collision avoidance process capabilities of making multiple parallel collision avoidance
decisions regarding several target vessels and those
decisions are executed as sequential actions in the vessel navigation to avoid complex collision situations are
also presented.
The intelligent guidance in maritime transportation is

still underdeveloped when compared with the land
and air transportation systems. In conventional maritime transportation, the most important navigation
factor is still the human guidance, and wrong judgment and miss-operations by humans have resulted
in many casualties and environmental disasters. The
reported data shows human errors are still one of the
major causes of maritime accidents (Guedes Soares, &
Teixeira, 2001) and 7596% of marine accidents and
causalities are caused by some types of human errors
(Rothblum et al. 2002, and Anto & Guedes Soares,
2008).
Therefore, the implementation of intelligent and
autonomous capabilities and the limitation of human
subjective factors in navigation, to increase the safety
and security of ocean navigation, are proposed in
this study. Similarly, the replacement of human inference by an intelligent decision formulation process
resulting in feasible actions for navigation and collision avoidance can reduce the number of maritime
accidents and their respective causalities.
To avoid collision situations, all vessels should follow the law of the sea. The current law of the sea was
formulated by the International Maritime Organization (IMO) in 1972 by the Convention on the International Regulations for Preventing Collisions at Sea
(COLREGs). However, the reported data of the maritime collisions presented by Statheros et al. (2008),
show that 56% of major maritime collisions involve
RELATED WORK
The decision making process and strategies in interaction situations of avoidance of collision in maritime
transportation are presented in Chauvin & Lardjane
(2008). Sato & Ishii (1998) proposed combining radar
and infrared imaging method to detect the target
vessel position conditions as a part of the collision
avoidance system. Lisowski et al. (2000) used neuralclassifiers to support the navigator in the process
of determining the vessels domain. On a similar

approach, Pietrzykowski and Uriasz (2009) proposed
the notion of a vessel domain in a collision situation as
depending on the parameters like vessel size, course
and heading angle of the encountered vessels.
Kwik (1989) presented collision risk calculations
for a two-ship collision encounter based on the kinematics and dynamics of marine vessels. Yavin et al.
(1995) considered the collision avoidance conditions
of a ship moving from one point to another in a narrow zig-zag channel proposing a computational open
loop command strategy for the rudder control system.
Sutulo et al. (2002) studied the problem of predicting a ship trajectory based on simplified maneuvering
models. An alternative approach based on neural
networks is also proposed by Moreira and Guedes
Soares (2003). Smierzchalski and Michalewicz (2000)
modeled safe ship trajectory in navigation using an
evolutionary algorithm and considering static and
dynamic constraints for the optimization process.
Ito et al. (1999) used genetic algorithms to search
for safe trajectories on collision situations in ocean
navigation. Similarly, experimental results on the same
topic are presented in Zeng et al. (2001). Hong et al.
(1999) presented a collision free trajectory in ocean
navigation based on a recursive algorithm that is formulated by analytical geometry and convex set theory.
Similarly, Cheng et al (2006) have presented trajectory optimization for ship collision avoidance system
based on a genetic algorithm.
Statheros et al. (2008) give an overview of computational intelligence techniques that are used in collision
avoidance in ocean navigation. Liu and Liu (2006)
used Case-Based Reasoning to illustrate learning of
collision avoidance techniques in ocean navigation
using previous recorded data of collision situations.
Furthermore, an intelligent anti-collision algorithm for
different collision conditions is designed and tested
on the computer based simulation platform by Yang
et al. (2007). Zhuo and Hearn (2008) presented a study
of two vessel collision avoidance in ocean navigation
using a self learning neuro-fuzzy network based online
and offline training scheme.
A fuzzy logic approach for collision avoidance conditions with the integration of a virtual force field is
proposed by Lee et al. (2004). Similarly, automatic
collision avoidance facilities for ship system using a
fuzzy logic based controller is proposed by Hasegawa
(1987). Benjamin et al. (2006) propose behavior based
controls formulated with interval programming for
collision avoidance of maritime transportation. The
collision avoidance behavior is illustrated in accordance with the Coast Guard Collision Regulations.
Benjamin and Curcio (2004) present the decision making process of ocean navigation based on an interval
programming model for a multi-objective decision
making algorithm.
A computational algorithm based on the If-Then
logic is defined and tested under simulator conditions by Smeaton and Coenen (1990) regarding different collision situations. Cockcroft and Lameijer
Figure 1. Two Vessel Collision Situation.
(2001) present a detailed description of the COLREGs

rules and regulations, implementation, interpretation
of judicial authorities of vessel collision and near-miss
situations.
3 COLLISION AVOIDANCE SITUATIONS
IN OCEAN NAVIGATION
3.1 Two vessel collision situation
A two-vessel collision situation is presented in Figure 1, where the own vessel, which is the one that
is equipped with the collision avoidance system, is
located in the point O(k)(xo (k), yo (k)), at the kth
time instant. The ith target vessel, which is the
one that needs to be avoided, is located at point
Pi (k)(xi (k), yi (k)), with the estimated navigational trajectory of Pi (k)Bi (k).
The target vessel estimated trajectory Pi (k)Bi (k)
will intercept the own vessel domain with the closest distance of RDCPA (k), around the points of Bi (k).
The own vessel speed and course conditions are represented by Vo (k) and o (k); the ith target vessel speed
and course conditions are represented by Vi (k) and
i (k); the ith target vessel bearing and relative bearing conditions are represented by i (k) and i,o (k); the
relative speed and course conditions of the ith target
vessel are represented by Vi,o (k) and i,o (k). All angles
are measured with respect to the positive Y axis.
The own vessel navigational space is divided into
three circular regions with radii Rvd , Rb and Ra . The
radius Ra represents the approximate range to the target vessel detection when the own vessel is in a Give
way situation, i.e., when the own vessel has low priority for navigation and should take appropriate actions
to avoid collision situations. The radius Rb represents
the approximate distance to the target vessel, when
10
Figure 2. Multi-Vessel Collision Situation.
the own vessel is in a Stand on situation with the

higher priority for navigation should take appropriate
actions to avoid collision due to absence of the appropriate actions from the target vessel. The radius Rvd
represents the vessel domain.
3.2
Multi-vessel collision situation
The expansion of a two vessel collision situation

into a multi-vessel collision situation is presented
in Figure 2. The own vessel is located in the point
O(k). The target vessels are located at the points of
P1 (k), P2 (k), . . . , Pn (k) with the navigational trajectories of S1 (k), S2 (k), . . . , Sn (k) at the kth time instant
respectively. The own vessel trajectory S0 (k) will intercept the trajectories S1 (k), S2 (k), . . . , Sn (k) around the
points of C1 (k), C2 (k), . . . , Cn (k) at the time instants
of T1 (k), T2 (k), . . . , Tn (k) respectively.
3.3
Figure 3. Block diagram for Collision Avoidance System.
Tracking Unit and the Trajectory Prediction Unit. The

integrated Radar or Laser measurement system is considered for the Scan Unit where the real-time position
data of the target vessels is collected. The target vessels data are used in the Data Classification Unit to
identify each vessel and the Clustered Data Tracking Unit tracks each target vessel separately using
this data. Finally the collected tracking data is used
to predict each target vessels trajectory in the Trajectory Prediction Unit. Due to the target vessels constant
speed and course condition assumptions, this process
is simplified in this study. Further details on target
vessel tracking under ocean navigation conditions are
presented in Perera and Guedes Soares (2010b).
The main objective of the CRA module is to evaluate the collision risk of each target vessel with respect
to the own vessel navigation. This is achieved by the
Relative Trajectory Formation Unit and the Collision
Time and Point Estimation Unit. The inputs into the
CRA module are the measured/estimated position data
of the own vessel and target vessels. The outputs of the
CRA module are Range (Ri (k)), Bearing (i (k)), Relative course (i,o (k)) and Relative speed (Vi,o (k)) of ith
target vessel. These outputs will be used as the inputs
of the PDM module at the kth time instant.
In addition, the Time until the collision situation
Ti (k) of the ith target vessel will input into the SAF
module from the CRA module. The PDM module consists of a fuzzy logic based decision making process
Modules of collision avoidance system
A block diagram for the proposed Collision Avoidance System (CAS) is presented in the Figure 3. The
complete CAS consists of four modules: the Vessel
Tracking & Trajectory Prediction (VTTP) module, the
Collision Risk Assessment (CRA) module, the Parallel
Decision Making (PDM) module, and the Sequential
Action Formation (SAF) module.
The inputs into the VTTP module are the realtime position of the own vessel (xo (k), yo (k)), that
can be measured or estimated by the GPS integrated
Inertial navigational systems, in the Cartesian coordinates, and the real-time position of each target vessels
(Ri (k), i (k)) that is measured in Polar coordinates.
The Range (Ri (k)) and Bearing (i (k)) values of the
ith target vessel can be obtained using Radar or Laser
measurement systems by the kth time instant as further
discussed by Perera and Guedes Soares (2010a).
The VTTP module consists of four units: the Scan
Unit, the Data Classification Unit, the Clustered Data
11
that generates parallel collision avoidance decisions

with respect to each target vessel at kth time instant.
As the next step, the parallel ith decision of collision
avoidance, Di (k), will forward from the PDM module
to the SAF module. The main objective of the SAF
module is to organize the parallel decisions made by
the PDM module into sequential actions, Ai (k), considering the Time until the collision situation, Ti (k),
that will be executed on the own vessel navigation.
These actions are further divided into two categories
of Course and Speed control actions that will be implemented on the propeller and rudder control systems of
the own vessel.
Figure 4. Course CAAF and Speed CAAF.
PARALLEL DECISION MAKING MODULE
The collision avoidance decisions of Course change

and Speed change are categorized as: Course to Starboard o > 0; course to Port o < 0; no course
change o = 0; increase speed Vo > 0; decrease
speed Vo < 0; no speed change Vo = 0. The defuzzification process uses the centre of gravity method.
The further details on the Fuzzy logic based parallel
decision making module are presented in Perera et al.
(2010b).
The main objective of the PDM module is to make the

collision avoidance decisions considering the collision
risk warnings. An overview of the PDM module is also
presented in Figure 3. The module consists of 3 main
units: the Fuzzification Unit, the Fuzzy Rules Unit and
the Defuzzification Unit.
4.1 Fuzzification unit
The inputs from the CRA module, Range (Ri (k)),
Bearing (i (k)), Relative course (i,o (k)) and Relative speed (Vi,o (k)) of the ith target vessel at the
kth time instant are fuzzified in this unit with
respect to the input Fuzzy Membership Functions
(FMFs): the Range FMF (Ri (k)), the Speed Ratio FMF
(Vi (k)/Vo (k)), the Bearing FMF (i (k)), and the Relative Course FMF (i,o (k)). Then, the fuzzified results
from the Fuzzification unit will be transferred to the
Fuzzy Rules Unit for further analysis.
4.2
5.1 Action execution in navigation

The main objective of the SAF module is to transform the parallel collision avoidance decisions that
are generated by the PDM module into a sequential action formulation that can be executed in the
own vessel. This can be achieved by collecting the
PDM module multiple collision avoidance decisions
for the kth time instant, Di (k) (Di (k), DVi (k)), and
evaluating them using the Time until the collision
situation, Ti (k), from the CRA module with respect
to each target vessel.
Final results are arranged as a sequential formation of actions, Ai (k) (Ai (k), AVi (k)) involving
the course and speed action at given time instants
(Ti (k), TVi (k)). Figures 4 gives examples of the
sequential course collision avoidance action function (CCAF) and the speed collision avoidance action
function (CCAF). Di (k), DVi (k) and Ai (k), AVi (k)
represent the course and speed change decisions and
actions at kth time instant respectively.
The overview of the SAF module is also presented
in Figure 4. The SAF module consists of 2 units: the
Course and Speed Actions Formation Units. The main
objective of these 2 units is to formulate the collision
avoidance course and speed control actions.
Fuzzy inference and rules
A Mamdani type IF <Antecedent> THEN

<Consequent> rule based system is developed and
inference via Min-Max norm is considered in the
Fuzzy Rules Unit. The IF-THEN Fuzzy rules are
developed in accordance with the Knowledge Base
(i.e. the COLREGs rules and regulations). However
the expert navigational knowledge is also considered
in the Fuzzy rules development process. Overtaking,
Head-on and Crossing, the three distinct situations
that involve risk of collision are considered for the
development of the IF-THEN Fuzzy rules.
4.3
SEQUENTIAL ACTION FORMATION

MODULE
Defuzzification
The collision avoidance decisions, Di (k), for each target vessel are generated by the Defuzzification Unit.
The Fuzzy inference results from the previous unit
are defuzzified based on the output Course Change
FMF and Speed change FMF to obtain the Course
change decisions, Di (k), and Speed change decisions, DVi (k), that will be formulated for collision
avoidance actions in the own vessel navigation.
5.2 Bayesian network

The continuous Bayesian Network module that is formulated to update the parallel collision avoidance
decisions into the sequential actions is presented in
Figure 5. The Bayesian network consists of four
12
where 1 i is the normalization constant. The transition

model of the CRF is considered as a Gaussian pertur2
bation of the constant covariance
to the current
states of the CRF and can be written as:
where 2 i is the normalization constant. The conditional observation model for collision risk is assumed
to be a Gaussian distribution with the constant covariance z2 and can be written as:
Figure 5. Bayesian Network Structure for Collision

Avoidance.
nodes: Collision Time Estimation, Collision Risk,

Actions Delay and Collision Avoidance Actions. The
inputs of the Bayesian network are the CollisionAvoidance Decisions, Di (k), and the Time until the collision
situation, Ti (k), generated respectively by the PDM
and CRA modules .
The main objective of the CollisionTime Estimation
node is to estimate the Time until the collision situation, Ti (k), between the own vessel and each of the
target vessels. The node Collision Risk inferences the
collision risk with respect to each target vessel considering the Collision Time Estimation. The Actions
Delay node is designed to formulate the appropriate
time to take collision avoidance actions. The Collision Avoidance Action node will inference by the two
nodes of Actions Delay and Collision Risk and the
Collision Decisions as presented in Figure 5.
5.3
where 3 i is the normalization constant. Considering the Bayesian Rule, the CRF update from the
observations can be written:
5.4 Collision avoidance action function

The own vessel Collision Avoidance Action Function (CAAF) is modeled as a Gaussian distribution
2
(k)) with the mean i (k) and
i (k) N (i (k), i
the covariance 2i (k). The CAAF with respect to the
CRF can be written as:
Collision risk functions
The Collision Risk Functions (CRF) of the own vessel (i (k)) due to the ith target vessel in the kth
time instant are modeled as a Gaussian distribu2
tion i (k) N (i (k), i
(k)), with the mean value
i (k) that is estimated by the average time until
collision, Ti (k), which can be defined as:
where i (k) N ( , 2 ), is the time delay function

that is approximated by a Gaussian distribution with
a constant mean, , and covariance, 2 . The conditional CRF with respect to the CAAF as a Gaussian
distribution can be written as:
where |OPi (k)| is the relative range and Vi,o (k) the
relative speed of the ith target vessel with respect to
the own vessel at the kth time instant. Furthermore, the
2
covariance i
(k) is considered in this distribution.
It is assumed that the CRF, i (k), can be obtained
from noisy Observation Function of Zi (k) and can be
written as:
where 3 i is the normalization constant. The prior

distribution of the CAAF due to the ith target vessel
at kth time instant can be written as a Gaussian
distribution:
where zi (k) N (0, z2 (k)), is a Gaussian observation

noise with the mean 0 and the constant covariance
z2 (k). Hence the prior distribution of the CRF due to
the ith target vessel at kth time instant can be written
as a Gaussian distribution:
where 1 i is the normalization constant. The transition

model of the CAAF considered as Gaussian perturbation of constant covariance 2 to the current states of
the CAAF and can be written as:
13
where 2 i is the normalization constant. Considering

the Bayesian Rule, the CAAF update from the CRF
can be written as:
5.5
Implementation of collision avoidance actions
The implementation of the accumulated CAAF, Ai (k)

is divided into two sections of Course Control (Ai (k))
and Speed Control (AVi (k)) CAAFs as presented in
Figures 4. The CAAFs are generated from the collision
avoidance decisions of Di (k), from the course control
decisions (Di (k)) and the speed control decisions
(DVi (k)) as described previously. Hence, the accumulated CAAFs of Course Control (A (k)) and Speed
Control (AV (k)) actions can be written as:
Figure 6. Simulations of multi-vessel Collision Situation.
These accumulated CAAFs are implemented in

the computational simulations of collision avoidance.
The further details on the Bayesian network based
sequential action formulation module including
mathematical derivation of the CRF and CAAF are
presented Perera et al. (2010c).
6
accumulated CRF is presented in the x = 12000 (m)

axis as the first peaks in a time instant. Furthermore,
the respective accumulated CAAFs of course and
speed, with respect to the collision situation are presented in the same Figure in the axis of x = 9000 (m)
and x = 6000 (m). The accumulated CAAFs, the
course to starboard and speed reduction to avoid the
first target vessel are also presented in the same Figure.
In Figure 7, the collision avoidance system has
observed the second possible collision situations
and the accumulated CRF is presented in the
x = 12000 (m) axis as the second peak in a different
time instant. Furthermore, the respective accumulated
CAAFs of course and speed, with respect to each collision situations are presented in the same Figure in the
axis of x = 9000 (m) and x = 6000 (m). The accumulated CAAFs, the course to starboard and speed
reduction, to avoid the second target vessel are also
presented in the same Figure.
Figure 8 presents the sub-completion of the first
action segment of the CAAFs, consisting in speed
reduction and continuation of course change to starboard side by the own vessel. However, due the first
action segment of the CAAFs, the partially reduced
collision actions (i.e. the speed reduction for second
collision situation no longer required) between the own
COMPUTATIONAL SIMULATIONS
The computational simulations for the multi-vessel

collision situation are presented in Figures 6 to 16.
As presented in Figures 6, the own vessel starts navigation from the origin (0 (m), 0 (m)) and the first,
second and third target vessels start from positions
(4238 (m), 10238 (m)), (8790 (m), 10000 (m)), and
(10714 (m), 12400 (m)) respectively. All startup and
final positions of the own and the target vessels are
presented by vessel shape icons at the kth time instant
in the Figures.
It is assumed that the target vessels are moving in
constant speed and course and dont honor any navigational rules and regulations of the sea. The CRF
assessment is formulated by a Gaussian distribution
and presented in the x = 12000 (m) axis. Similarly,
the CAAFs for the course and speed changes formulated by the Gaussian distributions are presented in
the x = 9000 (m) and x = 6000 (m) axis respectively. The scaled Time axis (Actual Time 5 (s)) is
presented in the y axis, and the scaled Collision Risk
(%) , Course Actions (%) and Speed Actions (%) are
presented in the x axis.
In Figure 6, the collision avoidance system has
observed the first possible collision situation and the
14
vessel and the second target vessel could also observed

in Figures 8 and 9.
The completion of the first action segment of the
CAAFs in Figure 10 and the own vessel is about
to safe-pass the first target vessel in Figure 11 are
presented. One should note that, due the first action
segment of the own vessel, the collisions situation with
the second target vessel was eliminated.
The own vessel is about to safe-pass the first target
vessel trajectory in Figure 12 and about to safe-pass
the second target vessel trajectory in Figure 13 are

presented.
Furthermore, the collision avoidance system
observing the third possible collision situation with
the respective accumulated CAAFs, the course to port
and speed reduction to avoid the third target vessel are
also presented in Figure 13.
Figure 14 presents the sub-completion of the third
action segment of the CAAFs, consisting in speed
15
CONCLUSIONS
This paper introduces a decisions formulation and

action execution process of collision avoidance in
maritime transportation. As presented in the computational results, the autonomous Fuzzy-Bayesian based
decision-action formulation process could be used to
avoid complex navigational conditions. The successful
results obtained in this study show the system capabilities of collision avoidance involving multiple vessels
under various collision conditions in ocean navigation while still respecting the COLREGs rules and
regulations.
ACKNOWLEDGEMENTS
The first author has been supported by the Doctoral
Fellowship of the Portuguese Foundation for Science
and Technology (Fundao para a Cincia e a Tecnologia) under contract no. SFRH/BD/46270/2008.
Furthermore, this work contributes to the project of
Methodology for ships manoeuvrability tests with
self-propelled models, which is being funded by the
Portuguese Foundation for Science and Technology
(Fundao para a Cincia e Tecnologia) under contract
no. PTDC/TRA/74332/2006.
REFERENCES
Anto, P. ]and Guedes Soares, C. 2008, Causal factors in
accidents of high speed craft and conventional ocean
going vessels, Reliability Engineering and System Safety,
vol. 93, pp. 12921304.
Benjamin, M. R., and Curcio, J. A., 2004, COLREGS
based navigation of autonomous marine vehicles, in
IEEE/OES Autonomous Underwater Vehicles, pp. 3239.
Benjamin, M. R., Curcio, J. A., and Newman, P. M., 2006,
Navigation of unmanned marine vehicles in accordance
with the rules of the road, in Proceedings of the 2006
IEEE International Conference on Robotics and Automation, pp. 35813587.
Chauvin, C., and Lardjane, S. 2008, Decision making and
strategies in an interaction situation: Collision avoidance
at sea, Transportation Research, vol. Part F, no. 11,
pp. 259262.
Cheng, X. D., Liu, Z. Y., and Zhang, X. T., 2006, Trajectory
optimization for ship collision avoidance system using
genetic algorithm, in OCEANS Asia Pacific, pp. 15.
Cockcroft, A. N., and Lameijer, J. N. F., 2001, A Guide to
The Collision Avoidance Rules. Burlington, MA. USA:
Elsevier Butterworth-Heinemann.
Guedes Soares, C. and Teixeira, A. P. 2001, Risk Assessment
in Maritime Transportation. Reliability Engineering and
System Safety, Vol. 74, pp. 299309.
Hasegawa, K., 1987, Automatic collision avoidance system
for ship using fuzzy control, in Proceedings of 8th Ship
Control System Symposium, pp. 234258.
Hong, X., Harris, C. J., and Wilson, P. A., 1999, Autonomous
ship collision free trajectory navigation and control
algorithms, in Proceedings of 7th IEEE International
Conference on Emerging Technologies and Factory
Automation, vol. 2, pp. 923929.
reduction and continuation of course change to port

by the own vessel.
Figure 15 presents the completion of execution
of the CAAFs in order to avoid the third target
vessel. Finally the completion of all the CAAFs
with zero CRF and safe passing of the third target
vessel trajectory are presented in Figures 16.
16
IMO 1972, Convention on the international regulations for

preventing collisions at sea (COLREGs),, URL http://
www.imo.org/conventions/.
Ito, M., Zhang, F., and Yoshida, N., 1999, Collision
avoidance control of ship with genetic algorithm, in
Proceedings of the IEEE International Conference on
Control Applications, pp. 17911796.
Lee, S., Kwon, K., and Joh, J., 2004, A fuzzy logic
for autonomous navigation of marine vehicles satisfying colreg guidelines, International Journal of Control
Automation and Systems, vol. 2, no. 2, pp. 171181.
Lisowski, J., Rak, A., and Czechowicz, W., 2000, Neural network classifier for ship domain assessment, Mathematics
and Computers in Simulations, vol. 51, pp. 399406.
Liu, Y., and Liu, H., 2006, Case learning base on evaluation
system for vessel collision avoidance, in Proceedings of
the Fifth International Conference on Machine Learning
and Cybernetics, pp. 20642069.
Kwik, K. H., 1989, Calculations of ship collision avoidance
manoeuvres : A simplified approach, Ocean Engineering, vol. 16, no. 5/6, pp. 475491.
Perera, L. P., Carvalho, J. P., and Guedes Soares, C., 2010,
Eds., Autonomous guidance and navigation based on the
COLREGs rules and regulations of collision avoidance,
1st ed., Advanced Ship Design for Pollution Prevention.
London, UK: Taylor and Francis Group.
Fuzzy-logic based parallel collisions avoidance decision
formulation for an ocean navigational system, in Proc.
8th IFAC Conference on Control Applications in Marine
Systems, Rostock, Germany, pp. 295300.
Bayesian network based sequential collision avoidance
action execution for an ocean navigational system, in
Proc. 8th IFAC Conference on Control Applications in
Marine Systems, Rostock, Germany, pp. 301306.
Perera, L. P. and Guedes Soares, C., 2010, Laser measurement system based maneuvering target tracking formulated by adaptive competitive neural networks, in In
Proc. 2nd International Conferrence on Adaptive and
Self-adaptive Systems and Applications, Lisbon, Portugal,
pp. 8490.
Perera, L. P., and Guedes Soares, C., 2010, Ocean vessel
trajectory estimation and prediction based on extended
kalman filter, in Proc. 2nd International Conference

on Adaptive and Self-adaptive Systems and Applications,
Lisbon, Portugal, pp. 1420.
Pietrzykowski, Z., and Uriasz, J., 2009, The ship domain a
criterion of navigational safety assessment in an open sea
area, The Journal of Navigation, vol. 63, pp. 93108.
Rothblum, A. M., Wheal, D., Withington, S. , Shappell,
S. A. , Wiegmann, D. A., Boehm, W., and Chaderjian,
M., 2002, Key to successful incident inquiry, in 2nd
International Workshop on Human Factors in Offshore
Operations (HFW), pp. 16.
Sato, Y., and Ishii, H., 1998, Study of a collision-avoidance
system for ships, Control Engineering Practice, vol. 6,
pp. 11411149.
Smeaton, G. P., and Coenen, F. P., 1990, Developing an
intelligent marine navigation system, Computing &
Control Engineering Journal, pp. 95103.
Smierzchalski, R., and Michalewicz, Z., 2000, Modeling
of ship trajectory in collision situations by an evolutionary algorithm, IEEE Transactions on Evolutionary
Computation, vol. 4, no. 3, pp. 227241.
Statheros, T., Howells, G., and McDonald-Maier, K. 2008,
Autonomous ship collision avoidance navigation concepts, technologies and techniques, The Journal of
Navigation, vol. 61, pp. 129142.
Sutulo, S., Moreira, L., and Guedes Soares, C., 2002,
Mathematical models of ship path prediction in manoeuvring simulation systems, Ocean Engineering, vol. 29,
pp. 119.
Moreira, L., and Guedes Soares, C., 2003, Dynamic model
of maneuverability using recursive neural networks,
Ocean Engineering, vol. 30, no. 13, pp. 16691697.
Yang, S., Li, L., Suo, Y., and Chen, G., 2007, Study on
construction of simulation platform for vessel automatic
anti-collision and its test methods, in IEEE International
Conference on Automation and Logistics, pp. 24142419.
Zeng, X., Ito, M., and Shimizu, E., 2001, Building an automatic control system of maneuvering ship in collision
situation with genetic algorithms, in Proceedings of the
American Control Conference, pp. 28522853.
Zhuo, Y., and Hearn, G. E., 2008, A ship based intelligent anti-collision decision-making support system utilizing trial manoeuvers, in Chinese Control and Decision
Conference, pp. 39823987.
17

The importance of satellite-based vessel monitoring system (VMS) for

fisheries management: A case study in the Portuguese trawl fleet
T. Pilar-Fonseca, A. Campos, P. Fonseca, B. Mendes, V. Henriques & J. Parente
INRB, IP/IPIMAR, Lisboa, Portugal
M. Afonso-Dias
Centro de Investigao Marinha e Ambiental, Universidade do Algarve, Faro, Portugal
ABSTRACT: Satellite-based Vessel Monitoring System (VMS) data, consisting of vessel location and speed,
have been used for fisheries control. However, the usefulness of these data can be extended to fisheries research,
as the outputs provided by this system make it possible to map and quantify the fishing activity. One possible
application is the estimation of effective fishing effort. Moreover, the cross analysis of effort with the corresponding vessel landings, allows the estimation of landings per unit effort (LPUE), used as a proxy of stock
abundance in fisheries management. VMS data with 10-minute interval was processed for a Portuguese coastal
trawler operating off the west coast, using GeoCrust 2.0, a geographical information system which identifies
fishing trips and towing events (hauls) within trips. This information was then used to characterize vessel activity.
It is demonstrated how high-resolution VMS data allows to understanding fishing tactics and derive more precise
series of effort estimates.
1 VMS IN PORTUGUESE FISHERIES
with the corresponding catches from vessel logbook,

or with vessel landings, allows the estimation of landings per unit effort (LPUE), often used as a proxy of
stock abundance in fisheries management.
For each vessel, a series of trajectories can be produced by mapping the geo-referenced information,
each defining a fishing trip, starting at a fishing port
and ending at the same or in other port. These trajectories will tend to form patterns consisting in a
succession of points, closer or more distant according to the slower or faster vessel speed. In addition to
providing geographical location of the fishing activity, depending on the amount of data, these patterns
can also identify and characterize the main operational
phases of a trip.
Satellite-based VMS information is available for all
Portuguese coastal trawlers operating mostly in Western Iberian waters, ICES division IXa (DGPA, 2010),
with occasional trips in divisions VIIIc and IXb. A
total of 83 vessels were active at the end of 2009, corresponding to 15385 GT (gross tonnage) and 69737
of engine power (kw). During this year, this fleet
accounted for 12 and 17% approximately of the total
national landings in weight and value, respectively.
Fishing pressure is high, the fishing activity taking
place all year round and exploiting a large number
of fish, cephalopods and crustaceans. These include
pelagics such as horse mackerel (Trachurus trachurus), mackerels (Scomber scombrus and Scomber
japonicus), and blue whiting (Micromesistius poutassou), demersal species such as octopuses (Octopus
vulgaris and Eledone cirrhosa) and European hake
The Portuguese vessel monitoring system (MONICAP), created in 1987, was the first implemented
in Europe, being presently operating in vessels with
length overall (LOA) higher than 15 meters (EC,
2003). Global Positioning System (GPS) data for
each vessel, consisting of a succession of geographical locations (latitude, longitude), dates, times, speed
and course, are received by a blue-box (satellitetracking device installed on board Community fishing
vessels), recorded and automatically transmitted via
satellite (Loran-C) to the Portuguese Fishery Directorate (DGPA/IGP), the national monitoring authority responsible for marine surveillance. Until 2004,
records were obtained each 10 minutes, but since then,
according to European Union (EU) regulations the frequency of data transmission may be at least once every
two hours, if the national Fisheries Monitoring Center
(FMC) has the possibility of polling the actual position
of the fishing vessels (EC, 2003), which is the case of
Portugal.
These geo-referenced data have been exclusively
used with the purpose of fisheries control, in particular to check whether vessels are operating outside the
six nautical miles line off the coast as required by law.
However, the usefulness of MONICAP data can be
extended to fisheries research purposes, as the analyses of data provided by this system makes it possible
to map and quantify the fishing activity. One possible application is the estimation of series of effective
fishing effort. Moreover, the cross analysis of effort
19
(Merluccius merluccius). A different fleet component exploits deepwater crustaceans including the rose
shrimp (Parapenaeus longirostris) and the Norway
lobster (Nephrops norvegicus).
Herein, we demonstrate the utility of satellite-based
VMS information to map the activity of a Portuguese
coastal fish trawler mainly involved in the capture
of horse mackerel and cephalopods, identify fishing
tactics and estimate effective effort.
DATA ANALYSIS
A collection of unprocessed VMS data corresponding

to a 12-month period of activity (2003) was selected for
this trawler. The spatial pattern of the vessel activity is
displayed in Figure 1, showing that the vessel operated
over approximately 193 nautical miles in the western coastal area between Matosinhos and Figueira,
between 40.0 and 41.6 N in latitude and 8.7 and
9.5W in longitude, with only a few trips south of this
area. Operational fishing depth was between 30 and
200 m.
Criteria for filtering these data for effort estimation purposes are primarily based on the separation
between fishing from non-fishing points, according
to each vessel speed profiles. The speed-frequency
histogram for trawlers typically displays the trimodal
pattern in Figure 2, where the speed profile for the
vessel in study is shown using the raw data corresponding to 2003. The second low speed peak will
correspond to fishing activity (trawling), in this case
between 3 and 5 knots approximately, while the third
peak corresponds to faster movements (e.g., steaming,
9 to 11 knots). Low speeds of pre-processed data can
also correspond, for instance, to steaming near port or
shooting the trawl.
The geo-referenced information, consisting of trajectories with 10-minute intervals between records,
each corresponding to one fishing trip (FT), was
then processed using GeoCrust 2.0 (Afonso-Dias
et al., 2004). This dedicated geographical information system (GIS) was designed to analyze fleet
spatial dynamics through semi-automatic identification of the fishing trips and trawl hauls, thus requiring
expertise and time from an operator.
Once the identification process by GeoCrust is
complete at the haul level, the haul characteristics
(i.e. duration, distance, and geographical location)
are defined. The operator classifies each trip with
a two-level quality index, indicating if the original data consists of high/low quality data regarding
completeness. Fishing trips with incomplete VMS
data are considered of low quality (partially valid
fishing trip) since this can affect haul identification,
and therefore, the effective effort estimates.
GeoCrust identified a total of 849 hauls (corresponding to 11563 points) within 128 high-quality
FT (21270 points). These FT are analyzed and interpreted to identify the main operating phases: steaming
from port to fishing ground; towing; steaming between
Figure 1. Map of the geographical distribution of vessel

activity (unprocessed data). Each point represents a position
of the vessel during 2003. Approximately 25000 VMS points
(with 10-minute interval) were used.
Figure 2. Speed profile (nautical miles/hour, x-axis) for

the vessel in study using pre-processed VMS information
for 2003.
fishing grounds; and steaming back to the port. For

practical purposes, only two main activities were considered, trawling and steaming. Accurate computation
of fishing effort estimates were obtained from the
system including trip and hauls duration, as well as
distances trawled (nautical miles, nm).
Within this fishery, a number of different species
can be targeted by the same vessel. The methodology to estimate species directed effort involved the
cross analysis of effort with landings from individual fishing trips. The trips were previously classified
(Pilar-Fonseca et al., 2009), in a number of distinct
groups (landing profiles, LP) according to the main
target species landed. Effort and abundance indices
(landings per unit effort, LPUE) by LP could then be
obtained for fisheries management purposes.
20
Figure 4. Speed profile (y-axis) along time (x-axis) for one

fishing trip obtained from the GeoCrust system. Trawling
activity correspond to low speed, identified in black. This
54-hour trip started at around 5 am and ended at around 10 am
of the third day.
Table 1. VMS-based effort and landing per unit effort

(LPUE) estimates by landing profiles (LP) for the vessel
in study high quality data processed by GeoCrust 2.0. LP
named according to Pilar-Fonseca et al. (2009). LPUE is kg
landed/hour. Standard deviations in parenthesis.
Effort and operational

parameters
Total number of FT
Number of hauls*
Trip duration (h)*
Figure 3. a) Frequency distribution of starting and ending

time (black and gray, respectively) of trips by hour (x-axis).
b) Relationship between trip and haul duration (total hours
trawled and hours at sea) for the vessel in study during 2003.
Total number of trips identified: 128.
Number of hours
trawled per trip*
Number of hours in
navigation*
Speed in navigation* (kt)
Trawling speed* (kt)
Total trip duration (h)
Total number of hours
trawled
Total number of hauls
Proportion of trawling
time within trips*
RESULTS
During 2003, the trawler in study presented relevant

activity for three distinct LP, one of them targeting
exclusively horse mackerel (LP5), a second one where
horse mackerel is targeted along with bentho-pelagic
species (LP3), and a remaining one targeting benthic
species, with relevance to the Octopodidae (LP1).
Trips usually start before dawn, and the vessel steam
offshore for a variable period, depending on the distance to fishing grounds (Figure 3). When reaching
the fishing area, speed is reduced and the trawl is set
at low speeds. Setting, as well as hauling-back, are
operations limited in time. However, these operations
could be identified in this case (Figure 4) due to the
high data quality and completeness.
Total hours trawled by trip varied between 2 and
48 hours, with mean haul duration of 125 minutes
(s.d. = 54 minutes). For this vessel, trawling speeds
(at the trip level) varied between 3.3 and 4.4 knots
depending on the target species, after which haul-back
is initiated with the vessel stopped. The number of
hauls per trip varied between 1 and 21, with average number of 7. Hours at sea (trip duration) varied
between 5.6 and 76.1 (average = 22.7 hours) with
average steaming and trawling speeds estimates
around 6.4 and 4.0 knots (Table 1).
For the vessel in study, geographical location of
fishing activity was not found to differ much between
LP. However, differences in trip duration are evident,
with longer trips when benthic species are targeted
(LP1), progressively reducing towards LP5 and LP3,
LPUE
Octopodidae LPUE*
(kg/h)
Horse mackerel LPUE*
(kg/h)
Total landings in weight
per trip* (kg/trip)
Total landings in value
per trip* (Euros/trip)
Total landings in value
per hours trawled*
(Euro/h)
LP1
LP3
LP5
Total
53
8
28.1
(14.9)
18.9
(10.9)
9.2
35
5
15.7
(9.3)
8.1
(5.5)
7.6
26
6
22.4
(18.9)
12.5
(12.4)
9.9
128
7
23
(15.0)
14.0
(11.0)
9.0
5.9
3.8
1488
1001
7.0
4.1
549
282
6.6
4.0
582
326
6.0
4.0
2899
1770
430
0.65
172
0.50
168
0.53
849
10.9
5.6
4.7
7.0
6.2
61.8
130.3
49.0
886
866
1660
1034
2376
1849
3170
2364
58
140
170
108
*Average estimates.
where the main target is horse mackerel (Table 1).

This is possibly related to higher catches in the latter,
leading to more frequent landings. The proportion of
hours trawled by trip was found to be related with
trip duration, with about 65% of the time spent trawling in LP1 versus around 50% in the remaining LP.
Haul duration was found to be longer in LP1 targeting octopuses when compared to those targeting
pelagic fish.
21
Differences in trawling speed are also noticed

between LP targeting pelagic and benthic species,
although the magnitude of these differences is less
than 0.5 knots. Average trawling speed was found to
be around 4.0 knots for pelagic, while for benthic fish
it corresponded to 3.8 knots (Table 1).
Three different FT considered as representative of
the three LP above (LP1, LP3 and LP5), for which
observed effort approached the average LP effort estimates, were chosen (Figure 5). The figure illustrates
the existence of different fishing tactics. The vessel
activity in the first trip (LP1), where benthic species
were targeted, extends over a wide geographical area,
with a large number of hauls carried out at two different
sites following distinct isobaths. The remaining trips,
directed to pelagic species (LP3 and LP5), are shorter
with lower number of hauls, and spatially located in
fishing grounds at higher depths, being more distant
from the coast. Short duration hauls were carried out
at a higher trawling speed, which is directly related to
horse mackerel behavioral characteristics.
4
FINAL CONSIDERATIONS
By providing a more reliable account of effective

towing time, the use of VMS data allows for a precise estimation of fishing effort, and therefore of
LPUE, thus contributing for better resource management. Even considering the major step arising from the
use of VMS-derived effort data, reliability of LPUE
estimates could be further improved if a trustworthy
relationship between catch and individual hauls were
established. However, at present time, the accuracy
of the information contained in logbooks is, at the
least, questionable. The future adoption, within the
EU, of electronic logbooks, may contribute to partially
solve this problem, although the humanfactor
the correctness of entries by the skipper will persist.
The use of geo-referenced data can be extended to
address fishermen behaviour and fishing impacts on
the ecosystem, allowing for the integration of spatial
planning in fisheries management schemes. However,
the financial constraints related to data transmission
(satellite communication fare) that led to the adjustment of transmission rate from 10-minute to 2-hour
interval, will compromise the use of more recent VMS
data for the above purposes. In this fishery, where
mean tow duration is, as it was stated above, of 125
minutes, a 2-hour-interval between the consecutive
reception of VMS data points will preclude the correct
identification of single tows.
Figure 5. Map of the VMS information (trajectories) processed by GeoCrust at the trip level for three FT corresponding to the three main different LP. For trip 1, 2 and 3 targets
are: octopuses and benthic species; horse mackerel and
bentho-pelagic; horse mackerel. Trip duration: 38.1 h; 11.1 h;
14.9 h. Hours trawled; 25.8 h (68% of trip); 4.1 h (76%); 7.0 h
(47%). Trawling speed: 3.9 kt; 4.33 kt; 4.22 kt. Navigation
speed: 6.6 kt; 8.5 kt; 7.8 kt. A total of 12, 3 and 4 hauls were
identified. Black dots correspond to trawling activity.All trips
began and ended in the fishing port of Aveiro.
ACKNOWLEDGEMENTS
The fisheries data was supplied to IPIMAR by
the Portuguese General-Directorate for Fisheries and
Aquaculture (DGPA). The processed VMS data was
provided by the University of Algarve (GeoPescas
research project). This work is the continuation of
previous work under the Programme MARE: Fishing
22
Analyses in Fishery andAquatic Sciences (Vol. 2). FisheryAquatic GIS Research Group, Saitama, Japan, 323340.
DGPA, 2010. Recursos da pesca. Srie estatstica, Vol. 23AB, ano 2009. Direco-Geral das pescas e Aquicultura,
Lisboa, Julho 2010, 181p.
EC, 2003. Commission Regulation No. 2244/2003 of 18
December 2003, laying down detailed provisions regarding satellite-based vessel monitoring systems. Official
Journal of the European Union 2003. L333:1727.
Pilar-Fonseca, T., Campos, A., Afonso-Dias, M.,
Fonseca, P., & Mendes, B. 2009. Fleet segmentation of
the Portuguese coastal trawl fishery: a contribution to
fisheries management. ICES CM 2009/O:29.
Technologies (MARE, FEDER, QCA-III, 22-05-01FDR-00114: 2000-2007).

It was partially funded by the Portuguese Fundao
para a Cincia e a Tecnologia, through a PhD grant
(SFRH/BD/43409/2008) attributed to the first author.
REFERENCES
Afonso-Dias, M., Simes, J., & Pinto, C. 2004. A dedicated GIS to estimate and map fishing effort and landings
for the Portuguese crustacean trawl fleet. In T. Nishida,
P.J. Kailola, & C.E. Hollinworth (eds), GIS/Spatial
23

An algorithm for minimum-time ship-routing for the

North Indian Ocean region
Debabrata Sen
Department of Ocean Engineering & Naval Architecture, IIT Kharagpur, India
Chinmaya P. Padhy
WIPRO, Kolkata, India
ABSTRACT: This paper presents development of a ship weather routing algorithm for determining minimumtime route. The optimization is based on a form of Dijkstras algorithm. The developed algorithm is investigated
using various realistic wave data for the North Indian Ocean region obtained from the 3rd generation WAM
model. Illustrative minimum-time sea routes on Arabian sea and Bay-of-Bengal have been determined and
presented. The algorithm permits incorporation of all relevant practical and realistic constraints such as presence
of land boundaries, consideration of non-navigable water, effects of wind and current, voluntary speed reduction
due to multiple seakeeping criteria within the framework of the optimization scheme.
INTRODUCTION
and Lacey (1998), Ulusoy (2004) and Montes (2005)

among others. However, it must also be mentioned here
that although most of these works focus on the optimization algorithm using advance mathematical models, their use for developing an operational algorithm
for practical application are not often demonstrated
and thus their actual application remains somewhat
unclear. One of the reasons why the exact algorithm
and its application for practical ship-routing works are
not usually discussed in open literature is due to its
commercial value.
In this paper, taking advantage of these modern
developments in wave modeling, an optimum track
ship routing algorithm for ships operating in the Indian
Ocean region is described by using a network graph for
the Indian Ocean. A binary heap version of modified
Dijkstras algorithm has been used to determine the
optimum route given an input from atmospheric/wave
model and basic ship response function. The algorithm
is then applied for determining optimal routes along
the Indian coastline for coastal ship operations. Sample results presented demonstrate that the algorithm
is capable of considering a majority of the constraints
that are necessary for practical application.
Ship routing is concerned with determining an optimal track for ocean voyages based on wave and
weather forecasts of weather, and the vessels seakeeping characteristics. There are a number of different
approaches to calculate the shortest or in some sense
optimal path/route between two nodes representing
the start and destination ports. Within specified limits of weather and sea conditions, the term optimum
may be defined to mean maximum safety and crew
comfort, minimum fuel consumption, minimum time
underway, or any combination of these factors.
It is evident that development of optimal shipweather-routing requires a combination of the following three different areas:
i. Forecasting the sea-conditions (i.e. ocean-state
forecast)
ii. Estimating ship behavior in such ocean wave
conditions
iii. Development of an appropriate and efficient track
or path optimization algorithm.
The last aspect above, i.e. a track-optimization algorithm wherein some objective function is optimized
(maximized/minimized) combines the first two areas.
The optimal ship trajectory problem, or the
so-called problem of weather-routing of ships is a
fairly old problem, and quite a few works can be
traced over the past half a century. A glimpse of
the works done over the past 50 may be seen in
the works of Hanssen and James (1960), Haltiner
et al. (1962), Faulkner (1962), Zoppoli (1972), Chen
(1978), Mitchell and Papadimitriou (1986), Hagiwara
(1985,1989), Perakis and Papadakis (1988), Chen
BACKGROUND AND FORMULATION
Without loss of generality, the problem of optimal

ship routing reduces to optimization of certain objective function which in most cases can be related to a
reduction in speed under the following heads:
25
Involuntary speed reduction

Voluntary speed reduction
if a start and end (destination) nodes are defined, then

the algorithm finds the optimal path joining these
two nodes through grid points such that the total weight
is minimized.
Involuntary speed reduction is due to the increased

resistance in a seaway while voluntary speed reduction is the deliberate reduction in speed by the ships
captain in order to ensure that the ships wave-induced
responses are within acceptable safety limits, since it
is found that in general a reduction in speed (and also
heading to some extent) improves sea-keeping (i.e.
reduces motions). The optimization algorithm to be
used should be general enough so that it can handle
variety of multiple objective criteria. The optimal-path
problem is a problem of interest in many fields of
study, e.g. traffic engineering. A literature search on
general path-optimization algorithm reveals that one
of the available and easy to implement optimization
technique in this context is the Dijkstras algorithm,
which tries to minimize the distance between any two
node points in a given mesh/grid (see e.g. Cormen
et al. 2001). The distance can also be replaced by
any weight function. In the case of the ship routing
problem, the weights can be viewed as an objective or achievement function, which can be obtained
by combining weather information along with seakeeping characteristics of the hull. It was thus felt that
this algorithm could be successfully applied for the
present ship-routing problem.
Dijkstras algorithm finds the shortest path from a
point in a graph to a destination. It finds routes by
cost precedence. The algorithm begins at a specific
node and extends outwards. It can be used to calculate the shortest path between any two vertices in a
weighted graph, where each edge has non-negative
edge weight. Although most applications of shortest
path involve graphs with positive edge weights, modified algorithms (e.g. Prims or Kruskals algorithm)
are available where such a constraint on weights is not
necessary. For our case, the weight will never be negative (it can at best be very small), and therefore the
original form of Dijkstras algoithm appears adequate.
Dijkstras algorithm creates labels associated with
nodes. These labels represent the distance/time (cost)
from the source node to that particular node. Within the
graph, there exist two kinds of labels: temporary and
permanent. The temporary labels are given to nodes
that have not been reached. The value given to these
temporary labels can vary. Permanent labels are given
to nodes that have been reached and their distance/time
(cost) to the source node is known. The value given
to these labels is the distance (cost) of that node to
the source node. For any given node, there must be a
permanent label or a temporary label, but not both.
The algorithm begins by setting the permanent label
of the start node to 0 and the temporary labels of all
other nodes to 0. The algorithm then proceeds to select
the least cost edge connecting the node with a temporary value with the start node. The new node then has
its label changed to a permanent value, taking into
account the distance/time (cost) of travel up to this
node. This process continues until all the nodes of the
graph have permanent values set.
Based on this algorithm, a code has been developed,
where in a given mesh (grid) with associated weights,
2.1 Determination of the Weights

In order to apply Dijkstras algorithm for the ship routing problem, an area of the sea-surface encompassing
the possible route of the ship need to be discretized
by means of a grid formed by latitude and longitude
lines. The nodes or vertices can be taken as a central point in the grid or alternatively the intersection
of each latitude and longitude line can be considered
as a node. Typically the weather information (significant wave height and characteristic wave period) will
be available from satellite generated data and advance
wave modeling methods like WAM at each grid. As in
WAM output, these values are assumed constant over
the corresponding grid.
The next and most important task is now to determine an appropriated weight function for each grid
or node by combining the wave conditions (significant
wave height and direction). Once these weights are
known at each node/grid, the problem is now transformed to a state where Dijkstras algorithm can be
applied to find the optimal path for minimum (or
maximum) weight.
The weights wi, j between the path lines joining
adjacent nodes i, j will depend on the parameter that
need to be optimized. There can be several possible
parameters such as minimum travel time, minimum
fuel consumption, safe and comfortable travel etc.
In the present work, we will consider the optimum
ship routing for minimum travel time and discuss the
methods to determine the weight function based on
involuntary speed reduction. For this we first need to
determine the reduction in speed due to the added drag.
The speed of the self propelled ship in calm water is
given by the so-called self propulsion point. This is the
speed at which the thrust T , after taking into account
the thrust deduction factor t, i.e. T (1 t) equals the
calm water resistance RSW . The speed at which the
ship will travel for the total resistance RT including
the additional resistance from all other environmental sources (wave, wind, current) will similarly be at
the speed where RT equates T (1 t). This is schematically shown in figure 1 where V and VR are the calm
water and reduced speed. For a ship in operation, for
the given engine setting V is known. Therefore the
objective will be to determine VR given its calm water
speed V at a given engine setting (i.e. revolution per
unit time N ).
The thrust T produced by a propeller depends
on its propulsion characteristics and are usually represented by means of KT KQ J curves,
where KT = T /(N 2 D4 ) is the thrust coefficient,
KQ = Q/(N 2 D5 ) is the torque coefficient, and
J = Va /(ND) is the advance coefficient. Here Q is
torque, D is propeller diameter and Va is the velocity
26
In the above, S is some representative area of the hull

which is in proportion to the square of its linear dimension and C T can be defined as the total equivalent
resistance coefficient. In such a case it is possible to
directly determine the speed VR as follows:
or,
Noting that J = Va /(ND) and Va = (1 w)V , i.e.

V = Va /(1 w), we have
Figure 1. Illustration for self-propulsion point.
of advance, which is the velocity of water past the propeller. Va is related to ship speed by Va = (1 w)V
where w is the wake fraction for the hull.
For a given ship in operation at a given engine
setting, the parameters N , D, t, w are fixed. Thus
J is a direct function of V . Therefore effective
thrust T (1 t) = KT N 2 D4 (1 t) as a function of V
can be plotted if the propulsion characteristics, i.e.
KT KQ J is available. The ship speed will then be
given by the intersection of the effective thrust line and
the resistance line, as shown in Figure 1.
The above procedure is the most proper way to
determine the speed given the total resistance. However, for the present purpose of ship routing application, it has the following difficulties. It needs to
be recognized that the speed has to be determined
between each adjacent grids for all possible path lines
connecting the grids. This will a very large number,
and this number will increase rapidly if the navigational domain is large (i.e. the possible area of the sea
within which the route is expected to lie, e.g. for large
ocean crossings). The other issue is that for each path
line, the value of RT will be different depending on
the prevailing environmental conditions even if RSW is
constant. Thus for each possible path line, one has to
determine the curves (1 t)T and RT against speed,
and then determine the intersection of these two lines
to find VR . This procedure, besides needing the propulsion characteristics in terms of KT KQ J curves,
is therefore extremely time and computer intensive,
and may not be feasible for the present application.
We therefore need to consider some simplified and
approximate methods to determine wij , which is fast
and yet gives the weights within practically acceptable
limits of accuracy.
Before going to simplified methods, we note that
KT J curve can usually be approximated by a straight
line of the form:
or,
The above is a quadratic equation for Va :
with
Thus,
It can be shown that the admissible root for Va will

correspond to the positive sign in the numerator of (9).
Thus, the ship speed at which thrust (with the thrustdeduction factor) and total resistance are same (i.e. the
desired reduced ship speed) will be Va /(1 w).
The above procedure therefore will give directly the
desired reduced speed VR for a given total resistance.
One problem in this procedure is however in (2), i.e.
expressing RT in this form and more importantly finding an equivalent total resistance coefficient C T . While
it may be possible to come up with a value of C T if we
consider added wave resistance and current load, it is
difficult to consider wind drag in this form since the
concerned area over which the wind load applies is the
upper hull above waterline, not related to the wetted
hull as in the other resistance components. The other
issue is again the need to have propulsion characteristics as well as the time of computation for the large
number of weights to be evaluated.
Additionally, assume that it is possible to represent the

total resistance RT in the form:
27
Figure 3. Procedure to get reduced speed assuming constant

thrust.
Figure 2. Procedure to obtain reduced speed for constant

effective power.
wind data. Further, it also needs to be noted that prevailing weather conditions are all assumed constant
over a grid, which itself is an averaging process introducing some inaccuracy. It also needs to be noted that
the error introduced in assuming thrust to remain constant will have similar order of error for all possible
paths, and therefore the determined optimal path will
still remain optimal or near optimal. Thus determination of the reduced speed based on the assumption
that effective thrust is constant over the small range of
speed around the prevailing calm water speed may be
acceptable for practical calculations.
Based on the above assumption, the reduced speed
will be as depicted in Figure 3. If we write
We therefore attempt to devise an approximate

method of determining the reduced speed for each possible path-line which is simple yet within practically
acceptable limit of accuracy as far as application to the
ship weather routing is concerned.
The reduction in speed due to environmental factor is expected to be not very large, at least for most
part of the voyage. As a result, it may be reasonable to assume that over the applicable speed range
(i.e. over the range VR to V ), the propulsion characteristics remain same. This means, the effective
power PE is assumed constant. In calm water, PE is
given by RSW (V ) V , but in the presence of waves,
winds and current, the effective power will be given
by RT (VR ) VR = (RSW (VR ) + Radd (VR ))VR . Here Radd is
the resistance due to winds, waves and current, i.e.
Radd = RAW + RW + RC . This is illustrated in Figure 2.
In order to determine the reduced speed from this, we
note from (3.8) that the calm water resistance can be
expressed in the form aV 2 where a = 0.5SCSW . As
regards the additional drag, if an assumption is made
that the power required for this component at speed VR
is same as the power required at speed V , i.e. AB = CD
in Figure 2, then we get,
then
If a database is created by computing RT (V ) against

V for different values of k, then for any given k, one
can determine the corresponding speed V . Hence by
a process of interpolation, for a given additional resistance value, one can determine the reduced speed VR ,
as shown in figure 3. This procedure is however based
on expressing the additional resistance as k RSW . This
means, the additional resistance due to environmental factors are all a function of calm water resistance
regardless of speed, which means it has the implicit
assumption that the speed dependence of Radd is same
as that of RSW . Instead of this, it may be more reasonable to assume that Radd (VR ) = Radd (V ), as illustrated
in Figure 2. In such a case, if we have RSW expressed
as aV 2 , we can directly get the reduced speed as:
From above, we can get the reduced speed as:
To make further simplification, we now made an adhoc assumption that the effective thrust, i.e. (1 t)T
does not vary much over the range VR to V , and
therefore we can take it as constant as far as determination of reduced speed VR is concerned. We note that
the procedure to determine the total added resistance
are all based on approximate theoretical and/or semiempirical formulations, and therefore there is quite
some uncertainty in these values.
Additionally, there will also be some uncertainty in
the forecast weather information like wave data and
Both the approximate procedures given by (11) and

(14) will require determination of the additional resistance Radd arising from wind, waves and currents
for each possible path line between adjacent grids
28
using the prevailing environmental conditions at that

location, but for the given calm water ship speed V .
Once the reduced speed for each possible path line
joining nodes i and j is determined, i.e. VRi,j is found,
it is straightforward to determine the weight functions
as the time ti,j taken to cover the distance between the
nodes i and j given by Li,j :
Thus, by optimizing (minimizing) wi,j for the complete path, the minimum time travel route for the ship
can be found. Here the route optimization due only
for involuntary speed reduction, that is, due to additional drag arising from environmental factors have
been considered. It may be noted that in case of calm
water, w(i,j) reduces to t0 (i, j) = Li,j /V , which is the
travel time taken at constant calm water resistance, and
the minimum-time route becomes the shortest route.
In the present work, both (11) and (14) are used,
and it is found that the difference in the predicted
path from these two is marginal. Therefore finally the
approximate formula (14) is retained, as it is found to
be the most convenient and easy to apply.
The procedure can now be summarized as follows:
Figure 4. Route between Dammam Mumbai and

Mumbai Dammam by Ship 1.
Read at each grid the input wave conditions (significant wave height Hs , some typical spectral
period parameter like peak period or average zeroupcrossing period, and direction w ); also read wind
and current data (speed and directions).
Read the initial steady speed (V ) of the vessel.
For each path joining adjacent nodes i and j,
determine the ship heading angle s .
Determine for each grid, the relative heading angle
= s w .
Determine the added resistance Radd for speed V .
Here RAW which is the most significant component can be determined from interpolation from the
database, and the other components (RW and RC ),
if these apply, can be directly computed.
Determine the distance L between the nodes i.e. for
the path joining i and j. This is easily achieved from
the known lat.-long. values of the nodes.
Determine the weights for each grid from eqn. (11)
or (14). Note that this is the weight considering only
the involuntary speed reduction.
Figure 5. Route between Dammam Mumbai and

Mumbai Dammam by Ship 1.
run using the NCMRWF (National Centre for Medium

range Wave forecast, at Delhi) wind field data. Ship 1 is
a relatively small vessel of length 60 m, breadth 11 m
and draft 2.9 m, while ship 2 is of length 160.93 m,
breadth 23.1 m and draft 9.07 m.
Figure 4 show the route between Damman and
Mumbai, and the return route of Mumbai-Damman for
ship 1, while for the same routes for ship 2, results are
shown in Figure 5. The comparison of the results for
these two ships is depicted in Figure 6. These results
show that the routes depend on the ship type, and also
on their direction with respect to the waves.
In order to demonstrate the ability of the algorithm
to generate route circumnavigating a land mass, in Figure 7 we show the route from Mumbai on the west
coast of India to Visakhapatmen on the east coast of
India. The result is for ship 1. The algorithm here
simply assigns very large wave heights to those grids
which are land masses, As can be seen, the algorithm
could produce a route going around the land mass.
However, there is a problem with this track: this route
goes through the Palk Strait, the water between India
and Sri Lanka. This Strait is not open for navigation,
and therefore ships are not allowed to go through it.
RESULTS AND DISCUSION
A large number of results have been previously published for demonstrating the working of the algorithm,
and its correctness in achieving the minimal time route
(e.g. Padhy 2010), and thus these are not repeated here.
Here we apply the algorithm to determine minimumtime optimal route for two ships termed Ship 1 and
Ship 2 for routes lying in the North Indian Ocean
region, for which realistic wave data are generated
using the 3rd generation WAM. The wave model is
29
Figure 6. Comparisons between ship 1 and ship 2 routes.
Figure 9. Route between Adan Chennai based on WAM

climatological wave data for January.
Figure 7. Route between Mumbai-Visakhapatnam.
Figure 10. Route between Adan Chennai based on WAM

climatological wave data for July.
Palk strait is treated as navigable water. As can be

clearly seen, the routes generated are quite different.
Such studies are therefore for great use for planning
routes based on time of the year for the voyage. Similar result for a route lying in Bay of Bengal shown in
Figure 11 reveals that here the routes between winter
and monsoon months are not as large. This is mainly
because the region through which the route passes do
not show large change in the wave heights, although
in other parts of the ocean there are large changes in
wave heights.
One problem for this algorithm is that the route
produces often shows a zig-zag saw-tooth type nature
lacking smoothness, e.g. see Figure 7. This is a result
of generating the route by adding the nodes at the
centre of the grids. A smoothing-scheme is therefore devised to handle this problem, and the results of
applying this scheme is shown in Figure 13. It can be
seen that the generated track is now smooth.
The algorithm, as developed is also capable of
handling presence of current and wind fields: these can
Figure 8. Kolkata-Mumai route avoiding Palk-Strait.
Such constraints are also easily handled by thealgorithm, by simply blocking selected areas of ocean.
This is achieved by simply taking such areas as part of
land mass. Thus isolated areas on open ocean which
are not open to navigation can be treated as islands by
simply assigning very large wave heights associated
with those grids. Result shown in Figure 8.
Figure 9 and 10 show the result from the algorithm where the monthly average wave-data is used.
Although the algorithm is run for all twelve months,
here for brevity we show results for only two months,
January and July, the former being a winter month
with relatively calm sea and the latter being a monsoon month with high sea conditions. Note that here
30
Figure 14. Optimum Cochin Port Blair route with waves

and current.
Figure 11. Route between Chennai George Town based

on WAM climatological wave data for January.
Figure 15. Illustration for consideration of voluntary speed

reduction.
field. In a similar way, the effect of wind drag can also

be considered.
Finally the algorithm is extended for considering
voluntary speed reduction, in addition to the involuntary speed reduction in planning the route. To
account for the voluntary speed reduction within the
framework of the developed algorithm, the maximum
attainable speed at each grid considering both voluntary and involuntary speed reduction needs to be
found. This is schematically shown in Figure 15.
Let VRI represent the reduced speed due to involunV
tary speed reduction (due to added wave drag), VR1
and
V
VR2
represent the maximum speed limits for response
criterion 1 and 2 respectively. For wave height H1 , the
V
V
speeds VRI , VR2
and VR1
are given by points A, B, C
V
V
respectively, and since VR1
, VR2
> VRI , the net reduced
speed for this height will be VRI . For wave height H2 ,
the corresponding points are respectively D, E, F and
V
V
V
since here VR2
< VRI but VR1
> VRI , VR2
, the speed limit
V
here is VR2
. For height H2 , the points are G, H, I.
V
V
V
V
Here, both VR1
, VR2
< VRI and also VR1
> VR2
and
V
therefore the limiting speed will be VR2
. In other words,
for any wave height, the final reduced speed will be
V
V
the lowermost of the three speeds VRI , VR1
and VR2
.
Thus in general, we have:
Figure 12. Route between Chennai George Town based

on WAM climatological wave data for July.
Figure 13. Smoothedtrack between Cochin and Port Blair.
be considered by simply taking the additional drags

induced by these effects. Formulations are available
for estimating current drag and wind loads. Figure 14
shows a minimum-time route in a wave and current
31
Figure 18. Vizag-Port Blair route with voluntary speed

reduction, max roll 5 deg., pitch 2 deg. and heave 1 m.
Figure 16. Vizag-Port Blair route without voluntary speed

reduction.
CONCLUDING REMARKS
In this work, an algorithm has been developed for ship

weather routing application considering the prevailing weather (wind-wave-current) conditions and ship
behavior in waves. The specific application in this
work has been confined to routes lying in the North
Indian Ocean region, primarily because of the availability of information from ISRO satellites and from
sites such as NCMRWF of India providing ocean wave
related data over this region. The optimization procedure is based on a variant of Dijsktras algorithm,
which has been suitably modified for application to
the ship-routing problem.
The algorithm is general enough to consider a variety of optimization parameters since it is based on a
concept of a weight function associated with the path
that needs to be minimized/maximized. Therefore as
long as such a weight function can be determined combining weather conditions and ship performances, the
routing algorithm can in principle work. In this work,
the optimization criterion is chosen to be the time of
travel, so that the obtained path represents a minimumtime-travel route. For this, the procedure to determine
the appropriate weight functions has been explained.
It is found that the algorithm is versatile and robust
enough to handle all constraints that are usually present
in practical application of ship routing by ship operators. Specifically, the algorithm is found capable of
considering the following aspects:
Figure 17. Vizag-Port Blair route with voluntary speed

reduction, max roll 10deg., pitch 2deg. and heave 1 m.
where VR is the reduced speed, and VRiV , i = 1, 2, . . ., NR

are the maximum speeds based on NR number of
response criteria. Once VR for a given path line is
determined, the weights for the minimum-time route
is found as in eqn. (15).
Typical results with and without consideration of
voluntary speed reduction for ship 2 are shown in
Figures 1618. Figure 16 shows the route without voluntary speed reduction. Figures 17 is the route with
voluntary speed reduction based on a limiting set of
values of primary roll, heave and pitch motions (max.
roll 10 deg., max. heave 1m and max. pitch 3 deg.).
Figure 18 shows the plot with voluntary speed reduction, but now with a different set of value of max. roll
5 deg.. In these results it is found that voluntary speed
reduction has hardly influenced the route except over
a small part. This however is because of the moderate sea conditions: in these waves the ship is found to
hardly exceed the set criterion of the motions.
Consideration of presence of land masses in a direct

route between source and destination ports.
Consideration of non-navigable water and avoidance of zones not open for navigation due to existing
maritime rules and regulations. The ships Captain
also may at his discretion want to avoid certain areas
which he may perceive as risky, and this aspect
can also be considered by simply declaring for that
voyage those areas as non-permissible zones.
32
Cormen, T.H., Leiserson, C.E., Rivest R.L. & Stein C.

(2001), Introduction to Algorithms, The MIT Press,
Second Edition.
Faulkner, F.D. (1962), Determining optimum ship routes,
Jour. of Operations Research, Vol. 10, No. 6, pp. 799807.
Hagiwara, (1985), A study on the minimum fuel consumption
route-II: Simulation in the North Pacific Ocean, Journal
of Japan Institute of Navigation, Vol. 72, pp. 8796.
Hagiwara, H. (1989), Weather routing of (sail-assisted)
motor vessels, Ph.D. thesis. University of Delft,
Netherlands.
Haltiner, G.J., Hamilton, H.D. & Arnason, G. (1962), Minimal time ship routing, Jour. of Applied Meteorology,
Vol. 1, pp. 17.
Hanssen, G.L. & James, R.W. (1960), Optimum ship routing, Journal of Inst. of Navigation, Vol. 10, No. 3,
pp. 253272.
Mitchell, J. S. B. & Papadimitriou, C. H. (1986), The
weighted region problem, Tech. Rep., Dept. of O.R.,
Stanford University, Stanford, Calif.
Montes., A.A. (2005), Network Shortest Path Application
for Optimum Track Ship Routing, Naval Postgraduate
School Monterey CA, MS Thesis.
Padhy, C.P. (2010), Development of a ship routing algorithm
and its application to the North Indian ocean region,
Ph.D. Thesis, IIT Kharagpur.
Perakis, A. & Papadakis, N. (1989), Minimal time vessel
routing in a time-dependent environment, Transportation
Science, Vol. 23, pp. 266276.
Ulusoy, T. (2004), Stochastic Optimal Control of Seakeeping
and Routing of Ships, Department of Ocean Engineering,
MIT, Doctoral Thesis Proposal.
Zoppoli, R. (1972), Minimum-time routing as an N-stage
decision process, Journal of Applied Meteorology,
Vol. 11, pp. 429435.
Consideration of current field.

Voluntary speed reduction. The algorithm is versatile enough to consider a multiple criteria for
voluntary speed reduction.
Ship-routing has a large commercial application

and there are several weather-routing service providers
available internationally. However, most of these service providers, do not give much information on the
optimization method. In open literature also, information available on the basic method or algorithm
that can be used for practical application is scarce.
The commercial codes provide no information except
the claimed capability of their services, while most
research papers tend to deal with theoretical aspects of
optimization algorithms whose application for practical ship navigation is generally unclear.To this end, this
paper has presented the basis of a routing scheme that
is possible to be applied in practice by ship operators.
Although for the present development to be fully applicable as a decision making tool by ships Captain using
on board computers, further refinements and additional work will be needed, the paper has shown the
basic framework of a practical ship-routing algorithm.
REFERENCES
Chen H. (1978), A stochastic dynamic program for minimum cost ship routing, Ph.D thesis, Department of Ocean
Engineering, MIT, USA
Chen, H. & Lacey, P. (1998), Use of Operation Support
Information Technology to Increase Ship Safety and Efficiency, Trans. of the Soc. of Naval Architects and Marine
Engineers, Vol. 106, 105127.
33

Analysis of maritime traffic off the coast of Portugal

P. Silveira, A.P. Teixeira & C. Guedes Soares
ABSTRACT: This paper aims at studying the maritime traffic off the coast of Portugal. This study is based
on AIS (Automatic Identification Scheme) data, which is registered and maintained by the Portuguese coastal
VTS control centre (CCTMC), referring to the time period between the 9th July and the 9th August 2008.
The decoding of AIS messages is briefly described and computer programs are developed for the analysis and
visualization of the AIS information available for the present study. From the analysis of the AIS information,
the maritime traffic off the coast of Portugal is characterised and a statistical analysis of the ships in the Traffic
Separation Schemes (TSS) is provided. An algorithm is developed to assess the risk profile and the relative
importance of routes associated with ports.
INTRODUCTION
groundings and collisions, with an average value of

26, 21 and 5 events per year, respectively. It was
also verified that the largest number of accidents
resulted in ship sinking, followed by the number
of accidents where the ship was adrift and where
it had grounded. This study also showed that the
main causes of the registered accidents were mainly
related to material failure, sea and weather and human
error.
The need to ensure the safety of navigation has
led to the implementation of a coastal VTS (Vessel
Traffic Service) in Portugal. VTS systems improve
the safety of navigation, the safety of life at sea and
the protection of marine environment. The Portuguese
coastal VTS is equipped with several AIS (Automatic
Identification System) base stations that receive and
maintain records of messages transmitted by ship AIS
stations.
AIS data has become an important source of information for studying maritime traffic and associated
risks, in particular ship to ship collision risk, taking as examples the work of Montewka et al. (2010),
Ylitalo (2009), Goerlandt & Kujala (2010) and Mou
et al. (2010).
The present paper aims at performing a statistical analysis of the maritime traffic off the coast of
Portugal, using a sample of the AIS data collected by
the coastal VTS base stations between July 9th and
August 9th 2008. For this purpose a set of programs
have been developed and implemented to decode
the AIS messages and to visualize and characterize the
traffic patterns of the cost of Portugal. Based on the
decoded data, an algorithm has been also developed
to assess the risk profile and the relative importance
of routes associated with the ports of Leixes and
Lisbon.
The maritime areas under national jurisdiction of the

continental coast of Portugal are crossed by the busy
commercial routes connecting northern Europe and
the Mediterranean Sea. A significant part of the maritime traffic proceeding from the production centers
in the far East, namely China, passes through the Suez
channel, crosses the Mediterranean Sea to the strait
of Gibraltar and heads for the consumption centers
in northern Europe. Passing off the coast of Portugal,
these ships share the waters with vessels bound for
and leaving from national ports, such as fishing vessels, pleasure crafts, traffic from the Mediterranean to
North America, from Africa to northern Europe, from
northern Europe to South America, creating a complex
network of routes.
Maritime transportation involves risks such as fire
on board, running aground or collision, all of which
can result in loss of human life, loss of property and
spilling of dangerous or pollutant substances, with
potentially disastrous consequences for the environment and economic activities depending on the sea,
such as fisheries and tourism.The risks associated with
maritime transportation have been the object of various studies, as reviewed by Guedes Soares & Teixeira
(2001). The growth of the world fleet, both in number
of ships and in tonnage, tends to increase the number of accidents, and ship collisions are no exception
(Lloyds Register of Shipping 2010).
Maritime accidents on Portuguese waters have been
analyzed by Gouveia et al. (2007) with the objective of characterizing and assessing the main causes
of the maritime accidents. Based on the analysis of
857 maritime accidents it was identified a high frequency of some types of accidents such as founderings,
35
2 AUTOMATIC IDENTIFICATION
SYSTEM (AIS)
Table 1.
AIS equipments are designed to provide information

from a vessel to other vessels and coastal stations.
From December 31st 2004 all vessels over 300 GT
engaged in international voyages, all cargo vessels
over 500 GT and all passenger vessels are required
to be equipped with AIS (IMO, 2002).
2.1
Messages
Number
Percentage
Types 1, 2 and 3
Type 5
Type 18
Type 19
Total
122930797
6144137
286371
1466
129362771
95.028%
4.750%
0.221%
0.001%
Decoding AIS information
Table 2. Correspondence between matrix values

and pixel colours.
The AIS information used during this study was provided in raw format. Information contained in AIS
messages is encoded for transmission using appropriate protocols (NMEA AIVDM/AIVDO (Raymond
2009)), and had to be decoded in order to be usable. To
accomplish this, a program was developed in Python
programming language (Python Software Foundation,
2011) that decodes AIS messages of type 1, 2, 3,
5, 18, 19 and 24. Messages of type 1, 2 and 3 are
class A position reports, type 5 messages are ship
static and voyage related data from class A equipments, type 18 messages are standard class B position
reports, type 19 messages are extended class B position reports and type 24 messages, that were not found
in the sample provided, are class B static data reports.
The information decoded from the various message
types was:
the main ports were also discarded because it was not

the aim of this work to study the risk of ship collisions
inside the ports and to do so would imply changes
in the methodology used. The port areas from which
the messages were discarded do not match exactly any
legal port limits or jurisdiction areas.
2.3 Visualization of AIS information
Message type 1, 2 and 3: date/time; MMSI; navigation state; rate of turn; speed over ground; position
accuracy; latitude, longitude; course over ground;
heading;
Message type 5: date/time; MMSI, IMO number; ship type; length; breadth; position fix type;
draught;
Message type 18: date/time; MMSI; speed over
ground; position accuracy; latitude; longitude;
course over ground; heading;
Message type 19: date/time; MMSI; speed over
ground; position accuracy; latitude; longitude;
course over ground; heading; ship type; length;
breadth; position fix type.
In order to be able to visualize traffic patterns using the

decoded AIS messages, a set of programs was developed using Python programming language. These
programs create a matrix of which each element, initially given the value zero, is matched by a pixel of
a bitmap image. The side of each pixel is equivalent
to 0.001 of latitude or 0.06 nautical miles (111.12
meters). The matrix has 6001 lines and 2527 columns.
Line 0 matches latitude 36 N, line 6000 matches
latitude 42 , column 0 matches longitude 7 15W and
column 2526 matches longitude 10 30W.
Each element of the matrix is incremented when
the program reads a decoded AIS message sent from
a position corresponding to that element. Once the
program finishes reading all the decoded messages,
the color of the pixels of the bitmap image are changed
according to the value of the corresponding elements
of the matrix. Table 2 shows the correspondence
between values of the elements of the matrix and colors
of the pixels.
The resulting bitmap image is shown in Figure 1.
The decoded AIS messages are read from the file
containing class A position reports (message types 1,
2 and 3).
The decoded messages were stored in four files, one

containing type 1, 2 and 3 messages, another containing type 5 messages, another containing type 18
messages and another containing type 19 messages.
The distribution of decoded messages by message type
is shown in Table 1.
2.2
Decoded AIS messages by type.
Definition of the geographical area covered
AIS messages are transmitted using VHF radio waves.

The AIS base stations used by the Portuguese coastal
VTS are located in sites that ensure the coverage of
the VTS monitoring area. Having this in mind, the
area defined for the development of this study was
the area bounded by parallels 36.08 N and 41.86 N,
and by meridians 7.4W and 10.24W. All type 1, 2,
3, 18 and 19 messages sent from positions outside this
area were discarded. The messages sent from inside
MARITIME TRAFFIC REGULATION AND

CHARACTERIZATION
3.1 Ships routing systems

The main objective of ships routing systems is to contribute to safety of life at sea, protection of the marine
36
organizing traffic flow in areas, or around areas

where the passage of all ships or some specific types
of ships is dangerous or not desirable,
organizing traffic flow in, around or at a safe
distance from environmentally sensitive areas,
reducing the risk of grounding providing special
guidance in areas where the information about depth
is not reliable or where depth is critical,
guiding traffic away from fishing areas or organizing traffic flowing through fishing areas.
3.2 Traffic Separation Schemes (TSSs) and areas to
be avoided off the Portuguese coast
Traffic Separation Schemes (TSSs) are areas to be
avoided are both ships routeing systems. There are
two TSSs off the Portuguese coast, off Cape Roca
and off Cape S. Vicente. There is also one area to
be avoided, protecting the environmentally sensitive
Berlengas archipelago. TSSs are governed by rule 10
of the international regulations to avoid collisions at
sea (COLREGs). The Berlengas area to be avoided
shall not be crossed by vessels over 300 GT or vessels
carrying dangerous cargo, unless duly authorized by
competent authorities. Both TSSs have four separated
traffic lanes, four separation zones and one Inshore
Traffic Zone (ITZ). The four lanes in each mentioned
TSS are:
the northbound non dangerous cargo traffic lane,

the northbound dangerous cargo traffic lane,
the southbound non dangerous cargo traffic lane,
the southbound dangerous cargo traffic lane.
Figure 2 shows the TSS off Cape Roca and the Berlengas area to be avoided, Figure 3 shows the TSS off
Cape S. Vicente.
3.3
Characterization of maritime traffic off the

Portuguese coast
Figure 1 shows a high traffic density in the TSSs traffic

lanes and in the main routes connecting those traffic lanes. It also shows some concentration on the
approaches to the main ports and on some clearly visible routes, namely between the ports of Vigo, Leixes,
Aveiro and Figueira da Foz, and theTSS off Cape Roca,
between the ports of Lisboa, Setbal and Sines, and the
TSSs off Cape Roca and off Cape S. Vicente, between
the port of Lisboa and the archipelagos of Madeira and
Canarias, between the ports of Lisboa and Leixes,
and the Aores islands, between ports on the South
coast of Spain (like Cadiz and Huelva) and the TSS off
Cape S. Vicente, between some ports of Morocco and
the TSS off Cape S. Vicente, among others.
To analyze traffic in the area covered by this study
and specifically in the TSSs, a set of programs was
developed using Python programming language.
Table 3 shows the number of ships and average
speeds in the traffic lanes of the TSS off Cape Roca.
Table 4 shows some information about traffic in the
area covered by this study, collected from AIS data
Figure 1. Class A AIS traffic, 32 day period.
environment and safety of navigation in convergence

areas, areas with high traffic density, restricted waters
or areas affected by unfavorable weather conditions.
Routing systems aim at (Costa 2002):
separating opposite traffic flows, in order to reduce

the risk of head on collisions,
reducing the risk of collision between traffic proceeding along predetermined routes and traffic
crossing those routes,
simplifying traffic flow in areas of convergence,
37
Table 3. Number of ships and average speeds, off Cape

Roca TSS traffic lanes, 32 days time period.
Northbound lanes
Dangerous
Non.
dangerous
Average speed (knots)

13.13
13.75
Number of ships
292
1283
Table 4.
Southbound lanes
Dangerous
13.95
262
14.86
1237
General traffic information for the studied area.
Average number of ships:

Average speed of ships:
Average number of passenger ships:
Average speed of passenger ships
(v 1 knot):
Average length of passenger ships:
Average number of cargo ships:
Average speed of cargo ships (v 1 knot):
Average length of cargo ships:
Average number of tankers:
Average speed of tankers (v 1 knot):
Average length of tankers:
Average number of other ships:
Average speed of other ships (v 1 knot):
Average length of other ships:
Figure 2. TSS off Cape Roca and Berlengas area to be

avoided. Source: Portuguese Hydrographic Office (IH).
Non.
dangerous
139.92
12.81 knots
1.96
18.41 knots
233.17 meters
95.94
13 knots
146.83 meters
27.57
12.99 knots
161.73 meters
11.06
10.11 knots
88.1 meters
Figure 4. Ship distribution, TSS off Cape Roca, non dangerous cargo traffic lane, Northbound.
Figure 3. TSS off Cape S. Vicente. Source: Portuguese

Hydrographic Office (IH).
Figure 5. Ship distribution, TSS off Cape Roca, dangerous

cargo traffic lane, Northbound.
using the same set of programs. Average speeds were

calculated for ships with speed over 1 knot to discard
vessels at anchor or adrift.
The lateral distribution of traffic for each traffic
lane is shown in Figures 4, 5, 6 and 7, and the average
speeds of vessels using the traffic lanes is shown in
Figures 8, 9, 10 and 11. The lateral positions shown in

the pictures are normalized to each traffic lanes width.
Table 5 shows the mean value and standard deviation of the lateral position, and the percentage of
vessels passing out of the traffic lanes limits. Lateral
38
Figure 10. Average speeds, TSS off Cape Roca, non

dangerous cargo traffic lane, Southbound.
Figure 6. Ship distribution, TSS off Cape Roca, non

dangerous cargo traffic lane, Southbound.
Figure 11. Average speeds, TSS off Cape Roca, dangerous

cargo traffic lane, Southbound.
Figure 7. Ship distribution, TSS off Cape Roca, dangerous

cargo traffic lane, Southbound.
route between the port of Lisboa and the archipelago

of Madeira. The route was studied between the points:
Point 1: 38 30.86 N/009 30.96W

Point 2: 38 01.02 N/010 09.05W
The program registered that the route Lisboa

Madeira was used 26 times and that the route
Madeira Lisboa was used 12 times. It was also registered that this part of the route was crossed 1414
times from the NW to the SE, and 1521 times from the
SE to the NW. Figure 13 shows the number of crossings
and their position on the studied part of the route.
Figure 14 shows an overlapping of the route
between Point 1 and Point 2, and the chart in Figure 13,
on Google Earth.
Figure 8. Average speeds, TSS off Cape Roca, non dangerous cargo traffic lane, Northbound.
3.5
Characterization of routes associated

with ports
The routes used by ships approaching or leaving ports

can also be characterized using AIS information. A
program was developed to assess the relative importance of routes associated with ports. A reference point
is chosen for each port, and the bearing to that point of
AIS messages transmitted by ships entering or leaving
the port is recorded. The result for the ports of Leixes
and Lisboa is shown on Figures 15 and 16.
Observing Figure 16 it is possible to identify bearings with a greater number of messages, which should
correspond to the routes associated with the port of
Lisboa that are used more often. These major routes
related to a port can also be identified using Figure 1.
Figure 17 represents the positions from where AIS
messages were sent by ships inbound to and outbound
from Lisboa.
Table 6 represents the number of messages
received by ships using the major routes associated
with the port of Lisboa.
Figure 9. Average speeds, TSS off Cape Roca, dangerous

cargo traffic lane, Northbound.
position and speed were measured when crossing the

parallel 38 45 N.
The ships distribution according to their speed is
shown in Figure 12.
3.4 Characterization of a route
AIS information can also be used to characterize a
route, either by counting vessels using that route or by
analyzing traffic crossing that route.
A program was developed to study traffic using and
crossing a route, and it was applied to a part of the
39
Table 5. Ships out of traffic lanes limits, mean values and standard deviation of lateral position relative to traffic flow, off
Cape Roca TSS.
Ships out of limits
Non dangerous cargo Northbound

Dangerous cargo Northbound
Non dangerous cargo Northbound
Dangerous cargo Northbound
Lateral position
Left
Right
Mean
St. deviation
0.00%
0.00%
0.56%
0.38%
0.47%
0.34%
0.24%
0.00%
0.642
0.573
0.440
0.457
0.185
0.147
0.205
0.166
Figure 12. Ships distribution according to speed.

Figure 15. AIS messages received from vessels entering and
leaving the port of Leixes.
Figure 13. Crossings of route Lisboa-Madeira, from Point

1 to Point 2.
Figure 16. AIS messages received from vessels entering and

leaving the port of Lisboa.
CONCLUSIONS
The maintenance of AIS records provides a very useful

source of data for studies about maritime traffic.
The programs developed during this study provide
tools to decodeAIS messages, visualize traffic patterns
and to perform statistical analysis of traffic.
The data used during this study reports to the period
between July 9th and August 9th 2008. To improve
the quality of the analysis, the sample should be
taken from several periods during the year, so as to
identify and account for possible seasonal or cyclic
characteristics.
Figure 14. Crossings and route between Point 1 and Point 2,

Google Earth.
40
REFERENCES
Costa, A. 2002. Apontamentos da disciplina de Navegao e
Planeamento, ENIDH.
Goerlandt, F. & Kujala, P. 2010. Traffic simulation based ship
collision probability modeling, Reliability Engineering
and System Safety 96, 91107.
Gouveia, J., Anto, P. & Guedes Soares, C., 2007. Statistical analysis of ship accidents in the Portuguese
waters (in Portuguese), Riscos Pblicos e Industriais, C.
Guedes Soares, A.P. Teixeira e P. Anto (Eds), Edies
Salamandra, Lisboa, Vol. I, pp: 499516.
Guedes Soares, C. & Teixeira, A. 2001. Risk assessment
in maritime transportation, Reliability Engineering and
System Safety 74, 299309.
IMO 2002. Regulation 19 of SOLAS Chapter V Carriage
requirements for shipborne navigational systems and
equipment, International Maritime Organization, available at http://www.imo.org/
Ylitalo, J. 2009. Ship-Ship Collision Probability of the
Crossing Area between Helsinki and Tallinn, A student research project, Helsinki University of Technology,
Espoo, Finland.
Kujala, P., Hanninen, M., Arola, T. & Ylitalo, J. 2008.
Analysis of the marine traffic safety in the Gulf of
Finland, Reliability Engineering and System Safety, 94,
13491357.
Lloyds Register of Shippings 2010. World Fleet Statistics, available at: http://shipbuildinghistory.com/today/
statistics/wldfltgrowth.htm (October 12th 2010).
Montewka, J., Hinz, T., Kujala, P. & Matusiak, J. 2010.
Pro bability modelling of vessel collisions, Reliability
Engineering and System Safety 95, 573589.
Mou J., Tak C., Ligteringen H. 2010. Study on collision
avoidance in busy waterways by using AIS data, Ocean
Engineering 37, 483490.
Python Software Foundation, 2011. Python Programming
Language, http://www.python.org/, April 11, 2011.
Raymond, E. 2009, AIVDM/AIVDO protocol decoding, available at http://gpsd.berlios.de/AIVDM.html
(November 26th 2009).
Figure 17. Positions of AIS messages sent by ships inbound

to and outbound from Lisboa.
Table 6. AIS messages corresponding to the major routes

associated with the port of Lisbon.
Nr. of messages
Inbound Outbound Total
EST Roca + Azores
Madeira
Canary Islands
EST S. Vicente
Setbal + Sines
5650
1121
595
2040
1276
7908
1701
1230
1340
1283
% Total
13558 43.05
2822 8.96
1825 5.79
3380 10.73
2559 8.12
Total 76.66
41
Ship design

Experimental study of the performance of a rigid wing sail

J.H. Bettencourt
Presently at IFISC, UIB-CSIC, Palma de Mallorca, Spain
N. Fonseca & C. Guedes Soares

ABSTRACT: The performance of a rigid wing sail was studied by wind tunnel experiments. The lift and drag
forces were measured for a range of angles of attack and flap deflections in order to get information about the
sails performance envelope. The results obtained show that the rigid wing sail behaves like a traditional cloth
sail when sailing against the wind, but has limitations on its performance when sailing in favour of the wind.
These limitations are related to the rigid sails design as an airplane wing which does not account for operation
in the low lift to drag ratio regime, contrary to the design of traditional cloth sails.
1
INTRODUCTION
Traditional cloth sails are the most widely used sails in

the yachting world today. However, some outstanding
sailing vehicles use rigid sails instead of cloth sails.
These rigid sails are, in general terms, wings setup
vertically in relation to the flotation plane of the yacht.
Some noteworthy vehicles that have used rigid sails
in the past include the Stars & Stripes catamaran that
won the 1988 Americas Cup challenge against a soft
sail yacht, the Yellow Pages Endeavour that set the
500 meter sailing speed record of 46.52 knots in 1993
and the ICCT class catamarans that feature rigid wing
sails. In the present day, we can refer to the 2010 Americas Cup that was won by a trimaran equipped with a
wing sail.
Wing sails are associated with high performance
sailing vehicles; however there are not much published
scientific studies dedicated to the performance of
rigid wing sails.
The purpose of this work is to study the performance
of rigid wing using aerodynamic force data from wind
tunnel measurements of the lift and drag forces and to
study the operation of the rigid sail in upwind sailing.
2
Figure 1. Aerodynamic forces in a sailing yacht in upwind

sailing.
In equilibrium conditions this propulsion forces balances the hydrodynamic resistance produced by the
hull and appendages moving through the water. The
sails also produce another force, the heeling or lateral
force FL that acts in a perpendicular direction to the
vehicles track (hence perpendicular to the propulsion
force).
This lateral force is balanced by the hydrodynamic
lateral force of the hull and appendages and its heeling
moment is balanced by righting moment of the hull
and appendages.
Due to the motion of the vehicle, the incident wind
the wind seen by the sails is a vector composition of
the true wind vector and the vehicles velocity vector.
This apparent wind is then defined by its magnitude VA
and angle with respect to the vehicles centreline A ,
as in Figure 1, and its dependence on the true winds
magnitude and direction and on the vehicles speed is
given by:
UPWIND SAIL AERODYNAMICS
2.1 Aerodynamic forces

The dynamics of sailing vehicles is determined by
the balance of the hydrodynamic and aerodynamic
forces acting on them, depicted in Figure 1. When
sailing against the wind (upwind sailing), yachts move
through water due to the propulsion force FP develop
by the sails when the incident wind passes over them.
45
the same amount with respect to the vertical. In this

inclined plane, the expressions for the sail forces are:
where VT and T are the velocity and angle of the true
wind and V is the yachts velocity.
The aerodynamic forces produced by the sail when
exposed to the apparent wind are of the same nature as
those produced by a wing, i.e. the sail produces a lift
force L at a right angle to the direction of the incident
wind and a drag force D in the direction of the incident
wind. The relation between these aerodynamic forces
on the sail and the sail forces are given by:
where e = e is the sails effective angle of attack,

given in terms of the effective apparent wind angle
e = tan1 (cos tan A ). The effective apparent wind
angle is the apparent wind angle seen in a plane normal
to the mast, inclined by . By using e and assuming that the sail is insensitive to the flow along the
inclined mast, the aerodynamic force coefficients can
be determined in a wind tunnel without having to use
an inclined model (Jackson, 1996).
The trim variable represents the effects of the
change of sail shape by the action of the crew. Trimming of the sail by the crew acts mainly on the sail area,
the sails curvature and its torsion. The sail area and
curvature change by action of the crew are expressed
by two coefficients r and f that represent, respectively,
the relative reduction of the inclining moment of the
sail due to a reduction in the sail area and the relative
reduction of the lift force of the sail due to a reduction
of the sails curvature.
where is the angle between the yachts centreline and

its track and is usually small (5 ).
2.2 Aerodynamic force coefficients
It is customary to express aerodynamic forces in the
form of non dimensional coefficients. In this case, (3)
and (4) can be rewritten as:
2.3
Upwind sail efficiency
When a sailing yacht navigates upwind, its sails work

as wings.The efficiency of wing can be measured by its
lift to drag ratio, usually expressed as CL /CD . Then, an
efficient sail would have a high lift to drag ratio thereby
producing lift while keeping a low aerodynamic drag.
In principle, the same measure could be applied to
a sail, but it would not be completely correct since the
purpose of a sail is not to develop lift at a low drag
value, but to generate a propulsive force making the
sailing yacht reach the highest possible velocity.
For weak winds this requirement amounts to, given
A , to generate the largest possible propulsion force FP .
In strong winds however, there is a limit to the propulsive force a sail should generate given that the inclining
moment produced by the lateral force FL should not
exceed the maximum moment the yacht can uphold.
A common way to visualize the ratio between these
two aerodynamic forces is the sails polar curve (Figure 2). The polar curve of a sail shows the lift to drag
ratio of the sail for a range of angles of attack. In Figure 2, the lower point on the polar curve shows the
point where the sail generates more lift per drag while
the second point shows the point where the sail produces more propulsion force per lateral force. The two
points are distinct and generally, the larger the apparent wind angle, the large will be the difference between
these two points. For a fixed trim sail, the polar curve
is built by plotting (CD , CL ) for the range of angles
of attack. For a variable trim sail, the polar curve is a
performance envelope obtained from (CD , CL ) curves
for each trim setting.
where = A is the angle of attack of the sail. The

corresponding forces can be recovered from their nondimensional value by multiplying the coefficient with
1/2V2A A where is the air density and A is a reference
sail area.
In the form of (5) and (6), the lift and drag coefficients are a function of the sails angle of attack. This
is true for a given sail shape and reference apparent
wind vector, but for a given sail and apparent wind
conditions the more general form of the aerodynamic
force coefficients is given by Jackson (1996):
where q is the dynamic pressure and Rn the Reynolds

number.
This expression describes the general dependence
of the aerodynamic forces on the incident flow, the
yachts sailing attitude and sail shape. The first two
terms and also the third in (7) are related to the variation in velocity and direction of the incident wind
along the sail height. This variation arises due to the
variation of the true wind speed along the sail height,
as can be seen in (1) and (2), where the dependence
of VT with height is implicit. The angle is the heel
angle of the yacht and its effect on the aerodynamic
forces is usually taken into account, in wind tunnel
studies, by considering the flow in a plane inclined by
46
15 kn, Flay and Vulevitch (1995) indicate z0 = 0.2 mm

whereas for a rural open field area z0 is about 0.01
0.15 m (Brederode, 1997).
The apparent wind is the result of the combination
of the real wind and the yachts velocity V. Due to
the vertical variation of the real wind profile (10) the
apparent wind varies along the mast not only in absolute value but also in direction (see (1)). This change in
the apparent wind direction with the height along the
mast is called the apparent wind torsion. The amount
of variation of the speed and direction of the apparent
wind depends on the sailing regime. Upwind sailing
results in much smaller variations than downwind sailing. Thus, the modeling of the apparent wind is more
difficult to achieve when testing downwind sails. Also,
modeling the variation of direction of the incident flow
along the mast is more complicated than modeling the
speed variation.
Claughton (1994) refers that only the apparent wind
gradient is modeled in the reported wind tunnel tests.
Flay (1996) built a wind tunnel that is capable of modeling the apparent wind torsion by the use of guiding
vanes that change the direction of the incoming flow
along the height of the test section.
Turbulence is another aspect of the incident flow
that requires additional modeling. Since the crew of a
yacht is capable of reacting rapidly to low frequency
turbulence (wind gusts), the effect can be modeled as
a high-pass filter applied to the spectrum of the atmospheric turbulence (Flay & Jackson, 1992). This means
that in the wind tunnel, only the high frequency content
of the turbulence spectrum needs to be modeled.
The modeling of wind tunnel turbulence is usually
based on matching the turbulence intensity of the modeled flow. The turbulence intensity is defined as the
ratio of the standard deviation to the mean value of the
flow:
Figure 2. Polar curve of a sail showing the different aerodynamic efficiency measures.
Since the goal of the sail is to propel the yacht at the

highest possible speed in the desired course, the evaluation of sail performance must be completed taking
into consideration not only its aerodynamic properties
but also the yachts hydrodynamics in order to identify the most favorable hull-sail combination. This task
is usually accomplished through the so called Velocity Prediction Programs see Van Oossanen (1993)
and Milgram (1993) for a review of the underlying
methodology.
3
FLOW MODELLING IN WIND TUNNEL

TESTS OF SAILS
The wind tunnel is a widely used tool to study of sail

aerodynamics and to obtain sail performance data.
Some applications of the wind tunnel to sail analysis include: evaluating different sail rig alternatives
(Campbell & Dijkstra, 2000); design of sails and sailing rigs for competitive sailing yachts (Flay, 2002);
producing data for numerical model validation (Locke
et al, 1996).
The requisites related to the wind tunnel characteristics are not particularly demanding, when compared
for instance, with aeronautical applications where
much larger flow speed are required. However some
care must be taken when modeling the incident flow.
Indeed, flow similarity requires that the modeling
reproduces the characteristics of the velocity profile
and turbulence content of the incident flow. Sailing
yachts operate on the atmospheric boundary layer that
has mean velocity profile given by:
In equilibrium conditions the following approximation

is valid (Flay & Vulevitch, 1995): u = 2.5u ; replacing it in (10) with U = VT , the turbulence intensity
variation with height above the sea surface is:
For the apparent wind, Taylors hypothesis is used

(Flay & Vulevitch, 1995) and the turbulence intensity
of the apparent wind then differs from the true winds
due only to the difference in the mean values of true
and apparent winds.
4 WIND TUNNEL TESTS OF THE RIGID
WING SAIL
where u is the friction velocity, z the height above

sea surface and z0 the characteristic roughness of the
sea surface. This roughness is a function of the wind
speed because of the wave height dependence on
it and is lower than land values. For a wind speed of
4.1 Wind tunnel characteristics

The wind tunnel tests were carried out at the Laboratory of Structures and Constructions at the Department
47
Figure 4. Longitudinal view of the model placement in the

wind tunnel. The flow direction is from left to right.
Figure 3. Configuration of the WingSailCat rigid wing sails.
Table 1.
Rigid wing sail model dimensions.
Wing span b
Root chord cr
Tip chord ct
Flap span bf
Flap chord cf
0.785 m
0.397 m
0.255 m
80% b
20% chord
of Civil Engineering and Architecture of the Instituto

Superior Tcnico. The wind tunnel is a suction type
wind tunnel with a test section area of 1.5 1.5 m2
and 5 m of length. The flow is created by a fan located
downstream of the test section with nominal power of
132 kW rotating at a constant speed of 1450 rpm. The
flow speed is controlled by changing the pitch angle
of the fan blades trough a compressed air circuit.
4.2
Figure 5. Transversal view of the experimental setup.
The angle was read manually from the scale.The model

was fixed at the desired angle of attack by a locking
screw attached to a metal guide. The flap angle was
varied by a small guide installed on the model with
fixed angular positions.
Rigid wing sail model
The rigid wing sail model is a rigid wing equipped with

a partial span flap. This wing shape was chosen due
to the projected configuration of the sailing vehicle
WingSailCat that requires a straight trailing edge for
its forward element (see Figure 3).
The model dimensions are shown in Table 1. It is
a low aspect ratio (AR = 2.4) wing with a symmetric
laminar profile NACA64A014 with maximum thickness of 14% of chord located at 40% chord length from
the leading edge. The model is made of an internal
structure in balsa wood with a polyester foil cover.
4.3
4.4 Force measurement

The measurement of aerodynamic forces on the model
was accomplished through a balance capable of measuring forces in two orthogonal directions parallel and
perpendicular to the free stream velocity in the wind
tunnel section (lift and drag forces).
The force balance was designed for the wind tunnel
experiments and is based on the five degree of freedom
wind tunnel balance design of Ramaswamy (1996).
The working principle of the balance is the deformation under load of a cage section where extensometers connected into two Wheatstone bridge circuits
transform the deformation of the cage into voltage
signals.
The balance was installed beneath the wind tunnel
test section floor and was connected to the model by a
rigid support structure (see Figure 6). The voltage signals were acquired by an NI cDaq-9172 chassis in to a
laptop running LabView 8.0 data acquisition software.
The balance was calibrated before the installation
in the wind tunnel. The purpose of the calibration was
to quantify the relationship between the read voltage
signal and the applied loads and also the interaction
between the two Wheatstone bridges. The interaction
results shows an average bridge interation of about
Experimental platform
The experimental setup is shown in Figures 45. The

rigid wing sail model was placed in the center of the
test section 2980 mm from the beginning of the wind
tunnels work section. The measurement of the dynamical pressure of the flow was made by a Pitot tube
placed upstream from the model at 1700 mm from the
beginning of the work section.
The dynamic pressure of the wind tunnel freestream
flow was measured by an differential manometer
Micatrone MG-1000 FLD connected to a daq module NI-9203. The measured dynamical pressure was
corrected by a calibration expression.
The angle of attack was measured with a scale
installed in the support base of the wind tunnel model.
48
Figure 7. Lift coefficient.
was calculated from the longitudinal force corrected

for the longitudinal force of the models support base.
A uncertainty analysis of the test results was carried out by the ISO uncertainty analysis method. The
analysis resulted in a mean uncertainty of 19% of
the force coefficients mostly due to the uncertainty
in the measurement of the dynamical pressure (the
visual indication of the manometer measure had a 1 Pa
precision).
Figure 6. Wind tunnel force balance. The cage section is

composed of four bars in the middle of the balance.
10%, meaning the 10% of the applied load in one

direction will show up as a voltage signal in the other
direction. From this calibration an interaction matrix
was derived and the voltage signals during the tests
were transformed into loads by this interaction matrix.
RESULTS AND DISCUSSION
5.1 Lift and drag coefficients

The lift and drag coefficients of the rigid wing sail
model are shown in Figures 78. The maximum lift
coefficient varied between 0.7 with the flap at 0 to
0.95 with the flap at 30 .
The lift coefficient curve slope is 0.0534 between
0 and 10 of angle of attack. The maximum lift coefficient is reached, for all flap deflections, between 14
and 16 of angle of attack. The linearity of the lift coefficient with angle of attack loses its linear shape with
increasing flap deflection due to the shift to negative
values of (not tested) of the zero lift coefficient angle
of attack. However, only for a flap deflection of 20 do
we see a shift in the angle of attack of the maximum
lift coefficient.
The drag coefficient of the rigid wing sail model
shows a smooth increase for all flap deflections up
to 10 12 of angle of attack. After this angle of
attack significant flow separation is likely to occur
from the increase in the drag coefficient seen around
14 16 of angle of attack. This range is coincident
with the maximum lift coefficient range, showing that
when the maximum lift coefficient is reached the flow
separation is already notable.
The polar curves of the rigid wing sail model are
shown in Figure 9. For the low CL range (up to 0.6) the
use of the flap is not advantageous because it will incur
in a somewhat higher drag for the same lift. But for
4.5 Test settings

The tests were conducted for ten angles of attack ranging from 0 to 20 and four flap angles of 0 , 10 ,
20 and 30 . There were considerable difficulties in
using the same dynamic pressure for each wind tunnel
test so the Reynolds number of the tests varied from
64 104 to 1 105 during the test period. For each
angle of attack and flap angle, two tests were made
and the loads were averaged between these two tests.
Force and dynamic pressure data where acquired for
10 s after the flow in the wind tunnel had stabilized at
a sampling frequency of 10 kHz.
4.6 Test data reduction
The aerodynamic loads where reduced to coefficient
form by the usual expression:
where q = 1/2V 2 A is the test dynamic pressure and

A = 0.5(cr + cp )b is the planform area of the rigid wing
sail model. The lift coefficient CL was calculated from
the transversal force FY and the drag coefficient CD
49
Figure 8. Drag coefficient.
Figure 10. Wind tunnel and yacht axis systems.
Figure 9. Polar curves.

Figure 11. Propulsion force coefficient vs. apparent wind
angle.
high CL, the flap can give better performance because

it reaches CL at a lower angle of attack.
velocity vector (Vx ) and the lift (Lm ) is perpendicular

to it. The wind tunnel angle of attack is m . In the yacht
axis system, the drag (D) is parallel to the apparent
wind velocity VA and the lift (L) is perpendicular to it.
The angle of attack of the sail is . Note that = A .
The variation of the propulsion force coefficient
CFP and lateral force coefficient CFL with the apparent wind angle are shown in Figures 1112. The
sails angle of attack is kept constant by changing the
sheeting angle .
The propulsion force reaches its maximum value
around A = 100 while at this apparent wind angle
the lateral force is close to zero. This holds for all
flap deflection angles tested. This is due to the fact
that at this apparent wind angle the longitudinal components of the lift and drag forces of the sail have
the same direction and orientation and the transversal components have the same direction but opposite
orientations, canceling each other out (see Figure 13).
5.2 Analysis of the rigid wing sail performance

The analysis of the lift and drag coefficients can give an
insight in to the aerodynamic performance of the rigid
wing sail, but a correct performance analysis needs to
be made on the basis of propulsive and lateral forces
as a function of the apparent wind.
In order to do this, the results from the wind tunnel
need to be transformed into the aerodynamic forces
produced by the sailing yacht.
The transformation is based on (3)(4) or (5)(6)
but with set to zero since we do not know its value
(although it should remain small). Figure 10 shows
the relationship between the forces in the wind tunnel reference axis system and the forces in the sailing
yacht reference system.
The longitudinal wind tunnel axis (x) coincides with
the longitudinal axis of the yacht (X). The wind tunnel
drag force (Dm ) is parallel to the wind tunnel flow
50
Figure 12. Lateral force coefficient vs apparent wind angle.
Figure 14. Aerodynamic force in the yacht axis system.

Flap deflection 0 .
Figure 13. Components of lift and drag forces in yacht axis

system.
Figure 15. Same as Figure 14. Flap deflection 30 .
At A = 100 the advantages of the flap are easy to

see since we have an increase of the propulsion force
but only a marginal increase of the lateral force.
As the apparent wind angle reduces the gains in
using the flap are lower mainly because the lateral force
increases with increasing flap deflection.
Increasing A beyond 100 causes a negative lateral
force which is not desirable. For a fixed angle of attack,
increasing the apparent wing angle causes a rotation
of the force system that makes the sum of the aerodynamic forces components transversal to the yacht
negative.
To increase the range of use of the rigid wing sail
it is necessary to delay this event. For an apparent
wind angle of 115 and zero flap deflection, Figure
14 shows the orientation and magnitude of the total
aerodynamic force (the vector sum of L and D). The
lateral force remains negative up to an angle of attack
of 19 , meaning that the sail would be operating in a
stalled condition.
However, by increasing the flap deflection to 30
(Figure 15), the range of angles of attack with positive
lateral force increases to 16 20 . This increase is due
to the increase in the drag force of the sail in this range

of , with a nearly constant lift, what can be seen in the
polar curve of the sail for 30 degrees flap deflection
(Figure 9).
These results show that there is a limit to the operation of this rigid sail related to the appearance of a
negative lateral force. This limit depends on the aerodynamics of the high angle of attack range of the rigid
wing sail.
CONCLUSIONS
The wind tunnel tests conducted in this study allowed

the sailing performance evaluation of a rigid wing
sail. The aerodynamic lift and drag forces and also
the apparent wind vector play a key role in the development of the propulsion and lateral forces. The drag
force, D, plays a beneficial role for larger apparent
wind angles, A , since it holds the total aerodynamic
force, FA , on the positive side. But for lower apparent wind angles, the beneficial role of aerodynamic
drag is reversed and it becomes prejudicial. The role
51
Flay R. G. J. 1992. Wind Tunnel Tests on a 1/16th Scale

Laser Model, Ship Science Report No. 55, University of
Southampton.
Flay R. G. J. 1996. A twisted Flow Wind Tunnel for Testing
Yacht Sails, Journal of Wind Engineering and Industrial
Aerodynamics 63: 171182.
Flay, R. G. J. & Jackson, P. S. 1992. Flow Simulation
for Wind-tunnel Studies of Sail Aerodynamics, Journal
of Wind Engineering and Industrial Aerodynamics 44:
27032714.
Flay R. G. J. & Vulevitch I. J. 1995. Development of a
Wind Tunnel Test Facility for Yacht Aerodynamic Studies,
Journal ofWind Engineering and IndustrialAerodynamics
58: 231258.
Locke N.J., Jackson P. S. & Flay R. G. J. 1996. Lift and Drag
Distributions of Yacht Sails using Wake Surveys, Journal
of Fluids Engineering 118: 346351.
Milgram J. L. 1993. Naval Architecture Technology used
in winning the 1992 Americas Cup match, SNAME
Transactions 101: 399436.
Van Oossannen, P. 1993. Predicting the Speed of Sailing
Yachts, SNAME Transactions 101: 337397.
of the lift force, L, is opposite to that of D, contributing to the propulsion force at low A and to a negative
lateral force for A larger than 100 .
The maximum propulsive force is obtained for an
apparent wind angle around 100 and in this case
the use of a deflected flap increases significantly the
propulsive force. The advantage of using the flap
reduces as A reduces because, although the lateral
force increases, the projection of this force along the
propulsive direction tends reduce and even become
negative below a specific value of A .
REFERENCES
Brederode V. 1997. Fundamentos de Aerodinmica Incompressvel, Edition by the author.
Campbell I. M. C. & Dijkstra G. 2000. Wind Tunnel Tests
on Schooner Rigs and their use in Performance Prediction by VPP Calculations, 16th HISWA Symposium,
pp. 91105, 2000.
Claughton A. R., Campbell I. M. C., Wind Tunnel Testing of Sailing Yacht Rigs, 13th HISWA Symposium,
pp. 89106.
52

Considering disabled people in sailing yacht design

F. Cerveira, N. Fonseca & L. Sutherland
ABSTRACT: The paper discusses the design requirements for development of sailing yachts adapted to disabled
people. A significant percentage of the population has some kind of limitation, either in terms of mobility,
sensorial, or intellectual capabilities. In general this prevents the practice of sailing because the boats and yachts
are not adapted to deal with these limitations. This paper starts by identifying and categorizing the disabilities
and the consequences on the sailing activity. Then a list of design requirements is established and a procedure is
recommended to include these requirements into the design of a sailing yacht. Finally, the development of the
layout design for a 10 m sailing yacht for disabled sailors is presented as an illustrative example.
INTRODUCTION
disabled people indicates that they sail mostly single

handed and does not have the opportunity to be part of
a crew. Also, it is not easy for a disabled sailor to leave
sheltered waters for even small coastal passages.
The present paper intends to identify the most
important limitations to the integration of disabled
people onboard sailing yachts and organise them for
easier incorporation in the yacht design process. The
safety onboard is also considered here because disabled people often have more difficulties to cope with
emergency situations and face a higher risk of injury.
This is achieved via a review of the literature published
about injuries in conventional cruising yachts, which
identifies the most critical areas onboard, the potentially dangerous equipment and the frequent types of
injuries. A few comments are also given for injuries
particular to disabled people. The presentation of the
needs of disabled people is concluded with a review
of the vessels designed for disabled people and the
respective auxiliary systems, which show how other
projects have tried to solve these problems. To illustrate the discussion, finally a case study of a 10 meter
yacht designed for disabled people is presented.
Sailing is a popular activity worldwide, some people

practice it as a sport others for pleasure and disabled
people may often use it for their personal development. Unfortunately, conventional yachts are usually
designed for the middle age male population, and are
mostly of the times not even suitable for women, young
and old people. This is the starting point of the present
paper, based on the concept that all products and environments should be designed to consider the needs
of the widest possible array of users. This is called
Universal Design, which is a way of thinking about
the design that considers human limitations not as a
condition of few but as a common characteristic of
all, since we all change physically and intellectually
throughout our lives. If a design works well for people
with severe limitations it is expected to work better
for everyone. Therefore, the present work is focused
on disabilities, which are considered to be the most
severe type of human limitations, but the final result
is expected to benefit a much wider population.
In comparison with other sports, sailing vessels
require little mobility and the controls may be easily adapted to different kinds of operation, and these
characteristics make sailing a popular activity for the
disabled population. The majority of boats used by
disabled people are small in size and crew, often
dinghies commercially produced for this application.
These boats solve the mobility problem by seating the
sailor in a central position facing the front with all
the sails and directional controls within arms reach.
The controls are often modular and offer different possible operation according to the users needs. In terms
of larger vessels there are only a few individual projects
of yachts designed for disabled people and a few other
conventional yachts extensively modified for the this
purpose. The study of the sailing vessels available for
2
2.1
NEEDS OF DISABLED PEOPLE

Limitations of disabled people
A disability is often associated with several limitations that may change according to the demands of
the activity performed, for example a paraplegic sailor
may not have a mobility limitation if all the controls are within arm reach. Therefore it is important
to identify which are the limitations that common disabilities have regarding the operation of conventional
sailing yachts. However, there is little published information on disabled people sailing, the most relevant
53
Table 1.
Most common limitations to the integration of disabled people in sailing yachts.

Permanent adaptations
Disability
Physical
Paraplegia e
Quadriplegia
Amputations
Intellectual
Intellectual
disability
Acquired Brain
Injury
Cerebral Palsy
Sensory
Blindness or Visual
Impairment
Hearing Impairment
Secondary
Diabetes
Asthma
Non-permanent adaptations
Personal adaptations
Simple
Low
Direction/
Low
Large &
heel Balance Mobility High Aid to go Sail
Electronic Physical
Body
Colourful
angle aids
aids
boom onboard Controls equipment Performance Protected Controls
x
x
x
x
x
x
x
is initially difficult to control even by non-disabled

people. This system should be substituted by a driving
wheel or a vertical tiller operating in a more intuitive
direct manner.
Electric equipment is most helpful for people with
poor motor control, which is a limitation characteristic of some physical and intellectual disabilities.
Visual impaired people may also benefit from electric equipment to receive information. It is worthwhile
remembering that most disabled people have poor
physical performance due to the disability itself or a
lack of physical activity, requiring light operational
systems.
Paraplegic and quadriplegic people have no feeling in large areas of the body and are susceptible to
injuries of which they do not notice. Some intellectual
disabilities have poor motor control and thus may also
suffer from injuries due to collision with objects and
surfaces. These people would benefit from cushioning and other body protections. Large and colourful
controls would help people to hold and/or better see
equipment.
references on the subject are published online by

the Sailability organisation (www.sailability.org) and
the International Foundation for Disabled Sailing
(www.sailing.org).
The main limitations to the practice of sailing are
presented in Table 1 according to the respective disabilities. This classification is based on the previous
references and the first authors experience with disabled sailors.The disabilities are divided in physical,
intellectual, sensory and secondary, where the
secondary group require some considerations but does
not prevent the practice of sailing. These limitations
to the pursuit of sailing are categorised as permanent, non-permanent and personal adaptations,
depending on how they affect the yacht.
Table 1 aims to differentiate between disabilities
and the associated limitations, particularly because
intellectual disabilities often suffer from associated
physical limitations. For example, people with cerebral palsy often have poor motor coordination. Most
physical and intellectual disabilities give problems in
coping with the heel angle, often requiring balance
aid equipment onboard. Also, people with physical
limitations often need mobility aids, in particular for
boarding and disembarking from the boat. People with
intellectual and sensory disabilities, have problems to
predict the movement of the boom which may lead to
serious injuries.
The directional and sail controls should be intuitive
enough to be understood by people with intellectual
limitations. For example, the conventional horizontal
rudder tiller changes the boat direction in the opposite
direction to that in which the tiller is moved, which
2.2
Methodology to consider disabled people

in yacht design
The identification of the most common limitations to

the practice of sailing is made to simplify the designers
work, since then (s)he is not concerned with a group of
disabilities but with a group of design requirements.
However, a logical procedure must be defined to organise these limitations according to different stages of the
design process. The approach suggested here is to give
54
controls should be considered. All the considerations made in terms of the control systems are nonpermanent and may be changed according to the crew
needs. Modular solutions are recommended to adjust
to the individual requests.
Simple controls are required when the user cannot
understand intuitively the system operation or when
there is no physical ability to cope with it. Electric
devices are often employed because they reduce the
physical requests and are able to be operated by a
variety of interfaces. However, there are people who
benefit from several adaptations but still wish to sail
as a sport and search for intense physical activity, and
again a solution may be found with modularity, providing for example the possibility to operate the systems
manually or with the aid of an electric system.
an order of importance to the limitations, where the

most important limitations are those which affect permanently the yacht, followed by the limitations that
require non-permanent adaptations, and finally any
remaining details that should be considered.
Permanent adaptations are considered to be, for
example, a heavy keel to reduce the heel angle or a
general arrangement designed for wheelchair operation. Non-permanent adaptations are related with the
operational controls and other systems that may be
easily changed through the vessel life. The subsequent
paragraphs divide the integration of disabled people
in a sailing yacht according to three design stages and
identify which are the limitations to consider at each
stage.
2.2.1 First design stage permanent adaptations
The first design stage focuses on the adaptations identified in Table 1 which affect permanently the vessel.
The heel angle is an uncomfortable characteristic of
sailing yachts, because it increases the problems with
balance and mobility onboard. Therefore, a major
design consideration should be a low heel angle, which
affects the hull shape and appendages configuration.
When sailing, most people have balance problems,
but the disabled are the most affected generally due
to amputations, low trunk stability or poor motor
coordination. The hull and appendages should be
designed to reduce the three dimensional accelerations, and the general arrangement should include
balance equipment, such as chairs, hand rails, etc.
The mobility onboard a sailing yacht is a difficult
problem to address because in most cases there is not
enough space for conventional mobility aids. This limitation is increased by the heel angle and the three
dimensional accelerations experienced by the crew.
The easiest way to solve the problem is to place the
sailor in a central position facing front with all controls
within arm reach, which minimizes the need to move.
This was the solution implemented for small dinghies
adapted to disabled sailors. However, in a larger yacht
the crew must have access to different areas and the
previous solution is not suitable. Therefore, the mobility solution must be according to the size of the vessel,
the general arrangement and expected heel angle.
The mainsail boom swings from one side to the
other above the cockpit and deck and must be well
above head height, since most intellectual disabled
or visual impaired people may not predict its sudden
movement, for example due to an uncontrolled gybe.
Attention should be also given to the main sheet and
blocks which should be attached to a position from
where they cannot cross the cockpit. A collision of
a crew member with one of these equipments may
certainly cause a severe injury.
2.2.3 Third stage personal adaptations

After the control systems are defined a wide range of
small details should be considered. The third design
stage continues after the design project and build of
the vessel in order to account with the needs of each
particular user. This design stage is concerned with
personal equipment, for example cushions to protect the users body, Braille signs, personal grips for
equipment holding, etc.
The design sequence suggested here constitutes a
logical procedure to address a large group of problems identified as frequently affecting the integration
of disabled people. The subsequent section will present
various considerations to avoid the risk of injuries in
the general sailing population, which is an alternative way to improve any sailors integration, dis- or
able-bodied.
PATTERNS OF INJURY IN SAILING

YACHTS
Little literature exists on the causes of injuries sustained whilst sailing and most of the studies are
concerned with high performance athletes, which suffer from specific injuries mostly related with overuse
of particular areas of the body. For more representative,
normal sailing a relevant study was made of cruising sailors who spend most of the year on offshore
cruising yachts (Rouvillian, et al., 2007). Figure 1
reports the most frequent places where trauma occur
onboard cruising yachts. The deck is the most dangerous area due to the high exposure to the environment
and equipment without effective means of protection.
The cockpit is more protected than the deck but most
of the work is performed here with the associated risk
of injury. The companionway is also a critical area
because people are moving from one level to another
where it is easy to lose balance or slip and fall. Many
reported injuries also occur at the companionway by
colliding with moving equipment when coming out of
the interior. The kitchen is also of concern, in particular
when using a gimballed stove.
2.2.2 Second design stage non-permanent

adaptations
After defining the general arrangement, hull shape
and appendage configuration, the directional and sail
55
contact with the hot stove. The kitchen area should be

well designed with handrails available and the cooker
should also wear protective equipment while cooking
at sea.
The occurrence of lower back pain in offshore cruisers is most related to the raising of the anchor. All of
the injuries occured in small boats without a windlass
and all traumas were sustained when the subject was
pulling while standing upright, with the full weight of
the chain being held by the lumbar spine. If a windlass
is not available, it is suggested that the sailor should
change to a sitting position on deck with feet resting
safely on the pulpit. The lower back pain can be also
associated with poor positioning in repetitive motions
like trimming sails or lifting objects like spinnaker
poles or heavy sails (Allen, et al., 2006).
Preliminary studies have been performed on
injuries related with disabled sailors, but these are limited, and suggest that these people suffer from injuries
similar to those suffered by able-bodied people (Allen,
et al., 2006). In 1999 a survey of disabled sailors during the International Foundation for Disabled Sailing
World Championship, with 24 teams and multiple disabilities types, showed that the majority of the injuries
were chronic in nature (68%) with sprains and strains
being the most frequent types of injury.The crew members were at greatest risk of injury in the three person
class (96%), with equal distribution between the foredeck and mid-deck positions. The upper limbs were the
most frequently injured body region (60%), likely due
to an increased reliance on the upper limbs as a consequence of lower limb or spinal cord disability (Neville,
et al., 2009). When performing actions onboard it is
important that disabled people without trunk stability have means to support themselves, otherwise one
hand will be working while the other is holding the
position, increasing significantly the physical efforts
of the upper extremities.
Figure 1. Risk of injury in cruising yachts [%] (Rouvillian,

et al., 2007).
The anatomical regions involved in trauma have

been classified by increasing order of frequency as
pelvis/abdomen (2%), head/neck (13%), chest (14%),
upper limbs (35%) and lower limbs (35%), and these
results are similar to those reported from an amateur
around the world race, The British Telecom Round the
World Yacht Race 19961997 (Prince, et al., 2002).
Rouvillian et al. (2007) found that head and neck
trauma are often related with a moving spar such as the
boom. Accidents typically occur when there is a combination of helm error and gybing while the victim is
returning from the cabin, and in no case was a companionway sea hood fitted. This is a removable structure
placed in front of the cockpit to protect the crew from
spray, wind and sun. It appears that this hood forces the
sailor to bend down when descending the steps down
into to the cabin which movement protects the sailor
from a collision with the boom.
It was found that chest traumas normally occur
when a crew member is carrying an object up to the
cockpit. If a rolling motion occurs, the body rotates
about a single vertical axis, and the person is unable to
grab for support because the hands are occupied. The
companionway should be protected by two side walls,
which is much safer than a simple narrow ladder since
the person can lean against a wall to free a hand while
keeping three support points.
Hand injuries are generally caused by deck equipment with moving parts and it is recommended
that equipment such as pulleys, winches, windlasses,
propellers and motorised deck covers has suitable
protection. The lower limb trauma is as common as
the upper limb trauma and occurs mostly due to collision with deck hardware. This can be improved with
pathways clear from obstructions, with smooth and
not slippery surfaces, and by wearing protective foot
equipment.
The burns onboard due to solar exposure made up
28% of the injuries reported by Rouvilan et al. (2007),
although most of the crews had been very careful
to use sun protection methods. It was in the kitchen
that the most severe burns occur. The stove onboard
is gimballed, but this means that it also swings with
the motion of the boat, possibly resulting in dishes
falling off the stove or the cooks body coming into
EXISTING VESSELS FOR DISABLED

PEOPLE
The majority of boats used by disabled people are small

in terms of size and number of crew. Most of them
solve the mobility problem by seating the sailor in
a central position facing the front with all the sails
and directional controls within arm reach. The control systems of the sails and rudder are simple and
often compatible with electric devices. The most relevant existing projects for disabled sailors will now be
described.
The Access Company is the largest builder of sailing vessels for disabled people, these vessels search for
a balance between recreation, competition and therapeutic activities at a low price. Designed for one or
two sailors, the Access dinghies have light and simple
operative systems. The stability is provided by a heavy
centreboard and wide beam. The high freeboard and
the wide deck intend to keep the boat dry when heeling,
which is important for sailors that easily suffer from
56
Figure 2. Veritas K with

(www.disabledsailing.org).
lateral
entrance
open
hypothermia. The disadvantage of the high freeboard

and wide deck is a possible loss of view to windward
when heeling.
Nowadays more than 1200 Access dinghies are
sailed in fifteen countries and its popularity is increasing considerably. There are a few more dinghies
designed for disabled people, namely the Challenger,
2.4 M, Artemis 20 and Martin 16, and all of them also
solve the mobility and balance problems by seating the
sailor in a fixed central position.
In terms of larger vessels there are only a few individual projects. A recent project at the University of
S. Paulo designed a day-sailer yacht for disabled people (Simos, 2007). The POLI19 has a simple deck
layout to leave open the possibility to insert personal
adaptations. This creates a flexible base for future personalisation, while a rigid layout would restrict the
vessel to one particular type of limitation. The boom
is relatively high when compared with similar boats to
avoid head injuries. The POLI 19 has a large cockpit
of approximately 55% of the length overall to avoid
the need to leave the cockpit to access forward deck.
It has two seats at different heights with an inclination
angle relative to the horizontal to support people with
balance problems while sailing with heeling angle. The
stern is open and has a small platform for easy access
to the water to recover a man-over-board.
TheVeritas K is another interesting project, this time
of an ocean going vessel that aims to be wheelchair
accessible (www.disabledsailing.org). It has a centre
cockpit which protects the crew from the sea spray
and gives a clear view all around. All the systems are
controlled from the cockpit, with space for wheelchair
operation. The entry to the cabin is via a stairway or
a wheelchair lift platform which, when not in use,
lies flush on the cabin sole and does not hinder safe
access to the cabin by able-bodied crew members.
The saloon also has space for wheelchair operation
and access to all the equipment. Figure 2 shows the
interface between the vessel and shore via the starboard side, where a gangplank concealed in the hull
gives easy access to the cockpit for anybody, including
wheelchairs users.
The catamaran configuration intrinsically helps the
integration of disabled people because of the low heel
angle and the spacious cockpit, suitable for wheelchair
operation. A good example is Impossible Dream, an
18 meter ocean going wheelchair accessible catamaran (www.impossibledream.org.uk). It has a flat, wide
Figure 3. Six way Power helm seat (www.martin16.com).
deck to allow wheelchair users to move all around the

boat and there is extensive use of rails on deck and
inside the cabin. The Impossible Dream has three navigation stations, one inside the cabin and one at each
side of the deck. From here it is possible to steer the
boat, control the sails and engines, and operate all the
electric devices such as autopilot, GPS, Plotter, etc. For
comfort while skippering the boat, there are two chairs
inside the cabin fixed to a rail system. The interior
is also wheelchair accessible, with access to kitchen
equipment and bathroom, and a lift platform provides
the access to the compartments below decks.
4.1
Modern adaptive systems
A wide range of equipment adaptations have already

been developed to cope with disabled peoples limitations, in particular for the Paralympic classes
where numerous athletes improve their performance
by adapting the material to their needs. Some of
these interesting adaptations are provided on the International Foundation for Disabled Sailing web site
(www.sailing.org). Most recent developments have
been in electrical aids and equipment, and these
aspects and relevant examples are discussed below.
The design team of the Martin 16 dinghy developed an interesting seat for the helm position, which
allows people with mobility and balance problems to
steer the vessel comfortably for long periods of time.
This system, presented in Figure 3, is designed to provide ergonomic and secure seating for people with
or without disabilities (www.martin16.com). The seat
travels from one tack to the other, rotating through 180
degrees, allowing the sailor to adopt the conventional
position facing the sails or the front of the boat. To
accommodate the heel of the boat the seat also tilts 25
degrees forward and aft and can slide up to 8 inches
towards the driving wheel, adapting to specific personal limitations and comforts. A four point safety
harness secures and maintains the balance of the sailor
in a seaway.
The same group also developed a complete set
of electric adaptations to operate small vessels
57
Table 2.
Inclusion 32 main dimensions.
Displacement
Ballast
LOA
LWL
BMAX
Draft
Sail area
the interior demands and an efficient hull shape. Next

the appendage configuration was defined and finally
a performance prediction (VPP) developed. The solutions applied to the Inclusion 32 for the integration
of disabled people are now discussed according to the
limitations identified in Table 1.
Figure 4. Sip and puff interface (www.martin16.com).
(www.martin16.com). This system provides the interface between the major controls of the boat and people
with severe limitations. The main part of the system
is a joystick module, which contains the main computer and a joystick control. The left and right axis of
the joystick operates the helm drive motor to steer the
boat, while the forward and backward axis of the joystick operates a windlass to trim the sails. The joystick
module can control up to three motors and operate
several trimming lines. Other way to control this system is with the sip and puff module presented in
Figure 4, which provides a sensitive pneumatic control interface, to allow highly quadriplegic sailors to
control the system functions with their breath. The sip
and puff interface is a chest mounted control stalk
with two pneumatic straws within reach of the sailors
lips, one to control the helm and one to control the
windlass.
The development of such electronic equipment
has allowed important improvements to be made in
the adaptability of sailing vessels. It is important to
remember that a quadriplegic person, for example,
is limited in most daily life activities, but that with
an interface like the sip and puff system he or she may
go on the water and enjoy an active days sailing.
6400 kg
3400 kg
9.8 m
9.1 m
3.6 m
1.8 m
42.7 m2
5.1 Mobility
The mobility onboard was simplified by a single sole
level, common to the interior and cockpit and similar to a sports yacht. It was decided to use a single
sole level to avoid the use of the unsafe ladders or lifting platforms, which are not suitable for the wider
disabled population. In addition, it was decided to
accommodate for wheelchairs onboard since most disabled people are familiar with their operation and
the interface with shore is thus easier. However, to
enable the use of wheelchairs onboard it is necessary
to have a general arrangement with spaces of larger
than usual dimensions, defined according to the guidelines of Universal Design architecture (Goldsmith,
2000). Figure 6 presents the access routes available
for wheelchair users.
5.2 Heel angle
The second design requirement for the integration of
disabled people is low heel angle. For the present
project the maximum heel angle for normal sailing
conditions (15 knots true windspeed) was 10 which
was assumed to be the safe limit for wheelchair operation; for wind speeds above 15 knots the crew should
start reefing the sails. This was guaranteed by the hull
shape and the design of the appendages.
The hull shape was designed with a large waterline beam to increase the form stability, which is the
major stability component at low heel angles. In terms
of the appendages it was necessary to find a balance
between the two options of a lower than normal sail
area and a heavier than normal keel. The Inclusion 32
was designed with a conventional sail planform with
a sail area and aspect ratio slightly lower than those of
conventional yachts of this size to reduce heel angle,
but the external appearance is of a conventional cruiser.
In addition, it is always possible to reef the sails as the
wind increases. The aesthetically pleasing sail plan
is balanced by the unseen heavy keel and bulb designed
according to the results of a Velocity Prediction Program which had to be specifically developed as part
CASE STUDY INCLUSION 32
The ideas previously discussed are now illustrated

using the Inclusion 32 project. This M.Sc. project completed at the Technical University of Lisbon, Portugal
(Cerveira, 2011), developed a 10 m cruising yacht
capable of costal passages designed for disabled people and those who find conventional yachts difficult to
cope with. It has all the facilities necessary for a crew
of four people to live onboard for a few days, including
a galley and heads designed to be disabled friendly.
Table 2 presents the Inclusion 32s main dimensions
and Figure 5 shows the general arrangement.
The design process of the Inclusion 32 started with
the definition of a disabled friendly general arrangement and iteratively searched for a balance between
58
Figure 5. Inclusion 32 general arrangement.
Figure 6. Pathways for wheelchair users at the single sole.
of the project in order to cope with the limit imposed

on the heel angle.
5.3
Drive and sails controls
Steering is via a driving wheel because this both

requires less physical effort and has a direct and intuitive relation between steering movements and the
resulting change in the heading of the yacht. The steering positions are at the lateral extremes of the cockpit
in order to give a clear view forward via the cabin
top sides. Wheelchair users position themselves facing
forwards because they often have difficulties twisting
the torso and neck.
The Inclusion 32 has a mainsail, a self-tacking jib
and an asymmetric spinnaker, all with furling systems and controlled from the cockpit. The operation
of the sails is the most physically demanding task
and the sails control areas were designed to promote
ergonomic postures (MacLeod, 2000). As presented in
Figure 7 the body is close to the equipment, direction
of effort is aligned with the body centreline, and there
is room below the trimming surface for the users legs
and wheelchair arm rests in order to allow a position
closer to the equipment. Below each trimming area
a space to store loose ropes avoids them becoming
mobility obstacles.
5.4
Figure 7. Ergonomic position to trim the forward sails.
based on removable benches was designed. These have

an operation similar to that of the table of the Impossible Dream catamaran shown in Figure 8, and may
be hidden at sole level to give room for wheelchairs.
This is illustrated in Figure 9 where the users on the
port side use the modular bench, which is hidden in
the figure at the starboard side for wheelchair users.
5.5 Social environment
Most people like to go sailing with friends or family
as a social event. In conventional yachts the saloon is a
dark area, lower than the cockpit with little contact to
the exterior. The Inclusion 32 is designed with a single
sole level and the saloon adjoining the cockpit to promote the social contact between the saloon users and
Modularity
In order to include both wheelchair users and nonwheel chair users in the cockpit a modular solution
59
now summarised. The single sole level common to the

cockpit and cabin avoids the risk of injuries related
with the companion ladder. The need to go up to the
deck in adverse weather conditions is reduced since
the yacht operation can be performed entirely from
the cockpit. An extensive use of handrails is made to
support the crew balance while sailing. The main
boom is designed well above average head height and
the sheeting equipment is attached to the cabin top.
The anchor is operated by an electric windlass at the
bow, avoiding the back stress associated with pulling
the anchor.
Figure 8. Table stored at sole level (www.impossible
dream.org).
The present paper uses the Universal Design concept

to include a wide range of (dis)abilities in sailing
boats. The study is focused on the more severe human
limitations often associated with disabilities.
The most important limitations to the practice of
sailing are identified and organized in order of importance, where the most important limitations require the
inclusion of permanent design features of the yacht and
hence should be considered at the initial stages of the
design process. Therefore, the first design stage should
find suitable solutions for the heel angle, balance and
mobility problems.
Next the directional and sail controls should be considered; these are non-permanent characteristics and
may change according to the particular users needs.
It would be good practice to consider at this stage a
few modular solutions. The third design stage continues after the end of the design process of the yacht
itself, as it defines the adaptations required for specific
users. In addition to this procedure considerations in
terms of injury prevention should be taken to improve
the overall safety of the project.
The case study of the Inclusion 32 presented illustrates that with simple considerations it is possible
to provide the opportunity for disabled people to
actively participate in sailing a cruising yacht, hence
increasing the sailing population. Furthermore, the
decisions taken to include disabled people are also
found to improve comfort and accessibility for the
non-disabled.
Figure 9. Inclusion 32 cockpit and saloon view.
the crew involved with the sailing tasks, as illustrated

in Figure 9. This arrangement is designed to give the
opportunity to severely disabled people to go onboard,
sit comfortably in the saloon and yet still be close to
and feel part of all the activity.
The sails can be trimmed manually or by electric
winches while the yacht may be steered manually using
the wheel, or by an autopilot. The electrical systems
can be controlled with different interfaces as described
in the adaptive systems section. The Inclusion 32 guarantees the integration of a wide range of disabilities
with mobility based on the use of wheelchairs and
modular solutions.
Disabled people often have problems with communication either due to sensory limitations or difficulty
to twist the trunk to face someone, and therefore it
was decided to use an electric engine which has a silent
operation. In addition the energy stored in the batteries
may be used to operate the electric adaptive systems.
5.6
CONCLUSIONS
REFERENCES
Allen, J. B., De Jong, M.R., 2006. Sailing and sports
medicine: a literature review. British Journal of Sports
Medicine.
Cerveira, F. 2011. Development of a Sailing Yacht for Disabled People, Master of Science Thesis, Instituto Superior
Tcnico, Technical University of Lisbon.
Goldsmith, S. 2000. Universal design A Manual of Practical
Guidance for Architects. Oxford: Architectural Press
The Disabled Sailors Association, Verity K, http://www.
disabledsailing.org/data/dsahq/verity k/Verity K.html
Impossible Dream Sailing Catamaran, http://www.
impossibledream.org.uk/idream.html
Injury prevent
The most frequent injuries onboard sailing yachts were

previously identified and the preventive considerations made during the design of the Inclusion 32 are
60
Martin 16 Power-Assisted Self-contained, portable, powerassisted steering and sail sheeting system, http://www.
martin16.com/resources/autobrochure.pdf
Sailability Online Manuals, http://www.sailability.org/
manuals.cfm
IFDS International Association for Disabled Sailing, http://
www.sailing.org/disabled/classes&adaptations.php
MacLeod, D. 2000. The Rules of Work: A practical Engineering Guide to Ergonomics. Tailor & Francis.
Nevile, V., Folland, J. P. 2009. The Epidemiology and Aetiology of Injuries in Sailing. Loughborough University.
Prince, C. J. S., Spalding, T. J. W., Mackenzie, C. 2002.

Patterns of illness and injury encountered in amateur
ocean yacht racing: an analysis of the British Telecom Round World Yacht Race. British Journal of Sports
Medicine.
Rouvillian, J.M., Mercky, F., Lethuillier, D. 2007. Injuries
on offshore cruising sailboats: analysis for means of
prevention, British Journal of Sports Medicine.
Simos, A. N., Design and construction of a training-yatch for
the physically handicapped, (in Portuguese) So Paulo.
Escola Politcnica da Universidade de So Paulo.
61

Hull shape design of a compressed natural gas ship

E.B. Malta, M.C. Nogueira, A.A. Ramos & C.M. Sampaio
TPN-Numerical Offshore Tank, University of So Paulo, Brazil
ABSTRACT: In order to design the hull shape of a compressed natural gas ship (CNG), a parametric model
was integrated with forward resistance and seakeeping commercial codes. This integration allowed dynamic
results analysis using only a program interface with external codes experimentally verified. With regard to
motions in waves, potential and strip method were used to evaluate the RAO (Response Amplitude Operator) of
the hull forms in different conditions (with/without forward speed). The seakeeping has been defined through
acceleration and comfort indices based on spectral analysis. The forward resistance was determined using nonlinear potential method for the wave resistance and finite volume method without free surface effect for the
viscous resistance. Both numerical models were validated with results achieved in the IPT towing tank. The
evaluation of some design output parameters, generated with the parametric model, assist to determine which
variables influence the most in CNG ship forward resistance and seakeeping. The same methodology can be
applied in other type of hull shape designs.
INTRODUCTION
With regard to the motion in waves, Seakeeper

(Seakeeper, 2009), which is based on the strip method
in the frequency domain, was used to obtain the RAOs
(Response Amplitude Operator). The results reliability was assured with experimental results comparation.
Another possibility was to evaluate ship motions using
potential method with Rankine sources distribution
along the hull and free surface. Nevertheless, this
assessment requires more computational time which
makes it difficult to implement in the optimization process. Therefore, the RAO results of Seakeeper defined
acceleration and comfort indices based on spectral
analysis (Lloyd, 1989).
After defining the most relevant parameters for each
analysis and the intervals at which the model is valid,
some evaluation methods were applied to define the
parameters that most influence the design.
The lack of pipelines to transport natural gas required

the development of alternatives which can efficiently
transfer the production to the coast. One solution is
to use vessels capable of transporting the compressed
natural gas (CNG). In these ships design, hull lines
are set to efficiently carry a certain amount of gas,
reducing fuel consumption by reducing the forward
resistance and maintaining acceptable levels of motion
in waves.
Nowadays, numerical simulators are an important
tool to design any offshore systems due to its capacity of evaluating a large number of conditions in a
virtual environment, which would have higher costs
and take a lot of time to do in towing tanks. For that
reason, it was created a parametric model at Friendship Framework (Abt et al., 2001) together with other
commercial codes that were first verified with some
preliminary experiments. The main dimensions used
to build the model were lengths, beams and heights
as will be explained better at Design Methodology
chapter. Besides these main dimensions, other parameters were used to change different aspects of the hull
shape, varying slopes, angles and parameters related
to stern and bulbous bow (Kracht, 1978).
For the analysis of many hull shapes, it was created
integration between the parametric model and some
analysis programs for resistance forces and seakeeping. In order to evaluate the resistance force, it was
used Shipflow program (Flowtech, 2008) for the wave
generation resistance using XPAN subroutine based
on potential method. For the viscous resistance, it was
used Star-CCM+ , which performs the calculations
using the finite volume method.
DESIGN METHODOLOGY
Vessel design process involves a series of tests among

which there are: stability, forward resistance, seakeeping, maneuverability, hull-propeller-engine performance and structural analysis.
The need for these analyses during the design process motivates integrated models. Among the analyses
that can be integrated, the focus of this work was
the integration of the forward resistance and seakeeping evaluations. In summary, the methodology
is initiated using the routines that help to determine
the main dimensions using the information available
in references as Watson (1995) and Lewis (1989).
These dimensions are entries to the parametric hull
model where the analyses can be performed and the
63
Figure 2. Hull surface creation.
Figure 1. Design methodology.
results verified according to pre-established requirements. Thus the methodology for the vessel design is
summarized in Figure 1.
First, the initial main dimensions were obtained by
the regressions in Watson (1995) and Lewis (1989),
which are described in Matlab routines (MATLAB,
2005). For instance, these routines are used to define
the necessary displacement based on transport capacity and other design requirements, so the initial main
dimensions can be set to construct the hull surface.
With this parameterization and all model evaluations described above, it was possible to define a CNG
ship hull. The same methodology can be applied to
other vessels designs.
3
Figure 3. Rake, flare and stem parameters.
is forward of the stem. The stem is defined in the

z-axis direction, which provides the beginning of the
rake above the water. The dimensionless distance of
the flare is the distance in the deck at the y-axis, as
presented in Figure 3.
PARAMETER DESCRIPTION
For the hull surface construction, it was necessary to

establish some design parameters from the dimensionless variables input. These parameters and dimensionless variables are listed below, with a brief description
of their roles in the project.
3.1
3.3 Bulb parameters

The bulb is defined from sections B08 and B09 set
of points. The first point that was parameterized sets
the height (Hb), in which the bulb starts from the hull
stem.
The remaining points are positioned in pairs, one
point in the section B09, with their coordinates in a
XZ plane, and the other in section B08, with the same
coordinates in XY plane. The points are divided to set
its z coordinate as ZBA, ZBB and BMZ. The parameters to set y coordinates on the section B08 are BBA,
BBM and BBB, as shown in Figure 4. The bulb length
is obtained with the parameters LprAB, positioning in
the x-axis direction, along with the midpoints Lprm.
Input parameters
The hull was divided into three areas: bow, stern and
parallel body. It was created two parameters to define
their length: the bow length (L_bow) and the parallel
body length (L_body). The stern length is defined as
the total length minus bow and parallel body lengths.
Throughout the hull, points were placed to form
the body plan. These points were interpolated by ten
spline curves, named as B00 up to B09, which were
used to construct the surfaces as presented at Figure 2.
To set the angle with which the ship cuts the water,
there is a parameter called (bow_angle | L_bow),
which provides the relationship as the section B07
is positioned in proportion to the length of the bow
defined by the section B06.
3.2
3.4 Stern parameters

Input parameters that define the stern are transom
stern beam (Btransom/Beam), stern transom immersion (immersion of the stern/draft), which define
aft frames points (B00 to B02). It was created two
frames which positions are responsible for stern slope
(B02 and B03). B02 and B03s position were set as
Bow parameters
Bow parameters are bow length, rake, stem and flare.

The rake is the length in the x-axis where the deck
64
Figure 8. Sectional area curve.
center in the water line area; the line area; dgua

(water plane rea); and longitudinal and transverse
moments both in the area of water line.
Figure 4. Bulb variables.
3.6 Motion in wave evaluation

The performance evaluation of a ship in waves is essential, both to ensure the integrity of the vessel as to
ensure the conditions of comfort and safety of crew
and cargo.
The model of wave behavior was analyzed by integrating with the programs that uses method of panels
and strip theory for effecting their calculations.
The Seakeeper program is based on the strip theory in the frequency domain. The main advantage is
the reduced computational time to obtain the hydrodynamic properties of two-dimensional sections through
the hull, which after integration along the length of
the vessel, resulting in coefficients of added mass,
damping and excitation forces in waves. With these
hydrodynamic coefficients can be obtained from the
movements of the vessel and, consequently, the RAO.
In the case of the method of panels, it uses Rankine
sources in the time domain, the distribution of tiles
along the hull and free-surface to thereby solve the
potential flow associated, resulting in greater accuracy
at the cost greater computational effort.
From the results of both methods was performed
a comparison between the numeric RAOs and the
experimental tests performed in the Instituto de
Pesquisas Tecnolgicas (IPT), as shown in Figure 9
and Figure 10.
Thus, it is observed that although the method
panels presents a better grip with the tests, the strip
theory (Seakeeper) shows a good trend and it is more
appropriately to be use in the process of optimizing
behavior in waves.
From the results obtained in Seakeeper is possible
through spectral crossover between the RAO of the
vessel and a JONSWAP sea spectrum, obtaining the
performance against two criteria of comfort: Motion
Sickness Incidence (MSI) from Watson (1998) and
Subjective Motion (SM) from Lloyd (1989).
Figure 5. Input parameters of the stern side view.
Figure 6. Input parameters of the stern top view.
Figure 7. Input parameters of the stern front view.
parameters called initial stern slope and final stern

slope.
Stern tangent refers to B00, B01 and B03 stern slope
in the y direction.
3.5
Hydrostatic evaluation
The hydrostatic parameters are obtained through the

curve of the sectional area shown in Figure 8. These
parameters are: the center of buoyancy, which is
the center of the submerged volume; the center of
floatation (center of floatation), which is located the
3.7 Wave resistance evaluation

The subroutine XPAN is the module responsible for
calculating the potential flow. As the output parameter
65
Figure 11. Visualization of flow around the hull through

XPAN.
Figure 9. Comparison of Heave RAO.
Figure 12. Wave resistance comparison.
Figure 13. Computational domain.
3.8 Viscous resistance evaluation

In order to obtain the viscous resistance component,
hull submerged parts were modeled in Star-CCM+ ,
which performs calculations with finite volume
method.
The phenomenon of interest occurs in steady state,
which lead to the use of RANS (Reynolds Averaged Navier-Stokes Equation) as a simplification of
Navier-Stokes equations. To complete RANSE, standard k- turbulence model was chosen. Boundary layer
was modeled using wall function.
The computational domain dimensions, important
for the use of correct boundary conditions, were
defined based on Ahmed & Guedes Soares (2009) and
Cho et al. (2010). The boundary conditions that were
Figure 10. Comparison of Pitch RAO.
coefficient of hull wave profile is obtained by waves

generated by the hull as shown in Figure 11.
For the hull was made an initial comparison of
numerical results with the experimental results, the
coefficient of wave from the experimental test was
obtained from CT (total coefficient of resistance)
across regressions (Watson, 1998).
66
Table 1.
Dimensions of the computational domain.
Direction
Size (*Lbp)
Forward
Aft
Depth
Width
1
2
1.5
1
Table 2.
Boundary conditions.
Surface
Boundary condition
Ship
Entrance plan
Exit plan
Symmetry plan
Free surface
Lateral wall
Bottom plan
Full adherence
Evanescence
Symmetry
Impermeability
Figure 16. Prismatic mesh detail at the stern of the vessel

(side view).
Mathematical
expression
V=0
V = Vaway
p = patm
Vn=0
This model was compared with experimental test

performed at the IPT. For Froude numbers between
0.07 and 0.21, the difference between numerical and
experimental results was less than 4%.
4
MODEL RESULTS
The seakeeping and resistance analysis are presented

in this chapter.
4.1 Seakeeping results
The seakeeping analysis made possible to get an
insight of the relation between parameters variation
and comfort levels at the deck (SM and MSI). The condition considered was head seas at speed of 15 knots
with 3.5 m of significant height and 9 s of peak period
spectrum.
The parameter which influences SM and MSI the
most is the non-dimensional bilge radius (Figure 17),
causing 40% and 26% variation in MSI and SM,
respectively. Its reduction from 0.3 to 0.15 improves
about 16% and 9% of the initial hull MSI and SM.
Bow angle, parallel body length, and bow length
had minor influence in comfort level. The variation of
the three parameters together results in 12% and 7%
change in MSI and SM, respectively. Their increase
until the maximum range value improved 7.5% and
4% of the initial hull MSI and SM.
The section B07 inclination had similar results but
the comfort levels were increased. Non-dimensional
bow radius had practically no influence over seakeeping. A summary of this analysis is showed in
Table 3.
Figure 18 show MSI and SM, respectively, according to bilge radius and bow length, the two parameters
which influenced the most in seakeeping.
Figure 14. Block-structured mesh hexahedrical (side view).
Figure 15. Block-structured mesh hexahedrical (top view).
applied, shown in the table below, are widely used and

can be found in several publications.
Hexahedron mesh was used, which is recommended
scenario for external flow and which generation time
is much lower than polyhedral mesh. The grid was
block-structured, i.e., the domain is divided into subregions in which the element size is determined by
the user. The advantage is the ability to refine only
regions necessary for capturing the physical phenomena investigated, avoiding the generation of unnecessary elements and reducing the mesh size. The domain
was divided into four sub-regions (Figure 14 and
Figure 15).
In two sub-regions, the elements are larger in the x
direction, since this is the predominant flow direction,
with more intense velocity gradients that do not require
a mesh as fine as in the other directions to be captured.
A prism layer mesh was also generated in which
a wall function was used to model boundary layer
(Figure 16).
The mesh has approximately 500,000 elements,
being considered light.
4.2 Wave resistance results

4.2.1 Linear analysis of the hull
Through an analysis of all parameters of the bulb
was observed that the parameters associated with
longitudinal format (as shown in Figure 19) have influence a lot like each other. The same effect can be
noticed in other directions, thus, were chosen for the
study, the average parameters for each direction.
67
Figure 19. Influence in CWTWC for the linear.
Figure 17. Influences on MSI and SM.
Table 3. Trend of the hull parameters according to comfort

index.
MSI and SM tendency
Bow angle
Parallel length
Bow length
Bilge radius
B07 inclination
Bow radius

Figure 20. Variation of CWTWC depending on the width

and length of the midpoints of the bulb.

and height of the midpoints of the bulb.
Figure 18. Bilge radius and stern length influence on MSI

and SM.
In Figure 19 it is evident the growing influence of

the width of the bulb and descending of the parameters
that define width and height.
Changing ZbM from 0.7 (initial value) to 1 reduces
the wave coefficient about 2.3%, while the change of
the other two parameters together cause a reduction of
1.5%.
The parameters combined influence can be seen in
Figure 20 and Figure 21.
Wave making resistance had a 8.4% reduction
compared to the original hull. 500 hulls, generated
using Sobol, were evaluated. Figure 22 shows pressure
Figure 22. Linear wave generation of bow optimization

(Original at top and least wave resistance at bottom).
68
Figure 23. Body plan comparison with linear free surface

(Original at right and least resistance at left).
and length of the midpoints of the bulb.
Figure 24. Influences on CWTWC for free-surface nonlinear hull.
distribution on the mean draft of the original hull (top)

and the one with smaller wave resistance (bottom).
Figure 23 shows the comparison between the hulls
frames.

and height of the midpoints of the bulb.
4.2.2 Nonlinear resistance of the hull

In order to obtain more accurate results and to
verify if the linear theory provides sufficient results,
a non-linear analysis was made. Due to higher time
consumption, fewer hulls were investigated.
As in the analysis of linear free-surface, the parameters for the same direction showed a similar influence
on CWTWC.
In Figure 24, the influences of medium parameters
in CWTWC bulb, with its parameters of length, width
and height behaved very similar conditions with linear
free-surface.
The parameter BBM showed a greater influence
and provided an improvement of approximately 1.4%.
In the case of the BMZ, which had been the most influential in the linear case, the improvement was only
0.7%. The parameters combined influence can be seen
in Figure 25 and Figure 26.
Figure 24 shows pressure distribution on the mean
draft of the original hull (top) and the one with smaller
wave resistance (bottom). Figure 25 shows the comparison between the hull which has the lowest wave
resistance and the original one.
The hull with lowest wave resistance obtained
through linear and non-linear methods have similar
frames.
Figure 27. Non-linear waves generation of bow optimization (Original at top and least wave resistance at bottom).
4.2.3 Stern shape influence in non-linear wave

resistance
The stern parameters investigated to evaluate wave
generation resistance were:
Stern initial slope;

Parallel Stern;
Stern tangent;
Stern immersion/Draft;
Transom beam/Total Beam;
Stern final slope.
Initially, a study was conducted to determine which

parameters most influence on the wave coefficient.
69
Figure 31. Non-linear waves generation of stern optimization (Original at top and least wave resistance at bottom).
Figure 28. Body plan comparison with non-linear free
surface (original at right and hull with least resistance at left).
Figure 32. Comparison of the stern body plan (Original at

right and least wave resistance at left).
Figure 29. Influence of initial stern slope, parallel stern and

stern tangent in CWTWC.
Figure 33. Viscous

parameters.
resistance
evaluation
of
stern
one can note which parameters influence the most.

Small changes of block coefficient, longitudinal center
of buoyancy position, wetted surface and displacement
were made (up to 2%).
In Figure 33, points inside same red ellipse have
same final slope, the ones contained in a purple ellipse
have same initial slope and the ones contained in a
yellow ellipse have same transom breadth.
Also in Figure 33 evidences that final and initial
slope, respectively, are the parameters which influence
viscous resistance the most.
Viscous resistance reduction is observed along
with vortex size reduction and aft pressure increase.
Figure 34 and Figure 35 show stern pressure distribution and flow streamlines of the wave resistance
optimum hull and the least viscous resistance hull,
respectively. Their geometry differences can be seen
in the side view. In the back and perspective views of
Figure 34, an intense vortex can be observed at the
vessels shoulder. The vortex decrease in size can be
Figure 30. Influence of stern immersion stern, transom

beam and final stern slope in CWTWC.
Sobol was used to generate 298 hulls. Wave coefficient variation according to 6 stern parameters can be
observed in Figure 29 and Figure 30.
Stern initial slope, stern final slope and stern tangent are the parameters that most influenced wave
making resistance. They were set as variables in an
optimization process using the tangent method and a
28.8% wave resistance reduction (relative to original
hull) was achieved.
4.2.4 Stern shape influence in viscous resistance
The viscous resistance results of the 243 hulls generated with the variation of stern parameters (final
slope, initial slope, transom beam, immersion and
stern tangent) were organized in such a manner that
70
The ship resistance results have been made by

Transverse Wave Cut Method considering the freesurface (linear and nonlinear) and comparisons of
the two hulls (with and without bulbous bow). Hull
without bulb was evaluated the influence only of the
main dimensions with linear free-surface. For the case
of bulbous bow, different conditions of free surface,
generate different magnitudes of influence. For the
linear free-surface the parameter of greatest influence
was BMZ, with a reduction of 2.5% in wave making resistance, while for the nonlinear free-surface the
parameter BBM decreased the coefficient by 1.4%.
So with this parametric model is possible to optimize the hull with the definition of an objective
function in terms of ship resistance and seakepping. This methodology could be used in other vessel
types to design ship considering the most influenced
parameters.
Figure 34. Pressure distribution and streamlines (wave

resistance optimum hull).
ACKNOWLEDGEMENTS
The authors would like to thank TRANSPETRO for
all the support during this project.
REFERENCES
Abt, C., Bade, S.D., Birk, L. & Harries, S. 2001. Parametric Hull Form Design A Step Towards One Week
Ship Design. In 8th International Symposium on Practical
Design of Ships and Other Floating Structures PRADS.
Ahmed Y. & Guedes Soares, C. 2009. Simulation of free
surface flow around a VLCC hull using viscous and potential flow methods. In Ocean Engineering, Volume 36,
pp. 691696.
Benford, H. 1991. Naval Architecture for Non-Naval
Architects, Jersey City, NJ, 239p.
Choi, J.E., Min, K.-S., Kim, J.H., Lee, S.B. & Seo,
H.W. 2010. Resistance and propulsion characteristics of
various commercial ships based on results. In CFD Ocean
Engineering, Volume 37, pp. 549566.
Flowtech International AB. 2008. Shipflow User Manual,
Gotenburgo.
Kracht, A.M. 1978. Design of Bulbous Bows, Transactions
SNAME, Vol. 86.
Lewis, E.V. 1989. (Editor), Principles of Naval Architecture,
SNAME, 2nd Edition, New York, SNAME.
Lloyd, A.R.J.M. 1989. Seakeeping: Ship Behaviour in
Rough Weather, Chapter 21 Effects of ship motions on
passengers and crew, Ellis Horwood Limited.
MATLAB USER MANUAL, 2005.
Seakeeper user manual. 2009. Formation Design Systems
Pty Ltd.
Stromgren, C. 1995. A comparison of Alternative Bow
Configurations, Marine Technology, vol. 32.
Watson, D.G. 1998. Practical Ship Design, Elsevier Ocean
Engineering Book Series.
Figure 35. Pressure distribution and streamlines (least

viscous resistance hull).
seen in the back and perspective views of Figure 35.

With regards to aft pressure, the back view shows
higher intensity and homogeneity in the least viscous
resistance hull.
5
CONCLUSIONS
The hull shape for natural gas transportation was

designed considering seakeeping and ship resistance.
For that it was developed a parametric model in
the program FriendShip and it was performed an evaluation of the input and output parameters. This was
achieved through the parametric model, where the
surfaces were defined by points according to the size
of the main vessel.
With this model it was possible to monitor seakeeping trends of each parameter in the comfort indices
MSI and SM. By the model based on strip method,
comfort levels were more influenced by the bilge
radius, where it was possible to improve the original
hull at 16% in MSI and 9% in SM.
71

Short sea shipping route and stowage plan optimization

Centro de investigao Naval, Escola Naval, Alfeite, Almada, Portugal
A. Moura
University of Aveiro, Campus Universitrio de Santiago, Aveiro, Portugal
ABSTRACT: Short Sea Shipping is the name given to the use of ships to transport cargo within a country,
or between neighboring countries. The European Union considers all intra-European sea transport as Short Sea
Shipping, and has defined its efficient use as a strategic goal. This form of transport has many recognized
advantages over rail and road, but suffers from lack of flexibility. In this paper we propose a logistic model
for short sea shipping that greatly increases flexibility, and that may reduce costs substantially. The model is
described and formalized, and a genetic algorithm based optimization procedure is presented, to assign cargo
and routes to each ship, as well as provide stowage plans. The proposed method is used on a small problem that
illustrates the advantages that can be obtained.
1
INTRODUCTION
type of cargo, since if containerships are considered

the container stowage problem has to be incorporated,
while the cargo distribution problem for bulk carriers
is simpler.
Focusing in the containership problem, there are
three different challenges that are addressed:
Short sea shipping is considered by the European

Union as one of the pillars for its transport policy
since the 2000 European Conference of Ministers of
Transport EU_Comission (2001). During the last few
years, there have been several studies about the subject,
aiming to determine when short sea shipping is more
advantageous than road transport Medda et al. (2010),
how the maritime transport has lower CO2 emissions
than land transport Vanherle and Delhaye (2010), and
several others.
Gathering data from the available sources, short sea
shipping advantages and downsides can be generically
identified. On one side, it is commonly accepted by EU
member states that:
i. how to select navigation routes for the fleet;

ii. how to distribute the cargo by the available vessels;
iii. how to stow cargo on the vessel.
2
As mentioned in Avriel et al. (1998), the problem

of efficiently operating a fleet of containerships serving many ports is composed of several sub-problems,
among them finding the optimal sizes and optimal
routing of the ships.
Generically, this can be extrapolated to cargo transport by sea, and within this subject of optimization
two major research areas have appeared, namely the
container stowage problem (CSP) Avriel et al. (1998),
Wilson and Roach (1999) and the vehicle routing
problem (VRP) Christiansen and Nygreen (1998).
(1) it is environmentally sound;

(2) it contributes to road safety;
(3) it has low infrastructure costs (the sea itself is the
motorway);
(4) it can reach most of Europes peripheral regions.
On the other, one can also point out that there are
disadvantages:
(5)
(6)
(7)
(8)
(9)
STATE OF THE ART
the bureaucracy attached to customs;

port services costs and efficiency;
travel duration;
inflexibility of routes;
dependency on environmental factors.
2.1 Container stowage problem (CSP)

As far as the CSP is concerned, following the same
approach to the problem of references Avriel et al.
(1998), Wilson and Roach (1999) and Martins, Lobo
et al. (2009), it is defined as the problem of assigning
different containers to slots on board a containership,
bearing in mind their origin and destination, as well
as the different ports of call of the vessel during her
journey, aiming to reduce the inevitable containers
The work that has been developed, tackles disadvantages (7) and (8), proposing a managing model for
a fleet of cargo ships with no pre-defined routes, and
where these are defined dynamically, depending on
cargo arrival at ports, due dates and delivery deadlines.
The model complexity depends on the vessels and
73
manoeuvres at every port. All these manoeuvres have

a related cost and time duration, and may be motivated
by several reasons, namely:
CONTAINER STOWAGE AND SHIP

ROUTING PROBLEM (CSSRP)
The Container Stowage and Ship Routing Problem

(CSSRP), as far as we know, was first presented in
Martins, Moura et al. (2010) and it incorporates both of
the previous problems applied to cargo transfer using
containerships.
Commonly, ships follow pre-determined fixed
routes visiting ports in a sequence that, in many cases,
is circular, and carrying large amounts of cargo. In
fact, as far as long distance transports are considered,
this is probably the most efficient way since ports are
apart.
On the other hand, as far as short sea shipping is
considered, ports are much closer, cargo quantities
can vary significantly as well as their origin and destination, and therefore, ship routing may affect the
overcome profit of a journey.
a. containers that are unloaded since they reached

their destination;
b. containers that have to be moved because they block
the access to others that have to be unloaded (there
are usually called overstows);
c. containers that are stacked on the top of cargo
hatches and that have to be moved to access to the
cargo bay (a particular case of the previous ones);
d. containers that are re-positioned to improve the
overall stowage (including stability), and make the
next port handling easier (which are known as
re-handles);
e. new containers that are brought on board to send to
another port of call.
When assigning containers, care must be taken to
consider several constraints related to the containership herself, such as space limitations, stability and
strength, speed, autonomy and seaworthiness; and to
the containers requirements and characteristics, which
among others are size, weight, structural resistance,
refrigeration requirements, etc.
In the literature there have been several approaches
and methods to solve the problem, such as the use of
Binary Linear Programming formulations were used
in Botter and Brinati (1992), Ambrosino, Sciomachen
et al. (2006); simulation based approaches can be
found in Wilson and Roach (1999); and other in many
other cases metaheuristics methods, as for example
Simulated Annealing (Dubrovsky et al., 2002) and
Genetic Algorithms Martins et al. (2009).
In particular, Genetic algorithms provided good
results and large flexibility when implementing a large
number of both crisp and fuzzy constraints.
3.1 Conceptual model

Assuming that in each port considered there are several containers each one with its origin, destination and
time delivery constraints (deadlines and due dates).
To transport them there is a fleet of ships available,
each one characterized with her own speed, cargo
transport limitation (weight, stability, strength). Hence
it is necessary to decide:
how to select navigation routes for the fleet;
how to distribute the cargo by the available vessels;
how to stow cargo on each vessel.
The first two sub-problems are similar to the routing problem mentioned before, however in this case
the transported cargo is not the same, and so route
planning must take into consideration that each container has its own origin and destination, which must
be taken into account when evaluating port sequence
and may require multiple calls to each port.
Further it also has to consider the ports distances
(arcs) and port (nodes). For each candidate route that
arises from the previous step CSP must be solved. In
other words, containers must be stowed taking into
account the previous both the containers position constraints (motion constraints and slot constraints) and
vessels constraints (cargo limit, stability).
All of the above must be done aiming to reduce the
net distribution cost, which is not only related with the
vessels operation and the visited portsfees, but also to
the costs of container handling in ports and the time the
vessel stays at port to do so (minimize containers shifts
at each port). Table 1 summarizes the characteristics
of this problem from a optimization point of view.
Figure 1 is an example of the model application.
One can see five ports (A, B, C, D, E) of which three
of them have containers (each rectangle stand for x
containers) to be delivered to the ports with the corresponding color within the time limit specified. To do
so there are two ships (1 black, 2 orange) that can
follow any route possible. In the figure two solutions
are specified, where the 2-orange ship follows always
2.2 Vessels routing problem (VRP)

The vessels routing problem referenced in the literature, as far as we know, is dedicated to cargo
transport in bulk carriers where only one kind of
cargo is transported per journey Appelgren (1969;
Appelgren (1971); Brown et al. (1987); Fisher and
Rosenwein (1989), though some models present time
and inventory constraints (Christiansen and Nygreen,
1998).
Such approaches use graph theory to solve the
problem. In the particular case of Christiansen and
Nygreen (1998), in which deadlines are concerned
(time windows), the problem is sub-divided into
two. It begins to consider two distinct input variables,
the ship routes and port sequence (pick up and delivery). Once several feasible ship routes are selected
there is an interaction between this solution and the
port sequence, that is formulated as a shortest path
sequence, using Dantzig Wolfe decomposition and
branch-and-bound search. In both sub-problems, time
is measured considering ports (nodes) and the travelled
distances (arcs).
74
Table 1.
CSSRP definiton of variables, constraints and cost function characteristics.
Input variables
n of containers
containers origin
containers destination
containers characteristics
(weight, dimensions, etc.)
distances between ports
n of ships
ships characteristics
Decision variables
Constraints
Cost function
ship routes
port visit sequences
containers location
within the vessel
to move a container all the ones

above must be moved
a container can only occupy a slot
at the bottom, on a hatch, or above
another one
vessels carrying limit
vessels stability and weight
distribution limits
if cargo exists at a port, than it
must be visited
minimize transportation
cost
minimize port visit cost
minimize harbors
containers shift cost
minimize time at port
V = {1, ..., kmax }

C = {1, ..., max }
Set of vessels (k) that may be used;

Set of containers c were max is
the total number of containers to
delivery;
k
Matrix (zx zy zz ) of slots placed in
vessel k. A slot is equal to 1 if it is
possible to load a container and 0
if not;
= {1, ..., max } Set of journeys () for any vessel
that represents the number of arcs
(i, j) travelled by a vessel in a route.
dij
Arc length (i, j) corresponding to
the distance between port i and j in
miles;
Lk , Bk , Tk , Dk , k , GT , Qk , s_ fuelk
Vessels k characteristics: length,
beam, draught, depth, displacement, gross tonnage, maximum
pay load weight, specific fuel
consumption;
velk
Vessels k travel velocity (assumed
constant);
q
Total weight of container ;
a
Deadline delivery time of
container ;
o ; e
Origin and destination port of
container , respectively;
l , w , h
Container dimensions of
container;
fi
Port i fixed tariffs in Euros;
rik
Cost of a movement (load or
unload) of a container placed in
vessel k in port i and journey ;
bik
Time of a movement (load or
unload) of a container placed in
vessel k in port i and journey .
Figure 1. CSSRP graphical representation.
the same route but 1-black ship can follow the dashed
route or the filled route. Further, after selecting the
route one must determine the cargo distribution that is
not represented.
4
MATHEMATICAL FORMULATION
Having introduced the concept of CSSRP and characterised the optimization problem and all that is
necessary to proceed with the formulation, then it
will be presented in several sections, beginning to
define the input, auxiliary and decision variables, the
cost function, and finally the constraints to which the
problem is subjected to.
4.1
Input variables
As mentioned before, the input variables are the ones

that characterise the problem, i.e. cargo, vessels and
ports. Therefore, assuming that:
the following variables can be defined:
4.2 Auxiliary variables
G(P, A)
P = {1, ..., p}
A directed graph;
Gs set of nodes. The nodes 1 to
p represent the ports that could be
visited;
A = {(i, j) : i, j P, i = j}
Gs set of arcs joint node i to j that
represents a journey performed by
a vessel;
This set of variables is put together to help to understand the problem and make the constraints and cost
function mathematical expressions easier:
uik = ucrew + uport Visiting cost in Euros by time unit
(hours) of vessel k at port i,
during journey (includes
crews and taxes costs);
75
ck = cfuelk + ccrewk
the second component includes the cost of containers shifts (including crane maneuvers and
time components);
the third is the cost of loading/ unloading containers.
Cost travel operation (per unit

distance, in Euros) of vessel k
(includes crews and fuel costs);
ik = 2 rik
Total cost to shift a container
placed in vessel k at port i during
journey ;
ik = 2 bik
Total time to shift a container
placed in vessel k in port i during journey , that depends of
the time needed to perform a
movement bik ;

k

Corigik = # : o = i
ik(zx ,zy ,zz ) = 1
(i,zx ,zy ,zz )
Number of containers to
load in vessel k at port i during
journey ;

k

Cdestjk = # : e = j
ik(zx ,zy ,zz ) xijk = 1
(i,j,zx ,zy ,zz )
Number of containers to
unload in vessel k at port i during
journey ;
ik
Number of shifts performed by
vessel k at port i during
journey ;
moveik = Corigik + Cdestik
Number of movements
performed by vessel k at port i
during journey ;
sik
Time at which vessel k arrives
port i in journey ;
tik = ik
ik + bik moveik
Service time of vessel k at port i
at journey .
As explained during the conceptual model discussion,

there are two sets of variables:
ik(zx ,zy ,zz )
To ensure that the solution found is viable, candidate solutions must comply with several constraints.
The ones introduced in the model are the ones stated
bellow; however, with further research there may be
the need to introduce some others.
Expression (3) presents the flow conservation constraint that aims to ensure that if a vessel k arrives to
port i then in the next journey he will part from that
same port.
Expression (4) guarantees that the ports service does

not begin before the vessel arrives into the port. The
service time depends on the time and distance travelled
by the vessel and the summation of the service time
in each previously visited port by the same vessel
(if the actual port is not the first port visited). M1 is
a Big-M constant.
4.3 Decision variables
xijk
4.5 Constraints
This is a binary variable that defines

vessel k route. Equals 1 if vessel k visits port j immediately after port i in
journey and 0 otherwise;
This is a binary variable that defines the
container position in vessel k. Equals
1 if, after vessel k leaves port i the slot
(zx zy zz ) is occupied by the container
and 0 otherwise.
Expressions (5), (6) and (7) are related to the containers and their delivery obligations. The first guarantees
that the containers deadline is not violated, the second one the weight capacity of the vessels is not
exceeded, and the third guarantees that each container
is transported by one and just one vessel between ports.
4.4 Cost function

As mentioned before, and indicated in table 1, the aim
of the problem is to minimize cost. In the formulation
we incorporated into three main components, to which
we are multiplying a cost, since it may be of interest
of the decision maker to reduce a specific kind cost in
future analysis. Hence, expression (2) adds up:
the first component includes the operation cost
between two sequenced ports and the fixed tariffs
of the port which is visited;
76
Figure 2. Chromosome creation.
Figure 3. Scenario definition.

Table 2. Results comparison of different routes to deliver
all containers specified in the example.
Expression (8) binds the container loading variables to

the vehicle routing problem variables, i.e. if a container
is placed inside vessel k then that vessel has to visit
the port to which the container is destined to go.
Expression (9) guarantees that if a slot is occupied

with a container then all the slots below are also
occupied.
generated was used, taking into account the same starting ports for each vessel. As Table 2 shows, in only one
round of visits to the ports is not possible to deliver all
containers, and therefore more than one visit to each
port was allowed. In such case, if a circular type strict
route planning is used after nine port calls there are still
100 containers that were not delivered. On the other
hand, if the ships are allowed to change their routes
as a function of cargo delivery needs, not only all containers were delivered but also only 68 containers in a
total of 1400 containers surpassed their deadlines.
These results are probably not the optimal solution
since Genetic algorithms not always provide the optimum solution and may be trapped in a local minimum
of the solutions space. Nevertheless, it is better than
most.
DISCUSSION AND RESULTS
A simulation of the routing problem for five ports

and two containerships (similar to figure 1 conceptual
drawing) was solved using genetic algorithms (GA).
Unfortunately, it was not yet introduced the container
stowage problem into the solution though this work is
underway.
The GA implementation was done using
MATLAB, with its own standard crossover and
mutation operators. As to encode possible solution we
defined chromosomes (set of genes) that incorporate
the routes for all ships, as represented in figure 2.
Figure 3 presents the scenario used as example, where each rectangle stands for 100 containers
distributed by the ports A to E, where at each port:
CONCLUSIONS
A new logistic model for the management of small

fleets of ships used in Short Sea Shipping was proposed. This model dynamically assigns routes to ships,
assigns cargo to them, and provides an optimized
stowage plan. This paper presents the full mathematical formalization of the problem, seen as a linear
optimization problem. It also presents a genetic algorithm based solution to it, which proves to be very
efficient. The proposed method was applied to a small
problem to illustrate the details of the method and the
improvement that can be achieved.
The results show that the proposed model is much
more flexible than traditional models, and can provide
better results.
50% of the containers must be delivered to the port

furthest away, 30% to next furthest, and 10% to
each one of the other two ports;
80% of the containers have a deadline to be delivered of 15 days, 10% a deadline of 8 days and the
last 10% a deadline of 20 days.
The containerships used for cargo transport are the
vessels AXE and HEAVY that are based on two real
ships, and in this scenario are assumed to be positioned
at ports A and C respectively.
The results were promising after 56 generations,
when a initial population of 30 routes randomly
77
REFERENCES
EU_Comission. 2001. White Paper on European Transport

Policy for 2010: Time to decide. COM (2001) 317 Final,
Brussels, Commission of the European Communities.
Fisher, M. L. and M. B. Rosenwein. 1989. An interactive
optimization system for bulk-cargo ship scheduling.
Naval Research Logistics 36: 2742.
Martins, T., V. Lobo, et al. 2009. Container Stowage Problem
Solution for Short Sea Shipping. 14 Congresso da Associao Portuguesa de Anlise Operacional. Universidade
Nova de Lisboa, Monte da Caparica.
Martins, T., A. Moura, et al. 2010. Genetic Algorithms
approach for containerships fleet management dependent on cargo and their deadlines. IAME 2010 Annual
Conference. C. Edies. Lisbon.
Medda, F. R., E. Pels, et al. 2010. Incentive mechanism
for the development of short sea shipping. 2010 Annual
conference of the International Association of Maritime
Economists. Lisbon, Cargo Edies, Lda.
Vanherle, K. and E. Delhaye. 2010. Road versus short sea
shipping: comparing emissions and external costs. 2010
Annual conference of the International Association of
Maritime Economists. Lisbon, Cargo Edies, Lda.
Wilson, I. D. and P. A. Roach. 1999. Principles of Combinatorial Optimization Applied to Container-Ship Stowage
Planning. Journal of Heuristics 5(4): 403418(16).
Ambrosino, D., A. Sciomachen, et al. 2006. A decomposition heuristics for the container ship stowage problem.
Journal of Heuristics 12(3): 211233.
Appelgren, L. H. 1969. A column generation algorithm for
a ship scheduling problem. Transportation Science 3:
5368.
Appelgren, L. H. 1971. Integer programming methods for
a vessel scheduling problem. Transportation Science 5:
6478.
Avriel, M., M. Penn, et al. 1998. Stowage Planning for Container Ships to Reduce the Number of Shifts. Annals of
Operations Research 76: 5571.
Botter, R. C. and M. A. Brinati. 1992. Stowage Container
Planning: A Model for Getting an Optimal Solution. IFIP
Transactions B (Applications in Technology) B-5.
Brown, G. G., G. W. Graves, et al. 1987. Scheduling ocean
transportation of crude oil. Management Science 33:
335346.
Christiansen, M. and B. Nygreen. 1998. A method for solving ship routing problemswith inventory constraints
Annals of Operations Research 81: 357378.
Dubrovsky, O., G. Levitin, et al. 2002. A genetic algorithm
with a compact solution encoding for the container ship
stowage problem. Journal of Heuristics 8: 585599.
78

Non-linear optimization applied to preliminary ship design

Joo da Cmara Ruas & Manuel Ventura
ABSTRACT: Ship preliminary design optimization has been studied in two main areas, the models that
allow the determination of the optimum characteristics and the optimization algorithms used. In this work,
special relevance is given to the nonlinear optimization algorithms, with a focus in the evolutionary and directsearch methods. First, the results of a set of selected methods are compared using a simple bulk carrier ship
model commonly used in the literature, next a new optimization process is proposed and is validated with the
mentioned model.
INTRODUCTION
et al. (1965) the computer has done in 4 minutes what

could take 1,700 man hours.
The main goal of this paper is to analyze and compare a set of non-linear optimization methods to apply
in preliminary ship design, pointing out advantages
and disadvantages.
The work is organized as follows. First, the optimization methods selected are presented, identifying
limitations and some techniques that have been developed to overcome them. The concept of Multi-Criteria
Decision Making (MCDM) is introduced. Next, the
optimization results of a bulk carrier model presented
by Sen and Yang (1998) are compared with those from
literature. None of the optimizers used reached the
results from the literature so it was decided to join two
optimizers as it was done in Ray and Sha (1995), Lee
(1999) and Wolf et al. (2004). Based on these results,
a new process of optimization, using the combination of 2 optimization algorithms is proposed. Finally,
some conclusions and suggestions for further work are
presented.
During decades the initial dimensioning of ships

during the preliminary design stage was based on diagrams and statistics obtained from samples of similar
ships (Murphy et al., 1965). Since the second half
of the XXs century more numeric methods become
popular due to the computing resources available. The
initial approaches were based on the systematic variation of a set of design variables. Each combination of
variable values defined a possible ship that was next
submitted to a set of constraints to verify if it was
technically feasible. The optimum ship was selected
from the feasible ones based on an objective function
such as the ships weight or initial cost. This method is
not considered as an optimization because the generation of new sets of variables is not directly obtained
from the results of the previous iteration. Later, methods based on direct search algorithms started to be
applied. More recently evolutionary methods based
on concepts inspired on natural phenomena, such as
genetic algorithms, particle swarm optimization, ant
colony optimization or artificial immune system have
been tested in ship design applications.
Most engineering design optimization problems
are a compromise between several objectives. The
preliminary dimensioning of a ship is one example
where design it can be useful to have multi-objective
optimizers. Although most initial optimization algorithms are single-objective they can be used in multiobjective problems using an objective function that is
the result of a weighted sum of several objective functions. Evolutionary algorithms are multi-objective,
and so the result is a set of points with equal quality
and to help on the decision were created the MultiAttribute Decision Making (MADM) techniques.
The progress described was mainly allowed by the
fast increase of the computers speed of processing,
which speeds up the calculation. According to Murphy
OPTIMIZATION METHODS
The determination of the main characteristics of a ship

is a problem that requires a non-linear method. In this
category there are three main types of methods: the
direct search, the gradient based and the evolutionary. The gradient based methods are not frequently
used because they need derivative information of the
objective function, which can be quite complicated to
obtain.
Direct search algorithms only use values of the
objective function on the initial point and nearby to
evaluate all directions and select the best direction
to make a move. The difference between these algorithms is on the operators that they use to get the
79
Table 1.
points, around the initial point, which should be evaluated. To this class of algorithms are often pointed
three problems: the convergence to global minimum
of the function is not always guaranteed, they can not,
in the original codes, support constraints and they
can not perform multi-objective optimization. It is
often stated that they are faster than the evolutionary
algorithms. In this work the direct search algorithms
tested are Hooke-Jeeves 1961), Rosenbrock (1960)
and Nelder-Mead (1965).
To guarantee the convergence of these methods it
was used a random number generator to create a sparse
set of starting points within the domain of the objective function, similarly to what is done in evolutionary
algorithms to create the initial population. Each starting point will be optimized and the best solutions will
be selected. To apply the constraints it was created an
external penalty function which should be connected
to the optimization program to penalize the solutions
that are not in the hyperspace wanted. The penalty
grows with the distance to the hyperspace wanted, if
the point evaluated is farther from the hyperspace, the
penalty will be greater and this brings in, even the initial point generated out of the hyperspace. To use these
codes for multi-objective optimization it was used a
weighted objective function, where the weights were
obtained with the Multi-Objective Decision Making
(MODM), comparison matrix followed by eigenvector
technique (Ray and Sha, 1995).
The evolutionary algorithms are based on natures
phenomena and aim to reproduce them computationally. These methods have different ways of optimizing
so it is appropriated to describe them separately. In
this work two evolutionary methods were studied, a
genetic algorithm (NSGA-II) and a particle swarm
optimization (MOPSO-CD).
Genetic Algorithms (GA) are inspired in the evolutionary theory of Darwin based on the natural selection
which assumes that the best genes survive. First an initial population is generated and then sorted from the
best to worst and to those individuals are attributed
a probability of being parents of the next generation
based on their fitness. This process is repeated until
the pre-defined number of generations or the fitness
stated is reached (Holland, 1975).
The Particle Swarm Optimization (PSO) is based
on a bird flock searching for food. The initial population of particles is spread through the search-space.
Each particle has its own memory and the swarm has
a global memory, the particles communicate with the
flock by the global memory. In the global memory it
is stored the best position reached by one particle, in
the individual memory it is stored the best position of
each particle. The point to where the particle should
fly is obtained calculating a new velocity, according
to individual and global memories (Raquel and Naval,
2005).
The evolutionary algorithms were created to perform multi-objective optimizations, and although they
can support constraints, they require a large computation time. When they are used for multi-objective
MCDM methods and applications.

Weight Calculation
Optimum Solution
Methods
CM
EM
WOF
TOPSIS
Direct Search
Evolutionary
yes
yes
no
yes
yes
yes
No
yes
optimizations the result is a set of points with equal

fitness, a Pareto frontier, so MCDM are needed to
select the optimum solution.
The MCDM are usually divided in two classes the
Multi-Attribute Decision Making (MADM) and the
Multi-Objective Decision Making (MODM). To clarify it is important to define better the objectives and
the attributes. An objective is one goal of the optimization and an attribute is the value of each solution for
an objective. So MADM only can be used when there
is a set of solutions, in other words, with the results of
evolutionary algorithms. The MODM can be used with
both classes of algorithms. The MCDM can be divided
in weights calculation to each objective or attribute,
Comparison Matrix (CM) and Entropy Method (EM)
and optimum solution methods which are Weighted
Objective Function (WOF) and Technique for Order
Preference by Similar to Ideal Solution (TOPSIS).
In Table 1 is resumed when it is possible apply each
method presented.
3
COMPARISON OF RESULTS
To validate the algorithms selected, the results are compared with those obtained by Xuebin (2009) using
Sequential Quadratic Programming (SQP). The preliminary ship model for bulk carriers used is taken
from Sen and Yang (1998), and includes 6 design variables, 3 objectives, 11 constrains and about 30 equations. The objectives are the Transportation Cost (TC),
the Light Weight (LW) and the Annual Cargo (AC).
The selection of the configuration parameters for
each method is very important for the results. So, for
direct search it was used a maximum number of 100
iterations, a minimum step length of 1010 , and 0.9 as
the first step.
The initial points for each method are generated
randomly. For each algorithm 10 runs with 500 points
were made. The Nelder-Mead algorithm needs 7 initial
points (n + 1, where n is the number of design variables). One of the initial points is generated with the
random generator and the other 6 are given (this points
are close to global minimum for TC).
For NSGA-II the dimension of the initial population
has 100 individuals and the number of generations is
1000, the probability of crossover was set 0.9 and the
mutation probability 0.33. For MOPSO-CD the number of individuals and generations are the same used
with NSGA-II, the mutation probability is 0.5 and the
non-dominated solutions in the archive, are 100.
80
Table 2. Algorithms results for transportation cost and comparison with Xuebins (2009) results.
Methods
SQP
Rosenbrock
Hooke-Jeeves
Nelder-Mead
NSGA-II
MOPSO-CD
TC [t/$]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
7.972
13 718
749 689
221.83
36.97
19.82
14.57
0.72
14.00
74 504
7 968
7.979
13 464.19
733 625.60
222.17
37.00
19.52
14.36
0.70
14.02
71 047.10
7 441.74
7.973
13 638.20
741 600.34
222.58
37.10
19.67
14.47
0.71
14.00
72 785.00
7 624.66
7.972
13 713.45
748 658.25
222.01
37.00
19.80
14.56
0.72
14.00
74 279.06
7 915.63
7.981
13 364.05
730 178.99
221.48
36.91
19.42
14.29
0.70
14.00
70 453.05
7 393.07
8.209
11 776.34
637 662.62
218.16
36.21
17.00
12.52
0.65
14.00
54 117.56
5 662.74
Table 3.
Comparison of direct search results for light weight and annual cargo with Xuebins (2009) results.
Minimization of light weight
SQP
Rosenbrock
TC [t/$]
9.474
9.4123
LW [t]
5 240.30
5 312.9809
AC [t]
386 500 386 939.8525
L [m]
150.73
152.4972
B [m]
25.12
25.4162
D [m]
13.84
13.8540
T [m]
10.39
10.3890
Cb
0.75
0.7344
V [kn]
14.00
14.0134
DWT [t]
25 000 24 998.3715
Ps [kW]
5 018
4 822.7718
Hooke-Jeeves
Maximization of annual cargo

Nelder-Mead SQP
Rosenbrock
Hooke-Jeeves
Nelder-Mead
9.4360
9.5745
10.026
10.1057
10.0259
10.1536
5 332.5935
5 351.4430
25 759
26 209.0167
25 758.7674
26 530.1051
386 682.4744 392 688.3097 1 039 267 1 037 374.3292 1 039 260.5577 1 034 000.3475
152.8956
152.0583
274.32
274.3200
274.3199
274.2913
25.4824
25.3430
45.72
45.7199
45.7199
45.7152
13.7026
13.9798
24.76
25.9971
24.7635
27.1496
10.2918
10.3918
18.03
17.9797
18.0343
17.9229
0.7380
0.7400
0.75
0.7500
0.7500
0.7459
14.0000
14.3345
18.00
17.9932
17.9999
17.9957
24 999.8433 25 023.2674
148 123
147 143.5649
148 120.1508
145 294.5224
4 841.5691
5 394.2144
32 377
32 261.0298
32 375.6503
31 796.5462
From all the tested algorithms the only one that

could reach a minimum equal to the already known
value was the Nelder-Mead and this fact is related
with to the selection of the initial points. This method
needs 7 initial points one of which is given by the
random generator and the others six points given to
algorithm are close to transportation cost minimum.
Even in multi-objective optimizations when the TC has
a larger weight the results of Nelder-Mead are slightly
better than the others, as it can be seen in Table 5 (the
weight vector is [0.6, 0.2, 0.2]). When the results of
Nelder-Mead are compared with those from SQP it is
possible to understand that the minimum is equal but
the design variables are not the same and this means
that the objective function probably has some kind of
flat region in this zone or the minimum was not
reach already.
From all the algorithms tested, the Hooke-Jeeves
is the one that kept a constant behavior, always close
to the targets but never reaching any. If there is no
better alternative this algorithm can be used but with
consciousness that the results are only close to global
minimum.
The Rosenbrock algorithm, except for the minimization of the light ship weight, has not presented
In Table 2 are presented the results obtained from

the evolutionary algorithms (NSGA-II and MOPSOCD) which were not the expected; none could reach
or do better than direct search algorithms. For direct
search, the Nelder-Mead algorithm reached the same
minimum presented in Xuebin (2009) but not with the
same values of the design variables. The result from
Hooke-Jeeves is closer to the minimum desired.
From Table 3 were excluded the results of evolutionary algorithms because the quality of the results
are very similar to the presented on Table 2. The result
of the light weight minimization and annual cargo
maximization by the Hooke-Jeeves tends to be close
to the target, but never reaching it. The Rosenbrock
method stayed close to the minimum in the light weight
optimization.
In the multi-objective optimizations the MODM
used was the comparison matrix followed by a
weighted objective function because this is the only
combination that can be applied to direct search
methods. On the multi-objective optimizations the
direct search methods showed better results, since they
almost could found lowest values of weighted objective function as it is possible to see on Table 4 and
Table 5.
81
Table 4.
Comparison of multi-objective results for weight vector [0.4, 0.2, 0.4].

Weight vector [0.4,0.2,0.4]
Methods
SQP
Rosenbrock
Hooke
Nelder
NSGA-II
MOPSO
TC [/t]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
WOF
9.342
5 385.50
386 947
154.39
25.73
13.84
10.39
0.72
14.00
25 000
4 590
0.8231
9.2563
6 116.5376
388 035.1774
166.6534
27.7534
15.0399
10.3890
0.6318
14.0024
25 001.1719
3 605.6824
0.8472
9.3268
5 422.4521
387 033.6993
155.2534
25.8721
13.8036
10.3625
0.7131
14.0000
25 001.2227
4 521.0396
0.8239
9.2178
5 593.2529
387 512.7990
159.2689
26.5417
13.8700
10.3899
0.6797
14.0004
25 006.1016
4 114.9004
0.8251
9.8976
6 222.5703
457 865.3409
159.7290
16.5450
14.5410
10.8260
0.7470
15.6220
28 923.6797
8 511.1406
0.9103
9.3200
5 628.7866
397 842.2862
157.6780
25.6824
14.1106
10.4561
0.7282
14.0009
25 883.0781
5 005.3921
0.8356
Table 5.
Comparison of multi-objective results for weight vector [0.6, 0.2, 0.2].

Weight vector [0.6,0.2,0. 2]
Methods
SQP
Rosenbrock
Hooke
Nelder
NSGA-II
MOPSO
TC [/t]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
WOF
9.184
5 657.50
387 590
160.88
26.81
13.84
10.39
0.68
14.00
25 000
4 146
0.9816
9.1178
5 818.8140
406 495.2195
160.7575
26.7257
14.1261
10.5883
0.6973
14.0189
26 695.1641
4 486.3672
0.9865
9.1882
5 666.4092
387 599.3657
155.2534
25.8721
13.8036
10.3625
0.7131
14.0000
24 999.8906
3 972.5833
0.9824
9.2270
5 564.6353
388 623.4802
158.4512
26.4065
13.8610
10.4027
0.6876
14.0000
25 006.1016
4 114.9004
0.9816
8.9990
7 276.1634
513 602.1627
171.1980
28.3830
15.5210
11.5530
0.7480
14.9000
35 764.2969
7 603.8379
1.0538
9.0461
6 081.3940
432 215.4910
161.0240
26.7639
14.6603
10.8848
0.7317
14.0440
29 100.3516
5 180.7207
0.9961
good results and it was the one with the lowest

performance from the direct search methods.
From the evolutionary algorithms MOPSO-CD
was the best, presenting better results than NSGA-II.
In Xuebin (2009) it is stated that for multi-objective
optimizations the evolutionary algorithms are better
than direct search methods. In many papers (Xuebin,
2009; Parsons and Scott, 2004) the weighted objective
function has been incorrectly formulated and this has
been leading to wrong conclusions. The direct search
methods results are often compared with evolutionary
algorithms results but using for the first a CM followed by WOF and for the seconds others methods of
MADM.
Generically a weighted objective function is presented as equation (1), where wk is the weight of
objective k, fk0 (x) is the minimum (maximum) found
in a single objective optimization and fk (x) the value of
each hypotheses of solution during the optimization.
The results presented on Table 2 and Table 3 from single objective optimization are introduced in equation
(1) and produces the following equation
Weighted Objective Function (WOF) if it was correctly

used should be, in the limit 1 because the weight sum
is 1 and the three fractions should be 1 in the limit
(if the same solution reaches the optimum transportation cost, light weight and annual cargo). It is possible
to see in the results presented that all values of WOF
are smaller than 1 and this means, according to WOF
definition that every solution is better than the ideal
one, which can not happen. When a WOF is used the
algorithm tries to reach the minimum of this function,
if there is only objectives to minimize the formulation
presented in equation (2) should be used, keeping in
mind that 1 039 267 must be a minimum. When there
are objectives to minimize and maximize this function
could not be used like is, the fraction correspondent to
maximization should be inverted. If this fraction is not
82
Table 6.
Comparison of results for different weighted objective function formulation presented on equation (3) and (4).
Equation (3)
Equation (4)
Methods
Rosenbrock
Hooke
Nelder
Rosenbrock
Hooke
Nelder
TC [/t]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
WOF
9.4493
6 007.1616
387 359.7281
167.7588
25.1333
14.6920
10.3442
0.6936
14.0001
2 500.1934
4 199.7573
0.3727
9.6593
6 246.6753
387 256.9873
164.8020
26.8282
15.4868
9.8768
0.6981
14.0002
2 500.5703
4 303.9883
0.3726
9.6265
6 567.6147
387 551.7091
179.1329
25.7336
14.1162
9.7963
0.6820
14.0000
2 500.3242
4 015.6990
0.3729
9.9736
25 481.4492
1 031 460.0104
274.3200
45.7200
24.1922
17.6345
0.7499
17.9146
144 520.5625
31 271.7344
1.0076
10.0258
25 758.2363
1 039 247.5449
274.3197
45.7198
24.7625
18.0337
0.7500
17.9997
148 114.4219
32 373.4082
1.0000
10.1725
26 607.2363
1 035 821.3138
274.3200
45.7176
27.1254
17.9437
0.7500
17.9816
146 389.5625
32 131.8867
1.0033
Table 7.
Comparison results for different MADM methods applied to evolutionary algorithm.

EM TOPSIS
CM TOPSIS
EM WOF
CM WOF
Method
MOPSO-CD
NSGA-II
MOPSO-CD
NSGA-II
MOPSO-CD
NSGA-II
MOPSO-CD
TC [t/$]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
WOF
wCC
wPL
wCA
9.8377
6 112.45
447 153
159.610
26.4071
14.5056
10.7565
0.7347
15.5489
28 029.4
7 901.19
0.8333
9.9611
6 984.32
507 901
166.286
27.6670
15.1160
11.2730
0.7470
16.9550
32 725.8
10 661.72
0.8592
8.8481
9 041.24
604 037
185.902
30.8858
16.9054
12.4067
0.7415
15.4947
45 100.6
9 783.74
0.6659
8.4889
10 188.97
647 486
194.960
32.1990
17.9880
13.0390
0.7480
14.9810
52 567.4
9 290.21
0.6718
9.6640
5 497.87
398 693
154.904
25.5156
13.8850
10.3329
0.7300
14.6518
25 061.1
5 722.75
0.8764
10.1714
6 485.81
476 984
161.465
26.9100
14.6430
10.9250
0.7480
16.1460
29 909.0
10 055.24
1.0620
9.3849
5 546.37
397 893
155.402
25.8058
13.9385
10.9704
0.7373
14.0698
25 883.1
5 005.39
0.8720
9.0467
6 760.47
468 811
169.777
27.072
14.751
11.024
0.748
14.386
32 087.1
6 165.37
0.9577
0.0112
0.7265
0.2623
0.0448
0.7278
0.2274
0.3333
0.3333
0.3333
0.3333
0.3333
0.3333
0.0112
0.7265
0.2623
0.0448
0.7278
0.2274
0.3333
0.3333
0.3333
0.3333
0.3333
0.3333
inverted, every value of f3 (x) obtained by a combination of design variables will produce a value smaller
than 1 039 267 because this is the maximum, if this
value is smaller this fraction result will be less than
1 and that is why the WOF assumes values smaller
than 1. For an ideal design variables combination the
WOF value, as formulated in equation (2), is 1 but
tends to 1 by lower values and because the optimizers tries to get the minimum, it will always search for
smaller values of f3 (x).
On the other hand, if the weight of TC and LW tends
to 0 the weight of AC tends to 1 because the weight
sum has to be 1. In this case the two different formulations, the used in literature with all fraction with the
value of one objective optimization in the denominator
and the proposed one where the maximization fraction
should be inverted, will lead to equations (3) and (4),
respectively.
NSGA-II
The results of this optimization are presented in Table

6 where it is possible to see that the values for annual
cargo are much higher to equation (4) than to equation
(3). From the comparison of multi-objective optimization with the in Table 6, it is possible to see that
the annual cargo values are closer to the equation (3)
results and not to equation (4) as it was expected. This
leads to the conclusion that with the WOF formulation
presented in equation (2) the 3 objectives are being
minimized and not the annual cargo carried maximized
and the transportation cost and the light ship weight
minimized, as expected.
In Table 7 are shown results of the several MADM
combinations made with the methods presented in
Table 1 for multi-objective optimization with evolutionary algorithms. The combination TD-TOPSIS has
given the smaller transportation cost, a higher annual
cargo and a reasonable light weight. The entropy
83
Table 8.
TC [t/$]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
Table 9.
Comparison of proposed optimization process results for single objective optimization.

Transportation cost
Light weight
Annual cargo
SQP
HN
SQP
HN
SQP
HN
7.972
13 718
749 689
221.83
36.97
19.82
14.57
0.72
14.00
74 504
7 968
7.9723
13 720.4668
749 720.3653
221.8665
36.9777
19.8207
14.5744
0.7199
14.0000
74 509.9531
7 966.0117
9.474
5 240.3
386 500
150.73
25.12
13.84
10.39
0.75
14.00
25 000
5 018
9.4745
5 240.4160
386 504.4386
150.7273
25.1212
13.8415
10.3890
0.7500
14.0001
25 000.2363
5 017.7534
10.026
25 759
1 039 267
274.32
45.72
24.76
18.03
0.75
18.00
148 123
32 377
10.0259
25 758.8848
1 039 264.5763
274.3199
45.7200
24.7636
18.0345
0.7500
17.9999
148 122.3594
32 375.9375
Comparison of proposed optimization process results for multi-objective optimization.

[0.333, 0.333, 0.333]
[0.4, 0.2, 0.4]
[0.6, 0.2, 0.2]
[0.2, 0.2, 0.6]
Methods
SQP
NH
SQP
NH
SQP
NH
MIGA
NH
TC [t/$]
LW [t]
AC [t]
L [m]
B [m]
D [m]
T [m]
Cb
V [kn]
DWT [t]
Ps [kW]
WOF
9.474
5 240.3
386 500
150.73
25.12
13.84
10.39
0.75
14.00
25 000
5 018
0.8526
9.4746
5 240.6807
386 512.7028
150.7316
25.1219
13.8411
10.3887
0.7500
14.0002
25 000.8046
5 017.9097
0.8526
9.3416
5 385.5
386 947
154.39
25.73
13.84
10.39
0.718
14.00
25 000
4 590
0.8231
9.3466
5 379.4438
386 949.6856
154.2359
25.7058
13.8416
10.3890
0.7196
14.0002
25 001.7871
4 607.5327
0.8232
9.1836
5 657.5
387 590
160.88
26.81
13.84
10.39
0.684
14.00
25 000
4 146
0.9816
9.2224
5 572.7285
387 433.3337
158.8981
26.4830
13.8415
10.3890
0.6823
14.0000
25 001.7344
4 144.4189
0.9814
9.4727
5 247.9
386 760
150.88
25.12
13.84
10.39
0.75
14.00
25 022
5 018
0.6612
9.4747
5 240.5000
386 501.1597
150.7302
25.1206
13.8416
10.3890
0.7500
14.0002
25 000.0117
5 017.8301
0.6608
previously results if the second algorithm receives

7 points near the same minimum it will find the
minimum.
MOPSO-CD with Hooke-Jeeves this is a resource
process since the Hooke-Jeeves never found a
minimum but it guarantees a good approximation.
Hooke-Jeeves with Nelder-Mead this is unusual
process because it joins two direct search methods,
the advantage of this process is that the first algorithm Hooke-Jeeves will return a set of points much
closer to minimum than MOPSO-CD.
method is not a good choice for this problem because

it is giving a lot of importance to light weight, always
above 70%, and less importance to transportation cost,
near 5% or less, which lead to a transportation cost
2 dollars per ton more than the minimum. The overall best solution was obtained with NSGA-II. This
combinations methods have produced very different
solutions but they where obtained with the same Pareto
frontier points.
Since none of the tested algorithms proved to be
good enough to reach the goal of the optimizations
already known, it was decided to do another approach
joining two algorithms, following what was done by
Ray and Sha (1994), Wolf et al. (2004) and Lee (1999).
In the referenced cases were used evolutionary algorithms with large convergence criteria to search for
possible minimums zones and then were used direct
search methods to find the global minimum. From the
overall analyses of the results the following combined
processes were selected:
From the 3 options tested, Hooke-Jeeves followed by

Nelder-Mead returned the best results. A comparison
with the known results from literature (Xuebin, 2009;
Parsons and Scott, 2004; Hiroyasu et al., 2002) showed
that this process can find all the minimums for the
different objective functions.
In Table 8 are shown the results obtained from
single objective optimizations (transportation cost,
light weight and annual cargo) and prove that the
proposed optimization process Hooke-Jeeves NelderMead (HN) can find the minimums and maximums
MOPSO-CD with Nelder-Mead, this is typically

the configuration found in literature and based on
84
REFERENCES
already known. These results were taken with only

one run of 500 points. For the light weight and annual
cargo the design variables are the same as founded by
SQP (Xuebin, 2009) if rounded. Optimum ship for
TC is almost the same but with lower shaft power
and higher AC values, which means this is a better
solution, although the difference can be considered
irrelevant.
The results from multi-objective optimization are
in Table 9 where it is possible to compare the proposed process results with the SQP (Xuebin, 2009)
and MIGA (Hiroyasu et al., 2002) results which are
the best obtained. The results for HN where obtained
with one run of 500 points. For the first weight vector the minimum is the same and for [0.4, 0.2, 0.4] is
slightly worse. For the last two weight vectors, the HN
performed better and could found new minimums of
these functions, but in both cases they are very close
to the old ones.
Hiroyasu, T.; Yoshida, J.; Sano, M.; Fukunaga, T.; Kataura, T.,
2002 Distributed Genetic Algorithms ga2k Specification, Doshisha University.
Holland, J., 1975, Adaptation in Natural and Artificial
Systems, The University of Michigan Press.
Hooke R., Jeeves, T., 1961, Direct Search Solution of
Numerical and Statistical Problems, Journal of the
ACM, Vol. 8, N 2, pp. 212229, April.
Lee, D., 1999, Hybrid System Approach to Optimum
Design of a Ship, Artificial Intelligence for Engineering
Design, Vol. 13, pp. 111.
Murphy, R. D.; Sabat, D. J.; Taylor, R. J., 1965, Least Cost
Ship Characteristics by Computer Techniques, Marine
Technology, Vol. 2, pp. 174202.
Parsons, M. G.; Scott, L. R., 2004, Formulation of
Multicriteria Design Optimization Problems for Solution
with Scalar Numerical Optimization Methods, Journal
of Ship Research, Vol. 48, N 1, pp. 6176.
Nelder, J. A.; Mead, R., 1965, A Simplex Method for
Function Minimization, Computational Journal, N 7,
pp. 308313.
Parsons, M. G.; Scott, L. R., 2004, Formulation of
Multicriteria Design Optimization Problems for Solution
with Scalar Numerical Optimization Methods, Journal
of Ship Research, Vol. 48, N 1, pp. 6176.
Raquel, C.R.; Naval, P.C., 2005, An Effective Use of
Crowding Distance in Multiobjective Particle Swarm
Optimization, Proceedings of the 2005 conference on
Genetic and Evolutionary Computation, June 2529,
Washington DC, USA.
Ray, T.; Gokarn, R. P.; Sha, O. P., 1995, A Global Optimization Model for Ship Design, Computers in Industry,
Vol. 26, pp. 175192.
Rosenbrock, H. H., 1960, An Automatic Method for Finding
the Greatest Least Value of a Function, Computational
Journal, Vol. 3, pp. 175184.
Sen, P.;Yang. J.-B., 1998, Multiple Criteria Decision Support
in Engineering Design, Springer, London.
Wolf, R.; Dickmann, J.; Boas, R., 2004, Ship Design
Heuristic Optimization Methods, American Institute of
Aeronautics and Astronautics.
Xuebin, Li, 2009, Multiobjective Optimization and
Multiattribute Decision Making Study of Ships Principal Parameters in Conceptual Design, Journal of Ship
Research, Vol. 53, No. 2.
CONCLUSIONS
The optimization process proposed in this paper has

shown by the results presented in Table 8 and Table 9
that produces good results and it can reach at least the
same result of the other optimizers. The direct search
methods have been disregarded but in the optimizations performed they proved to be so efficient or better
than the evolutionary algorithm.
From the comparison of direct search algorithms
with evolutionary algorithms one may conclude that
the latter are more efficient, but these conclusions are
based on an incorrect formulation of weighted objective function. So it is proposed for future optimization
that every time that has a maximization objective, the
fraction correspondent should be inverted.
The study of the application of the TOPSIS methodology to the direct search algorithm which leads to a
more fair comparison with evolutionary algorithms is
a topic for further research and provides another tool
for engineering design.
85

Planning a fleet of containerships for a given set of ports

C.A. Silva
Presently at Estaleiros Navais de Viana do Castelo, S.A., Viana do Castelo, Portugal
C. Guedes Soares
ABSTRACT: The potential growth of short sea shipping motivated the study of a methodology to be used as
a decision support tool in determining the composition of the fleet that is required for a given market scenario.
The model allows both the parameters regarding the demand of markets and the characteristics of the fleet to be
varied so as to evaluate the sensitivity of the response. The tool also provides enough information to prepare the
fleet deployment, establishing its routes and scales in the ports of a particular scenario. The adopted methodology
may be divided in two parts, being the first one related to the generation of all feasible routes, with parameters
specific to each route for each vessel. The second part is the development of a linear programming model to
minimize the total cost of the shipping operation. The models were structured to present three main points: the
evaluation of the fleet for each vessel class; the optimal route for each vessel and the frequency in each port.
To provide a supporting base towards the mandatory validation, a fictitious case study is presented and several
sensitivity analyses are developed to test the models robustness and coherence.
1
INTRODUCTION
Consequently, the data related to the considered vessels will be inserted in the model by means of input
parameters which allow posterior sensitivity analysis. The number of the initially available vessels is
defined as a model input parameter; therefore it is
considered a limited resources problem.
The problems structure is divided in two
parts: firstly the evaluation of all feasible routes
and calculation of parameters related with each operation (route) and each type of vessel considered
(such as the involved costs, number of round trips,
etc.). Such results will provide valuable input to
the second part which considers the development
and implementation of a mathematical model on
an appropriate programming language, whose main
objective is to establish a schedule of routes for each
vessel to minimizing operational costs.
The frequency of scaling each of the ports is automatically defined with the choice of the best routes
for all vessels, since the number of possible trips has
been calculated in the first model and the best route is
defined in order to meet the required annual demand
service. The procedure or script to solve the problem
is shown in Figure 1.
In order to enable testing different scenarios, by
varying any of the parameters mentioned above, the
models need to have robust structures. As such,
the final calibration and validation will be achieved
through an example and later by the development
of a parameter sensitivity analysis of the operations
This paper addresses the need of ship owners to

determine the composition of a fleet of container vessels to meet a required demand level of the market.
This is achieved by the development of a methodology implemented in a decision support tool, in which
the parameters related with the demand for goods
in different markets and the fleet characteristics can
be matched (for example, through the assignment of
routes, ports calls and vessels characteristics).
The developed model is from the point of view
of the ship owner. Thus, the models main objective
is minimizing the operations cost for a given minimum
service level. It is assumed that the navigation company will not operate using chartered vessels, either by
voyage or time (Sherali et al. 1999), since the objective
function would not apply to this particular situation.
Due to the periodic nature of line transportation,
the time period of one year is used as the reference
duration for the study, which includes the allocation of
ships in the transport operation, the setting of the
best route for each vessel trying to reach the expected
calls frequency (days required between visits to the
same port). Although the fleet design considers factors related with the vessels characteristics, such as
speed, capacity and costs, among others, directly influencing the calculation (Bausch et al. 1991, Brown et
al. 1987), the main focus of this work is not the
characterization of the fleet itself, but its operation.
87
Table 1.
Ascend
Route definition.
Descend
1 2 3 4 5 6 6 5 4 3 2 1
1
1
1
1
1
Figure 1. Model Structure.
X
X
X
X
1
1
1
1
1
considered (based on the scheduling models of

Fagerholt 2000 and fleet deployment of Perakis 1991).
2
IDENTIFICATION OF THE VIABLE ROUTES
Total
Due to the need of providing a reliable service, so

that the costumers may schedule their transport service ship owners have came to the conclusion that
it is preferable to consider cyclic trips, with different routes on each way/return trip, while maintaining
a constant port calls frequency (despite not being a
optimal solution, sometimes considering costs higher
than desirable for ship owners).
Following this basic concept it is considered that
a feasible route includes a sequence of ports or a
number of possible trips, represented by arcs (arc
refers to a voyage between two ports within the route,
i.e., a route is composed by several arcs), to be operated by a vessel. Since it is possible for the route
to have distinct paths in both directions (commonly
referred as ascending-descending directions or NorthSouth bounding), the route is composed of two way
trip itineraries.
In practical operations, each itinerary cannot consider alternate directions to the final destination port,
i.e. if the vessel follows a trajectory towards the north
or south, she should keep that direction until the final
stretch of her itinerary. In order to maintain continuity
of the voyages, the cyclicity of the route must be considered, meaning the arrival and departure ports must
be the same.
The methodology used to calculate the number of
viable routes is illustrated in Table 1 (for example, considering 6 ports). A similar logic is applied for different
number of ports.
The program to create routes was developed in
a manner that, for a given sequence or round trip,
a binary variable x for each port defines if it is
included or not on that specific route. This variable
will present a value of 1 for the ports that are
scaled and 0 otherwise. Through this method, the
ordering of ports will imply a logical sequence of
voyages, avoiding alternating directions, as described
before.
The number of viable routes for each row is given
in the right most column of table 1, being this value
X X X 1 1 X X X
X X 1
1 X X
X 1
1 X
1
1
X
X
X
1
1
1
1
X X 1 1 X X X
X 1
1 X X
1
1 X
1
X X 1 1 X X 1
X 1
1 X 1
1
1 1
1 X 1 1 X 1
1 1
1 1
1 1 1 1
X
X
X
X
1
1
1
1
1
1
1
1
1
1
Combinations
=
=
=
=
=
=
=
=
=
=
=
=
=
=
256
64
16
4
1
64
16
4
1
16
4
1
4
1
1
453
equal to the power of two (2) and exponent, the number of binary variables on the line. The formulation of
equation 1, which calculates the number of alternatives
(R) for this problem, is shown in:
where n is the number of ports considered.

The program was developed in Fortran, providing
automatically all viable alternatives and identifying for
each route the ports that the vessel must call.
It is possible to make an assessment of the aspects
that characterize each of these routes, based on this
knowledge (for example with respect to travelled distances, travel time, number of round trips, maximum
transport capacity during the adopted time horizon
and, ultimately, the routes fixed costs and profits).
3
DETERMINATION OF THE OPTIMUM

ROUTES FOR A FLEET
The previous results (the routes) will serve as input

for the optimization model, which is formulated as
a mixed linear programming mathematical model in
(commonly known as MIP problem). The second part
should be structured to select the best route for each
fleets vessel, while meeting the restrictions of demand
between each pair of ports considered origin destination. The demand parameter corresponds to one year
period, and therefore the required cargo transfer can
be performed in several trips by different routes and
using different ships. The selection and division of how
it will be transported should be the expected response
of the model.
The fleets optimum size will be obtained from
several operational scenarios, evaluating the balance
between the overall cost and the provided level of
88
3.4 Restrictions
service. Since a given number of ships with specific

characteristics are given as input to the model, it is
considered a problem of limited resources.
3.1
The vessels exclusivity restriction to a single route is

given by:
Indexes and parameters
Throughout the mathematical models presentation the

following index and parameters will be used:
This equation ensures that each vessel is assigned to a
single route, i.e., the unit cannot be assigned to more
than one route through the process. Once assigned to a
route, it is served until the end of the planning horizon
adopted by that vessel. Moreover, the restriction allows
more than one vessel to be assigned to the same route.
The restriction for meeting the required demand is
defined as follows:
r, indicating the route, consisting of a sequence of

ports in both directions (considering a round trip with
different itineraries in each direction). It ranges from
1 to R, where R is the number of routes available to the
model and depends on the amount of ports considered;
i, indicates the ports of origin, receiving a specific
port code which may have any denomination, as long
as it is in a standardized format;
j, which indicates the destination ports and must
receive any of the codes presented in the previous
presented index;
k, indicating the vessel. Ranges from 1 to n, where
n is the maximum number of vessels available. There
may be more than a vessel of the same type and in this
case, each vessel will receive a different k allocation;
Cr,k , Annual cost of route r, when operated by
vessel type k;
Capk , Capacity of the vessel k in TEUs;
Di,j , annual demand from port i to port j, in TEUs;
Nr,k , Maximum number of round trips by vessel
k on route r, calculated during the definition phase of
the viable routes.
This equation ensures the attendance of the annual

demand rate from port i to port j, considering all chosen
routes and all vessels. The number of units available
for the model will be greater or equal to those required
to cover all the required demand. Equation 3 considers
that the sum, for all routes and ships, of the product
between the number of trips made by each vessel on
the route and the amount of cargo between ports, must
be larger or equal to the annual cargo demand between
the ports of i and j index.
The restriction which establishes a relation between
the existence of the flow variable M and the existence of the arc on the route is given by:
3.2 Decision variables

The choice of routes will be done using a binary variable associated with a vessel and a route. (Xr,k ), which
takes the value 1 if route r is assigned to vessel
k and 0 otherwise.
A variable Mr,k,i,j defines the flow of cargo between
the port of origin to the destination port (indexes i and
j) on the route r in one voyage of vessel k. This variable
was created to have control over the amount of cargo on
a vessel in order to comply with the maximum capacity
of the vessel on each voyage. It is worth mentioning
that the model will consider the value for Mr,k,i,j as
being the same for all trips during the one year planning
time horizon. For this variable a relaxation is allowed,
being defined as real and not negative.
Only on chosen routes will there be cargo transferred

between ports (index i and j) by whichever vessel k is
assigned to the route of index r. Mathematically, Mr,k,i,j
is greater than zero when Xr,k receives the amount
equal to one.
The restriction on vessels capacity is defined in
equation 6, ensuring that for each pair of ports i and
j (between two consecutive ports on a single voyage),
all transferred cargo does not exceed the maximum
capacity of the vessel:
3.3 Objective function

The models main objective is to present an appropriate scheduling for each vessel, while respecting the
restrictions, by minimizing the overall transport operational cost. The total cost to be minimized is the
sum over all vessels (and consequently over all chosen
routes), of the product of the cost associated with each
unit in its route and the decision variable:
The multiplier shown inside the sum of each pair of

ports i and j corresponds to the characteristics of the
journey and the route where the pair i-j is inserted,
i.e. for each voyage within the route, all potential cargoes that may be present aboard the vessel must be
considered.
The great difficulty to express mathematically such
a restriction is due to the cyclic nature of the route,
i.e. the same port can be characterized as an origin
89
Figure 2. Cargo composition example.
Figure 3a. Capacity restrain Component a.
or destination. Figure 2 shows an example of a route

including five ports in which the cargoes in the vessel
during the voyage from port B to port C consider the
presented composition.
In short, for the voyage between the port of index i
and j, cargo is loaded in port i to be shipped to port j
and later ones, and the cargo loaded in the ports previous to i is shipped to port j and later. The sum for all
pairs source destination of the cargo within a specific voyage should be less or equal to the capacity of
the vessel assigned to the route.
The characterization of the routes periodicity
obtained from the table of existing routes, is achieved
through the vector T (r,l), where r identifies the route
and l the voyage defined by the initial port. This
vector takes the value 1 when the port is visited and
0 otherwise. Some auxiliary vectors are introduced
allowing the mathematical model to acknowledge the
upper and lower limits, regarding the origins and destinations for each component of the equation 6. The
size of these vectors varies with the number of ports.
The first vector VO defines the logical sequence
of the visited ports, varying from 1 to n on the ascending direction and from n to 1 in the opposite direction
(descending). The second vector VI defines the port
of origin for each voyage of the route, presenting the
value 1 for all ports included in the ascending direction and n for all ports in the descending direction.
The third and final auxiliary vector VF relates to the
destination port of each voyage, presenting the value
n for all ascending ports and 2*(n 1) for ports in the
descending direction. These vectors allow the model to
identify and classify all intervals included on the chosen route in regards to origin, destination and direction,
allowing the definition of the route as a cycle of trips.
The mathematical formulation for each of the possible components of this equation will be described
below in a more general and mathematically consistent manner. The conditions for the existence of each
restriction were defined according to the sense of
direction for both its origins and destinations. Eventually, one of the conditions shall always assume value
0 when indifferent (no cargos present). Considering
the case concerning a generic interval l in any route,
in which we must consider all the loads that potentially, could be present. The equation of constraint will
be composed by the components presented next. In the
figures included to facilitate the understanding of each
component (Figures 3a to 3g), the upper row represents
a away/ascending trip and the lower row represents the
back/descending trip (if a specific square is filled in
Figure 3b. Capacity restrain Component b.
grey the port is called, if it is blank then the port is not

called).
a) Cargos originated in ports prior to voyage l and
destined to ports visited and prior to the origin.
* [T (r, li )][ Not [T (r, N li )]] * [T (r, lj ) or

T (r, N lj )]
This component corresponds to the cargo originating from any port previous to the origin port D,
which is visited only in the ascending direction
(mathematically, [T (r, li )][ Not [T (r, N li )]]), and
destined to any port previous to the source port
C, in both directions (mathematically, [T (r, lj ) or
T (r, N lj )]).
b) Cargos originated on ports before voyage l and
destined to ports visited, before voyage l, but after
the origin.
* [T (r, li ) or T (r, N li )] * [Not [T (r, lj )]

[T (r, N lj )]]
This portion corresponds to the cargoes originating
from any port denoted with an index lower or equal
to D 1 visited in whatever direction (mathematically, [T (r, li ) or T (r, N li )]) and destination to
any port after the origin A, but before D and visited
in the descending direction (mathematically, [Not
[T (r, lj )] [T (r, N lj )]]), as shown in Figure 3b.
c) Cargos originated on ports prior to voyage l and
destined to all visited ports succeeded to voyage l.
90
Figure 3c. Capacity restrain Component c.
Figure 3f. Capacity restrain Component f .
Figure 3d. Capacity restrain Component d.

Figure 3g. Capacity restrain Component g.
denoted with an index after C + 1 (for both directions), i.e., all ports posterior to C are part of this
equation.
f) Cargos originated in ports after voyage l and
destined to ports after the origin.
Figure 3e. Capacity restrain Component e.
* [T (r, li ) or T (r, N li )] * [T (r, lj ) or T(r, N lj )]

This part, depicted in Figure 3c, corresponds to all
loads originated in any port prior to C (in both directions) and destined to ant port posterior to C (in any
direction). The solely necessary condition for this
component is that the ports of origin and destination
must be visited.
d) Cargos with origin in voyage l and destined to ports
before voyage l.
* [T (r, N li )][ Not [T (r, li )]] * [T (r, lj ) or

T (r, N lj )]
This component, depicted in figure 3f, corresponds
to cargo loaded in ports after C (visited only on
the descending direction) and destined to any port
denoted with an index greater than C and the origin
D (in any direction).
g) Cargos originated in ports after voyage l and
destined to ports prior to the origin.
* [T (r, lj ) or T (r, N lj )] * [Not T (r, N l)]

This component, illustrated in Figure 3d, refers to
loads originated in C (when visited only in the
ascending direction) and destined to ports denoted
with an index prior to C 1 (in any direction).
e) Cargos originated in voyage l and destined to ports
prior to voyage l, as represented in Figure 3e.
* [T (r, li ) or T(r, N li )]] * [T (r, lj )]

[Not [T (r, N lj )]]
This component refers to all cargo loaded in
any port denoted with an index greater than C + 1
(visited in any direction) and destined to ports
before the origin E (only visited on the ascending
direction, as represented in Figure 3g).
Therefore, the final expression regarding the capacity restrain is the sum of all components, a trough
g, as described in the following equation 7.
* [T (r, lj ) or T (r, N lj )]
This component refers to cargoes originated in C
(on the ascending direction) and destined to ports
91
Table 2.
results.
Four (4) ports and two (2) vessels optimization
Vessel K_1
Vessel K_2
Total cost
Average frequency
933.150,00
4 days
Capacity (TEU)
Route:
Occupation
Capacity (TEU)
Route:
Occupation
670
R_2
5%
900
R_23
70%
Table 3.
R_2 Port 1 Port 2 0

0
0
Port 2 Port 1
R_23 Port 1 0
Port 3 Port 4 Port 3 Port 2 Port 1
The non negativity restriction is defined in equation

8 and ensures that all values for the decision variable
regarding the transported cargo are positive.
Figure 4. Depict of the selected routes.
The binary values restriction only allows binary values

to be assigned to the decision variable regarding the
vessels allocation to a specific route.
Sequence of ports in the selected routes.
Table 4.
Route 2 characteristics.
Voyage
Origin
Destiny
Cargo
(TEUs/trip)
Occupation
L_1
L_6
Port 1
Port 2
Port 2
Port 1
11.82
58.82
2%
9%
cost, which is the models figure of merit, and the

average frequency of attendance (calculated routes).
Among the twenty-eight candidate routes included in
the input data model, the optimal solution was reached
with the selection of route number 2 and number 23, for
ships K_1 and K_2 respectively. The selected routes
are formed by sequences of ports shown in Table 3.
Figure 4 illustrates that route r_2, selected for vessel K_1, considers trips visiting only ports 1 and 2,
while the route R_23, selected for vessel K_2, starts
the voyage from port 1, then visit port 3 and port 4,
ending the journey upwards. After port 4, the vessel
heads home (port 1), visiting all the intermediate ports
(port 2 and 3).
The trips represented by route r_2 provided the
following occupation factors presented in Table 4.
The nomenclature regarding Trip is the same as
presented in the chapter regarding modeling, therefore
L_1 and L_6 intervals whose origins were the ports
1 and 6. From the results shown regarding occupation, it is possible to conclude in advance that vessel
K_1 was assigned to cover only a marginal portion in
regards to vessel K_2. Certainly, from these data, it
can be concluded that there is spare capacity to meet
this demand.
The route R_23 presented the following voyage
characteristics as shown in Table 5.
Table 5 shows the operating characteristics of each
voyage within the journey travelled by vessel K_2 on
route 23. These data shows an excellent use of the vessel when travelling towards the north (L_1 and L_3)
and a regular performance for the remaining, except
L_4 interval due to an imbalance between the considered load flows. The vessels make the journey from
south to north loaded while the return trips are with
less occupation.
IMPLEMENTATION AND RESULTS
Since the model considers integer and real decision

variables (variables Xr,k and Mr,k,i,j respectively) a
mixed linear programming is considered. In order to
solve this problem a model was prepared using the
CPLEX solver, included in the GAMS programming
language package and commonly used in these type of
problems (Vananan 2004, Maturana 1995 or Hansen
2006). The model is structured to depend solely on
external input files, which define the viable routes,
costs of routes by type of vessel, number round trips
made by each unit on each route and, finally, the
demand for cargo between the ports involved in the
model. Thus, it is possible to vary certain conditions,
such as number of ports and therefore the number of
routes, quantity, speed, capacity and cost of shipping,
cargo search and demand, etc.
Consider now a fictional scenario with four ports
and two ships. This scenario has twenty-eight different
routes that the model will be in charge of scheduling to
each vessel. Some scenarios were run to validate and
verify the sensitivity and impact of changing some of
the input parameters. First, in order to verify the efficient utilization of the available fleet, scenarios were
run varying the amount of cargo that should be moved,
i.e., the demand will be modified according a parametric method. Then, scenarios were tested varying the
capacity of the vessels, setting a fixed value of demand,
with the aim of making the fleet more appropriate for
that specific search and demand condition.
For the considered initial conditions, the model has
reached the optimal solution presented in Table 2.
Table 2 presents the allocated routes for each vessel
and the resulting average occupation, plus the total
92
Table 5.
Route 23 characteristics.
Voyage
Origin
Destiny
Cargo
(TEUs/trip)
Occupation
L_1
L_3
L_4
L_5
L_6
Port 1
Port 3
Port 4
Port 3
Port 2
Port 3
Port 4
Port 3
Port 2
Port 1
99%
74%
37%
41%
67%
1, 2, 3, 4
2, 4, 7
8, 9, 10
5, 6, 8, 9
3, 4, 5, 8
Table 6.
Cargo flow distribution in route 23.
Voyage
Origin
Destiny
Occupation
1
2
3
4
5
6
7
8
9
10
Port 1
Port 1
Port 2
Port 2
Port 3
Port 3
Port 3
Port 4
Port 4
Port 4
Port 3
Port 4
Port 3
Port 4
Port 1
Port 2
Port 4
Port 1
Port 2
Port 3
29%
13%
16%
40%
3%
5%
20%
8%
25%
4%
Figure 5. Cost/Occupation evolution.
To better understand the above results, it is important to check the cargo composition in each voyage.
The last column of Table 5 lists the ports which contribute with cargo present in that particular voyage,
being this decomposition presented in table 6 below.
Table 6 shows an operating result of how the loading should be done for the vessel in each section. For
example, when docked at port 1, the shipment will consist of cargoes that go into the third port, filling 29%
of the capacity of the vessel and cargo that will go to
port 4, filling 13% of the capacity. A second example,
if the vessel is at port 4, it will be loaded with cargo to
port 1 (8%), port 2 (25%) and port 3(4%).
Thus, by considering the two previous tables with
outputs it can be assumed that the model helps in
building the vessels cargo layout in order to meet its
capacity and meet the necessary cargo demand level
minimization total operations cost.
Figure 6. Scaling/Number of vessels evolution.
a good and viable solution, which was seventy percent over current demand (1.7 * D). The results of this
analysis are available in Figures 5 and 6.
From these figures, it is possible to identify the
more attractive and convenient scenarios of operation
by analysing the relationships between the different
results. The total cost is increasing in line with the
increasing demand, however, it is noticed that for some
levels this value does not change, since for a given set
of assigned routes the required demand can be met by
fleets increasing occupation. Occupation was calculated as an average between the two vessels when both
are used, thus, these values are closely related to the
total capacity of vessels used. The scaling periodicity
is also very dependent on the number of ships used.
Figure 5 shows that for demand values between 20%
and 50%, the total cost remains almost the same, with
gradual increases in occupation of the fleet. For the
value of 60%, the cost remains similar, but fleet occupation has fallen slightly, equaling the level of 40%.
This phenomenon occurs due to change in the chosen
route. This means that, the selected route for the first
case considers a handling capacity equivalent to a maximum value of 50% of the initial demand. For greater
values, this same route no longer meets the constraint
regarding demand and therefore the model chooses
another route with similar costs, which changes the
sequence of visits to ports and thus the value of the
occupation as well.
DEMANDS PARAMETRIC VARIATION
Consider now some scenarios by varying the demand

for a fixed fleet in order assess what is the appropriate cargos handling level for the available fleet in
order to provide satisfactory results in terms of average occupation, total cost, ports attendance frequency
and number of ships assigned. The sizes of ships for
the creation of this set of scenarios are the same as the
model previously studied, 670 and 900 TEUs.
Scenarios were run varying the demand within ten
percent intervals, starting first from thirty percent of
the initial value (0.3 * D) until the threshold value for
93
Table 7. Tested fleet configurations.

500_500
500_600
500_700
500_800
500_900
500_1000
500_1200
600_600
600_700
600_800
600_900
600_1000
700_700
700_800
700_900
700_1000
800_1000
900_900
400_500
to meet demand up to 70% higher than that used in

the standard model. However it should be borne in
mind that a good use of the fleet or good occupation
levels implies a declining level of service in assisting
ports (increase in the scaling intervals). In the end, the
decision of interesting, viable and profitable operating ranges will depend on the conditions required and
desired by the owner.
Figure 7. Graphic Costs increase/Total fleet capacity.
In the same Figure 5 it is important to observe the

behaviour of the occupation curve between points corresponding to 0.9 and 1 of the original demand. There
is a steep drop explained by a sudden increase in transport capacity, i.e. from this demand up, you can only
meet the requirements with two vessels. The introduction of another unit in the system implies an increased
availability and the consequent drop in occupation and
increased cost. With the gradual increase in demand,
that space or availability will be filled until the system requires another vessel, which occurs at around
1.7 times the standard demand.
Figure 6 presents the same analysis in order to assess
the level of service associated with each scenario. The
introduction of a vessel in the system causes great
impacts in both the overall cost, occupation, but mainly
in the time between calls in ports. Naturally, an operation using only one vessel tends to provide a low cost,
a great occupation level, but a poor frequency of service. The graph illustrate that up to a level of 0.9 times
the demand, it is possible to operate the service with
only one vessel at a periodicity of scales of about 17
days. For greater values of demand, it is necessary to
introduce an additional vessel, which brings benefits in
service by reducing the interval down to approximately
4 to 8 days.
In these optimizations, the available fleet was composed by two vessels, one of 670 TEUs and another
of 900 TEUs. The model selected the most convenient
route for each value of demand and by observing the
results of the previous analysis it was noticed that the
characteristic of the fleet greatly influences the cost
of fleet utilization and service level. This aspect has
lead to the evaluation of the relative increase in total
cost (considering the previous data) compared with the
capacity used by the fleet (Figure 7).
It was found that up to the value corresponding to
0.9 times the demand, only one unit is used, but for
the values exceeding 0.6 times the demand, the vessel must be replaced by the one with higher capacity,
hence the cost increase at 0.6. Between 0.9 and 1.0 of
the original demand, another great increase is observed
due to the introduction of an additional vessel on the
fleet. The other most significant increases that occur
after the use of two vessels are due to modification of
routes assigned. In summary, it is concluded that the
resources available for this operation would be able
FLEET CAPACITY PARAMETRIC

VARIATION
For this case, it is possible to perform the analysis

of the fleet in order to define the best combinations
of classes of ships needed to meet the requirements of
the operation (similarly as exemplified by Boykin et al.
1985 for a chemical fleet operation). For a given level
of demand (assuming the default demand matrix), it
was verified the system behavior by varying the size
or capacity of vessels that could be attributed to the
operation of maritime transport.
In the analysis presented in the previous section it
was found that for the standard level of demand, the
fleet is over dimensioned. In order to reduce costs and
maintain the same level of service or even improve the
operation, other alternative fleet compositions were
studied to support the owner in selecting ships for the
operation. In practice, when searching for a vessel to a
new line of operation in maritime transport, the owner
must have information on this operation and especially
the desired vessels class. To assist in this process, the
model allows scenarios to be tested by changing the
distinct fleet characteristics such as, speeds, capacities
and costs. For this analysis the following fleet configurations presented in table 7 were considered and the
model executed in order to access the best route for
each pair of vessels. As such, the pair of numbers in
table 8 refers to the capacity in TEUs of vessels K_1
and K_2.
Similarly to the analysis presented in the previous
section, several tables were created referring to costs,
average fleet occupation, time between attendances,
number of assigned ships and routes assigned to each
of the mentioned alternatives. Results are shown in
Figures 8 and 9.
The graphs show that the fleet is very sensitive to the
values of the required demand. For the same amount of
capacity available, it is necessary to properly distribute
it between the two ships considering the cargo flow
94
Table 8.
Sequence of ports of each selected route.
R_12 0
Port 2 Port 3 Port 4 Port 3 Port 2 0
R_23 Port 1 0
Port 3 Port 4 Port 3 Port 2 Port 1
Figure 8. Fleet dimensioning total costs/Time between

scales.
Figure 10. Itineraries for the 400 and 500 TEUs vessels.
Therefore, the high value for the cost is due, among

others, to a high ports fees cost, because the numbers
of docking ports on the route represent a large share
in the overall cost of the route. On the other hand, the
high number of calls reduces significantly the intervals
between visits to ports and therefore the occupation of
the vessel.
It is noticed that for this case, a vessel of 1,000 TEU
class is fully capable of fulfilling the requirements of
demand at the lowest cost (leaving the smaller vessel
unused). In case of adopting the two units solution, the
settings tend to be more favorable to vessels of similar
sizes, up to 600 TEUs, always trying to balance the
flow of cargo.
Similarly to the analysis presented above, each presented scenario has its peculiarity with advantages and
disadvantages that should be considered as needed
according the context in which they wish to serve the
segment of short sea shipping. In short, it is possible to
verify the versatility of the model developed, allowing
analysis of various aspects and points of view.
Figure 9. Mean occupation/Number of vessels.
needed to meet demand. It is worth mentioning that

the model was designed so that if the voyages cargo
flow exceeds the capacity of the vessel, the assignment
is deemed unfeasible and, therefore, rejected. Thus,
when the transport requires two ships, one should seek
to maintain balanced the flow between the vessels.
From Figures 8 and 9, it can be observed that the
behaviour regarding the occupations and average costs
are similar to those existing in the previous section,
however a very significant difference in the number of vessels used. In this analysis it is possible to
make a deeper study regarding the utilization of the
fleet, for example take the last case presented using
a vessel of 400 TEUs and a second of 500 TEUs.
Although two small vessels are being used, the costs
are the highest, the average occupation is lower and
the intervals between visits are the best. Such factors
can be explained when analyzing the routes to which
the vessels were assigned, according to Table 8 and
Figure 10.
For this particular case, routes that make many stops
along the way were selected, i.e., the route R_12 visits
all ports except port 1 in both directions and also the
route R_23 visit all the ports in both directions, except
for port 2 on the ascending direction.
CONCLUSIONS
The developed mathematical model can assist the user

in strategic decisions procedures for maintaining a
given service level while using the resources in an
optimal manner.
The solutions and outputs created by the model have
both strategic and operational characteristics. Strategic
in a manner that it presents the required fleet dedicated to this service, establishing the routes needed to
meet demand. Operational in a manner that it presents
the composition of the cargo existing in the vessel for
each voyage of the chosen route, i.e., all loading conditions, including occupation by the vessel for each
voyage.
The various scenarios tested by the models show
evidences for consistent results reaching optimal
results for each configuration set. Because it is a
model involving binary variables any change in the
input parameters can produce quite different results,
95
allowing the development of a marginal analysis of

the results. The study allows a thorough evaluation
of the strategies employed and it offers alternatives
for increasing services, or provides the basis for new
business development in this segment of maritime
transport.
The methodology adopted for the development of
this model allowed drawing up a very simple, clean
and efficient model, compared with the complexity of
the problem. The feeding of information on the model
by means of aggregated data significantly reduced the
degrees of freedom and therefore reduced greatly the
number of model variables.
Several improvements may be considered when
addressing future recommendations, such as considering variable costs, demand variation trough the
planning time horizon, profits or revenues maximization, different vessels performance according each
voyage.
Boykin, R.F. & Levary, R.R. 1985. An interactive decision

support system for analyzing vessel voyage alternatives,
Interfaces 15: 8184.
Brown, G.G., Graves, G.W. & Ronen, D. 1987. Scheduling
ocean transportation of crude oil, Management Science
33: 335346.
Fagerholt, K. 2000. Evaluating the trade-off between the level
of customer service and transportation costs in a ship
scheduling problems, Maritime Policy Management 27:
145153.
Hansen P., Mladenovic N. & Urosovic D. 2006. Variable
neighborhood search and local branching, Computers &
Operations Research 33: 30343045.
Maturana D. & Eterovic Y. 1995. Vehicle routing and production decision support systems: Designing graphical
user interfaces, International Transactions in Operational
Research 2:233247.
Perakis, A.N. & Jaramillo, D.I. 1991. Fleet deployment optimization for liner shipping Part. 1 Background, problema
formulation and solution approaches, Maritime Policy
Management 18, 183200.
Sherali, H. D., Al-Yakoob, S. M. & Asan, M. M. 1999.
Fleet management models and algorithms for an oil-tanker
routing and scheduling problem, IIE Transactions 31:
395406.
Vananan M. Wang Q., Batta R. & Szczerba R.J. 2004. Integration of COTS software products Arena & CPLEX for
an inventory/logistics problem, Computers & Operations
Research 31: 533547.
REFERENCES
Bausch, D.O., Brown, G.G. & Ronen, D. 1991. Elastic set
partitioning A powerful tool for scheduling transportation of oil and gas, Advances in Operations Research in
the Oil and Gas Industry e.technip: 151162, 1991.
96

Design development of an oil and garbage collector and boom layer vessel
J.A. Valente
EUROSHIDE, Lisboa, Portugal
Pedro Antunes & Luis Batista

VeraNavis, Lisboa, Portugal
S. Ribeiro e Silva
ABSTRACT: Based on shipowner requirements conceptual and preliminary designs of an Oil and Garbage
Collector and Boom Layer Vessel (OGC&BLV) were initially carried out by a ship design office aiming to
prepare a construction proposal to be presented by a Portuguese shipyard to a Russian company subcontracted
by a local port authority. Unfortunately due to changes and different orientations in the Russian company the
project was stopped and the two OGC&BLV were never built. However, at a later stage a more detailed design of
the main machinery and some of the auxiliary systems has been conducted at university, using base information
supplied by the ship design office and technical data available from prospective suppliers of major equipments
to be installed onboard, including the main engines, generator set, oil skimming booms, recovery brush pack,
hydraulic actuators, and oil containment booms. This paper presents the final configuration of the OGC&BLV
and describes the most interesting features and technical design challenges faced during design development of
this specific and multifaceted type of vessels.
1
INTRODUCTION
the tender. Unfortunately due to changes and different orientations in the Russian company, the project
was stopped and the vessels were never built. More
recently, development of the detailed designs of the
main machinery and auxiliary systems has been conducted at Technical University of Lisbon for academic
purposes as part of Machinery and Marine Systems
course of the MSc in Naval Architecture and Marine
Engineering.
The base information collected before by the
ship design office and technical data available
from prospective suppliers of major equipments to
be installed onboard OGC&BLV was extensively
utilized.
Estaleiros Navais de Peniche (ENP) participated in an

international tender, to design and build two vessels
for collecting oil and garbage in harbors and coastal
waters. The tender was promoted by a Russian company subcontracted by local port authority and had two
main stages:
a) The first stage consisted of a technical proposal,
without any reference to costs involved. This first
stage was to be used to select a restricted number of participant shipyards that would pass to the
following stage;
b) The second stage consisted on presentation of an
updated technical construction specification and
the respective commercial tender.
2 THE SHIPOWNER REQUIREMENTS
ENP commissioned a ship design office, designated as

EUROSHIDE for designing the vessels. EUROSHIDE
is a company whose main activity is in the area of ship
design in the field of Naval Architecture and Marine
Engineering, another partner designated VeraNavis
was found to collaborate in this project. To elaborate the project, the ship design office also divided
the design in two stages to enable the shipyard to
deliver the documents the Russian company required
in each stage.
ENP entered the first stage of the tender, passed to
the second stage and finally won the second stage of
The ship service would be to collect oil and garbage

from the sea surface, so that oil skimming booms
would be permanently transported onboard and then
installed on both sides of the vessel at polluted areas.
In addition, the vessel would be fitted as well with
oil containment booms 200 meters long to contain oil
spills. The ship service area would be port areas and
coastal waters of the Baltic Sea.
The vessel would be of the closed steel displacement
type. The hull and decks would be general transversally stiffened and longitudinal girders would assure
97
Table 1.
Initial main characteristics of the OGC&BLV.
Table 2.
Final main characteristics of the OGC&BLV.
Parameter
Value
Unit
Parameter
Length overall
Beam
Draught (design)
Depth
Autonomy
Endurance
Storage of collected oil products
Cargo hold capacity
20
56
1.01.5
2.50
not less than 350
not less than 23
not less than 30
not less than 15
m
m
m
m
nm
day
m3
m3
Length overall
Length between perpendiculars
Beam
Depth
Draught (design)
Draught (scantling)
Storage of collected oil
products
Cargo hold capacity
Deadweight (at design draught)
Endurance
MCR (Propellers)
MCR (Pump jet)
Speed (at design draught,
propellers)
Speed (at design draught,
pump jet)
continuity of the structure. Main dimensions and operational characteristics were initially established by the
shipowner as follows.
2.1
Main characteristics
Value
Unit
21.5
20.1
6.0
2.6
1.5
1.8
30
15
50
3
2 260 kW at 1800
2 310 kW at 2200
9.010.0
m
m
m
m
m
m
m3
m3
ton
day
rpm
rpm
kn
8.09.0 kn
Pre-defined main characteristics of the OGC&BLV

are summarized in Table 1.
Table 3.
2.2
Speed
During sea trials, the ship at design draught and clean

hull, in deep and smooth waters and wind speed less
than 4 m/s (Beaufort 3), would need to reach the
maximum speed of 810 knots.
2.3
Intact and damaged stability
Intact and damaged stability characteristics would

need to be fully compliant with all the rules defined
by the Russian Maritime Register of Shipping for the
applicable service type and operational area of the
vessel.
2.4
Type
Value
Unit
Crew & effects

Provision stores
Ship stores and spares
Diesel oil
Fresh water
Dispersant
0.6
0.5
0.9
9.0
2.5
1.5
ton
ton
ton
ton
ton
m3
2.7 Special equipments

The vessel would be equipped with an oil and garbage
recovery system expected to reach in practice an oil
recovery capability in the range 80 m3 /h.
Oil containment boom reel would be installed on the
aft of the main deck of the vessel in to store 200 meters
of the rigid containment booms transported onboard.
Seakeeping
Seakeeping quality of the vessel operating near

the Baltic coast of Russia was initially defined by
the shipowner as sailing without any restriction up
to Beaufort 4 and sailing with restrictions up to
Beaufort 5.
DESCRIPTION OF THE VESSELS

DESIGN SOLUTIONS
3.1 Main characteristics
2.5 Accommodation
After analyzing the shipowner requirements, and after

making some design iterations, it was adopted an
arrangement with superstructure forward for allowing an ample working deck aft and a long forecastle
for enhanced seakeeping characteristics. In addition,
accommodations should be located in the forecastle
and machinery spaces in the aft part. Cargo tanks,
holds and pollution collecting equipment should be
all located as near as possible to amidships in order to
minimize their influence on the vessel trim, resulting
on the main characteristics shown in Table 2.
Onboard the vessel, the following accommodations

were foreseen:
Crew
2 men;
Special staff 24 men.
2.6
Consumables and stores of the OGC&BLV.
Propulsion and auxiliary systems
As main engines, two diesel were foreseen driving the

two propellers. Alternatively, two water-jet type units
could be used.
A diesel generator was foreseen for producing the
electric power.
The main engine command and control was defined
to be carried out from the wheelhouse.
3.2 Deadweight and capacities

As shown in Table 3, consumables and stores for three
days endurance have been determined.
98
Figure 2. Body plan of the OGC&BLV.
Figure 1. General arrangement of the OGC&BLV.
3.3
Crew and accommodations
In accordance with the general arrangement shown in

Figure 1, the accommodation spaces distribution will
be as follows:
One cabin with 2 berth for 2 crew;

One cabin with 4 berth for 4 special staff;
One Combined galley, mess and daily room;
One WC with shower;
One wheelhouse.
3.4 The influence of hull form selection on

resistance & propulsion and construction
Figure 3. Power curve of the free-runnning OGC&BLV.
The ship with a length between perpendiculars of

20.1
meters and a top speed of 10 knots has a ratio
V/ L = 2.2, which is relatively high, and, therefore,
special attention to the hull form was devoted from
the start. As illustrated in Figure 2, the hull form has a
low block coefficient with the LCB aft of amidships.
Figure 3, shows the power curves for propeller and
pump-jet propulsion, estimated using Holtrop method
corrected with correlations to similar size existing
vessels.
The curves shown in Figure 3 are obtained assuming a propulsive efficiency of 0.50 for propeller
propulsion and 0.40 for pump jet propulsion.
From Figure 3, it can be seen that speed at 85%
MCR will be 9.53 kn for both types of propulsion.
The basic operating requirement for a main propulsion system is to propel the vessel at the required
sustained sea speed for the required endurance. This

propulsive action is obtained from the torque generated at the prime mover (main engine), which is then
converted into low-speed drive for the propeller shaftline by means of a gearbox, and finally reaches the
propeller itself. In complement, the propulsion system
should also provide the vessel stopping, backing, and
maneuvering capabilities. In some special-purpose
vessels such as this one, there is more than one single speed requirement, and, therefore, the power plant
for the OGC&BLV must be designed according to
two basic and distinct operational conditions at sea:
free-sailing and conducting an oil recovering operation. Hence, the OGC&BLV power requirements are
clearly established by an extreme free-running operation, but consideration to the boom towing mode of
99
Figure 5. Midship sections of the OGC&BLV.
Figure 4. Effective power curve of the OGC&BLV during

oil recovering.
operation must also be given. Figure 4, shows the

power requirements in the oil recovery condition.
In the oil recovering condition, the curve was
obtained by the same method as free-sailing. However, total resistance was obtained by adding the drag
resistance of the boom to the hull resistance on freesailing condition. The resistance curve of the boom
(Rb ) is simply given by: Rb = 1/2 U2 Ab CD ,
where it has been assumed that CD = 2.0 (bluff body)
and Ab = 10 m2 . As shown in Figures 3 and 4, the
boom resistance is the largest component of the total
resistance during oil recovering condition, about 95%
of total. At the maximum speed (4 knots), the estimated effective power required during oil recovering
condition is about 100 kW.
Notice should be also given to the fact that in oil
recovery condition it should be made possible to couple a Power-Take-Off (PTO) to one of these gearboxes,
where the total required power of the PTO to feed
all the hydraulic systems is estimated as 82.3 kW:
Skimming Brush Pack 10 kW, Boom Drum 3 kW,
Deck Crane 20 kW, Channel Watertight Doors 2.8 kW,
Archimedes Oil Pump 28 kW, Dispersant Tank Pump
8 kW, and Anchor Winch 10.5 kW.
The propeller design was performed with the
following assumptions:
Brake power according to Figure 3;

B-Series;
4 Blades;
Fixed Pitch Propeller (FPP);
Free-sailing condition.
Therefore, main characteristics of the propeller,

optimized for the selected engine and gearbox, are as
follows:
dimensions obtained are in good agreement with

the recommended hull bottom and rudder clearance
margins.
Note that, according to shipowner OGC&BLV
could have two distinct types of propulsors:
2 FPF;
2 Schottel Pump Jets.
However, despite Schottel Pump Jets can provide enhanced maneuvering capabilities in confined
waters, the propulsion with FPF has larger efficiency
(about 20% higher) and a lower cost. Moreover, concerns about working conditions of Schottel Pump Jets
inside an oil spill have been raised.
Lastly in terms of cost optimization, as illustrated
in Figure 2, the vessel has an hydroconic hull with
4 chines on each side, for an easier and cheaper
construction process.
3.5 The ship structure
The ship hull and superstructure are manufactured
from a mild steel grade A plating and framing. Transverse framing has been adopted with a frame spacing
of 500 mm.
The largest peculiarity of the OGC&BLV structure
is to have two watertight side channels, which may
also be opened to the sea through watertight doors.
As it can be observed in the amidships sections shown
in Figure 5, specific structural design solutions are
required in this area, and this peculiar aspect of the
OGC&BLV will have some significant impact as well
on ships stability during oil and garbage recovering
operations.
A pair of holland bar shape bilge keels, 0.3 m
wide with its trailing edge located 10 m in front of the
aft perpendicular and 0.5 m above the baseline should
be welded at the bilges. The length of the bilge keels
is approximately 5 m.
3.6 Ship stability
Diameter: 880 mm;

BAR: 0.55
Pitch: 880 mm;
QPC: 51%;
Prop RPM: 745;
Eng RPM: 1800
Notice should be given to the fact that in order

to prevent induced vibrations problems propeller
First, the OGC&BLV must have adequate trim and

draughts while floating, and in addition should also
comply with the stability criteria in all service conditions in which the ship may find itself during operation
and life, even when damaged or mishandled. Considering that the OGC&BLV has two basic and distinct
operational conditions at sea: sailing and conducting
100
Table 4. Typical Load Conditions of the OGC&BLV.
1
2
3
4
5
6
7
8
9
10
11
12
13
Loading Condition
Disp.
Draft
Trim
Ballast
Observations
Departure no cargo closed

Departure no cargo open
Departure no cargo open
Departure with cargo closed
Departure with cargo open
Departure with cargo open
Arrival no cargo closed
Arrival no cargo closed
Arrival no cargo open
Arrival with cargo closed
Arrival with cargo closed
Arrival with cargo open
Arrival with cargo open
83.55
83.55
85.65
120.6
120.6
123.6
74.59
76.49
74.89
111.9
112.6
111.9
113.6
1.13
1.24
1.27
1.43
1,53
1.57
1.05
1.06
1.15
1.41
1.41
1.54
1.57
0.02
0.45
0.00
0.12
0.62
0.00
0.42
0.00
0.12
0.12
0.00
0.35
0.00
no
no
53% FP
no
no
92% FP
no
52% AP
no
no
24% AP
no
51% AP
Departure no cargo
with & without
Ballast
Departure with cargo
with & without
Ballast
Arrival no cargo
with & without
Ballast
Arrival with cargo
with & without
Ballast
Table 5. Dynamic stability and stability criteria of the

OGC&BLV.
Load Disp.
Cond. (ton)
GMt Area
Area
RA
Angle
(m) 0 /30 30 /40 at 30 Max RA
1
2
3
4
5
6
7
8
9
10
11
12
13
1.45
0.91
0.36
1.05
0.68
0.6
1.36
1.21
0.73
1.00
0.94
0.63
0.63
83.55
83.55
85.65
120.55
120.55
123.62
74.59
76.49
74.89
111.89
112.63
111.89
113.58
0.176
0.103
0.101
0.149
0.033
0.039
0.158
0.152
0.075
0.141
0.137
0.072
0.073
0.099
0.070
0.071
0.099
0.069
0.072
0.031
0.031
0.050
0.090
0.039
0.059
0.061
0.555
0.365
0.364
0.531
0.350
0.360
0.476
0.463
0.266
0.493
0.454
0.297
0.302
33.1
50.0
50.0
43.4
46.3
47.2
31.9
33.2
50.0
42.4
42.4
46.4
46.5
an oil recovering operation, where both recovery side

channels can be either closed or totally open to the
sea (flooded), then it was clearly needed to distinguish these two hydrostatic equilibrium situations in
the stability study. Apart from these two operational
scenarios, other loading conditions could be devised
between limiting arrival and departure conditions, and
with or without cargo (oil or garbage) embarked.
Table 4 shows a summary of all the typical loading
conditions, where the term openedand closedrefers
to the collecting channels being flooded or intact,
respectively. As shown in Table 4, in some loading
conditions small amounts of ballast were required in
the peak tanks in order to have adequate trim.
Table 5 clearly shows the negative effect in the
dynamic stability of the side opening of the collecting
channels.
3.7
Seakeeping
Seakeeping criteria: Defining acceptable limits of

operation, then the seakeeping quality of a ship operating in an ocean area and with a specific mission can be
quantified. In the present work the seakeeping quality has been assessed by an operability analysis that
accounts for:
The ship mission, through the use of seakeeping criteria, which represent acceptable limits of
operation.
The hydrodynamic and inertia characteristics of the
ship, through the use of a seakeeping program to
calculate the motions transfer functions.
The wave climate where the ship operates, through
the use of the probability distribution of short
term seastates together with spectral techniques to
calculate the ship responses in those seastates.
To define the seakeeping criteria it is necessary to

account for two aspects: the ability of the ship to physically carry out its mission (in spite of the adverse
seastate), and the working conditions onboard that
must ensure that some comfort standards are met.
The seakeeping criteria depend on the type of ship
and mission. Several seakeeping criteria related with
the absolute motions, relative motions, accelerations,
slamming, green water on deck, etc., have been proposed and are available in the literature. Apparently
there are no criteria specific for OGC&BLV, and for
this reason the authors have used a set of criteria that in
fact are based on technical intuition for these types of
ships, or have been adapted from other type of vessels.
For instance, Odabasi et al. (1991), Sarioz and Narli
(2005) and Ribeiro e Silva et al. (2006) prescribed
a similar value of 0.2 g for the vertical acceleration
criterion, at the bridge and working deck locations for
fishing vessels and navy patrol boats, to the one that
has been adopted in this study.
Table 6 presents the criteria selected and the positions in the ship where the related responses are
calculated. The positions are represented in a reference system with origin on the baseline in the vertical
of the AP, the x-axis points to the bow, the y-axis points
to port and the z-axis points upwards
The following ship responses are used to define
criteria: roll, pitch, vertical and lateral accelerations
at the Bridge and at the Working Deck (in the table g is
101
the acceleration of gravity). All the criteria are defined

in terms of the root mean square of the response (rms).
Ship responses in waves: The first step in the calculation method is to obtain the transfer functions of
the rigid body motions and also of the relative motion
and accelerations in selected points in the ship. The
transfer functions represent linear ship responses to
harmonic waves. This investigation uses a seakeeping code based on the strip theory formulation from
Salvesen et al. (1970) to calculate the potential flow
hydrodynamic coefficients and harmonic wave exciting forces.The equations of motion are solved to obtain
the five degrees of freedom oscillatory motions (strip
methods neglect all surge forces). Since the seakeeping
criteria are defined also in terms of accelerations felt
onboard, these quantities are calculated for all relevant
points on the ship.
To determine the response of the ship to real
(irregular) seastates a spectral formulation is adopted,
following the early proposals of St Dennis and Pierson
(1953). The Pierson-Moskowitz spectral form for
fully developed seas (Pierson and Moskowitz, (1964))
describes the irregular seastates in terms of significant wave height and peak period. The resulting ship
responses in irregular seas are represented by the
response spectra and the related standard deviation.
Table 6.
Seakeeping criteria for the OGC&BLV.
Response
Location (x,y,z)
Criterion
Roll (rms)
Pitch (rms)
Vert. accel. at Bridge
Lat. accel. at Bridge
Vert. accel. at Work. Deck
NA
NA
(16.5,0.0,4.65)
(16.5,0.0,4.65)
(0.0,3.0,3.25)
6
3
0.2 g (rms)
0.1 g (rms)
0.2 g (rms)
The seakeeping criteria are defined by a maximum

allowed standard deviation of the response. Comparing the ship responses in irregular waves with the
seakeeping criteria, it is possible to check if the significant wave height for which the ship will be operating
satisfies the criteria. From a scatter diagram, which
represents the wave climatology for a given ocean
area (Hogben (1986) and Longuet-Higgins (1952)),
the design point sea state can be easily defined.
Seakeeping performance results: Calculations were
done using annual wave statistics, in order to assess
seakeeping performance at the specific design point.
The OGC&BLV will operate at the Russian coast of the
Baltic Sea, which corresponds to a peak wave period
of 10.45 seconds and a significant wave height of 1.5
meters for Area nr. 31 of European Wave Statistics
database. Two different conditions were considered,
namely zero speed and the maximum speed of 8 knots.
Tables 7 and 8 present the results of the ship seakeeping performance at the design point. These results
allow identification of which headings and speeds
that the ship can sail fully operational, meaning that
the ship satisfies the seakeeping criteria of Table 6.
Different rows show the results for each heading of
the vessel with respect to the design point wave spectrum. The last row presents the average value, which
represents the global seakeeping performance.
The results in Tables 7 and 8 show that the operability of the OGC&BLV is reasonable for zero speed and
is improved at 8 knots due to viscous effects on roll
damping. Analyzing different headings one observes
that the worst conditions occur in headings around
beam waves (120 , 90 , 60 ) and this is related with
the roll motion.
Observing Tables 7 and 8 for specific headings, it is
clear that the roll criterion and the lateral acceleration
Table 7. Seakeeping performance of the OGC&BLV at the design point, sailing zero advance speed with (top) and without
(bottom) bilge keels.
Sea direction
Angle
Criterion
Nautical Term
Roll
Pitch
VA 1
LA 1
VA 2
LA 2
Speed 0 Kts without Bilge Keels

180
Head Seas
Bow quartering Seas
150
120
Fore quartering Seas
90
Beam Seas
60
Aft quartering Seas
Stern quartering Seas

30
0
Following Seas
Average
0.00
4.30
8.49
10.31
8.43
4.08
0.00
5.09
2.96
2.67
1.74
0.08
1.58
2.44
2.71
2.03
0.17
0.19
0.23
0.17
0.18
0.14
0.12
0.17
0.00
0.26
0.55
0.65
0.45
0.20
0.00
0.30
0.17
0.20
0.29
0.34
0.29
0.14
0.14
0.22
0.00
0.12
0.27
0.35
0.31
0.13
0.00
0.17
Speed 0 Kts with Bilge Keels

180
Head Seas
150
Bow quartering Seas
120
90
Beam Seas
60
Aft quartering Seas
30
0
Following Seas
Average
0.00
3.03
5.67
6.76
5.64
2.96
0.00
3.44
2.96
2.67
1.74
0.08
1.58
2.44
2.71
2.03
0.17
0.19
0.23
0.17
0.18
0.14
0.12
0.17
0.00
0.16
0.33
0.34
0.18
0.08
0.00
0.16
0.17
0.17
0.18
0.12
0.13
0.11
0.14
0.14
0.00
0.06
0.13
0.19
0.20
0.09
0.00
0.10
102
at the Bridge and Working deck failed for the unappended hull while for the hull appended with bilge
keels the criteria are practically satisfied in all the wave
encountering conditions. Another question arises if the
criteria selected for roll motion and lateral acceleration
are adequate for OGC&BLV, or if larger or smaller values should be adopted. A more in deep investigation
of the ship activity under well determined practical
conditions (e.g. evaluation of personnel performance
at the Working Deck when anti-slip deck covering is
applied) is necessary to derive appropriate seakeeping criteria. As a general conclusion, the seakeeping
performance of the OGC&BLV design is seen as
reasonable, however, it can be improved by reducing the roll motion amplitudes, and, as shown in here,
the simplest possibility that deserves to be considered
is the use of a pair of bilge keels to reduce roll motion.
3.8
Main machinery and auxiliary systems
This section describes first the propulsion plant design

of the OGC&BLV, providing not much great detail due
to lack of space. Figure 6, illustrates the shaft-lines
made up of a 4 blade bronze propeller, a stainless steel
AISI316 propeller shaft with 80 mm diameter with
taper cone 1:10, a flange C45 steel shaft coupling, a
bronze stuffing box bolted to the chrome stainless steel
stern tube, two rubber cutlass shaft bearings in which
one is outboard supported by a base strut bracket.
In terms of auxiliary systems, preliminary estimates of the system flows, pressures, and temperatures
were first made to support component selection. Then
approximate locations of major components and piping runs relative to the ship arrangement, and to each
other, were determined during the initial design phase
so that piping-system diagrams could be produced for
Table 8.
keels).
Figure 6. Propulsion plant arrangement, including twin

shaft-line layout of the OGC&BLV.
Figure 7. Machinery space arrangement, including piping

layout of the engine lubricating-oil service and diesel engine
fuel service systems of the OGC&BLV.
Seakeeping performance of the OGC&BLV sailing 8 knots advance speed with (top) and without (bottom) bilge
Sea direction
Angle
each auxiliary system. The machinery space encloses

part of the main machinery system and several other
auxiliary systems.
Moreover, the development of piping systems
within a machinery space and the machinery space
arrangement are inseparable because the locations of
many machinery components are influenced by piping
system considerations, and piping must be arranged
with knowledge of the arrangement restrictions that
Criterion
Nautical Term
Roll
Pitch
VA 1
LA 1
VA 2
LA 2
Speed 8 Kts without Bilge Keels

180
Head Seas
Bow quartering Seas
150
120
90
Beam Seas
Aft quartering Seas
60
30
0
Following Seas
Average
0.00
2.82
4.28
4.28
3.31
1.79
0.00
2.35
3.37
3.02
1.93
0.11
1.57
2.23
2.41
2.09
1.13
1.09
0.76
0.16
0.03
0.01
0.12
0.47
0.00
0.27
0.36
0.13
0.02
0.01
0.00
0.11
0.82
0.91
0.53
0.13
0.02
0.01
0.01
0.35
0.00
0.21
0.26
0.06
0.03
0.01
0.00
0.08
Speed 8 Kts with Bilge Keels

180
Head Seas
150
Bow quartering Seas

120
90
Beam Seas
60
Aft quartering Seas
30
0
Following Seas
Average
0.00
2.44
3.88
4.02
3.20
1.75
0.00
2.18
3.37
3.02
1.93
0.11
1.57
2.23
2.41
2.09
1.13
1.09
0.76
0.16
0.03
0.01
0.12
0.47
0.00
0.26
0.35
0.13
0.02
0.01
0.00
0.11
0.82
0.91
0.53
0.13
0.02
0.01
0.01
0.35
0.00
0.21
0.26
0.06
0.03
0.01
0.00
0.08
103
govern other components in the space. As an example,

Figure 8 shows the distillate fuel fill and transfer diesel
and the engine lubricating-oil service systems. The two
diesel engines are fed with distillate fuel from the daily
service tanks by gravity while the lubricating-oil service circuits has a motor-driven pump to circulate oil
through filters and coolers to the engine. These two
systems are quite simple, since oil sumps are integral with the two high-speed diesel engines selected
and the fuel fill and transfer circuits simply receive
and store fuel, deliver it to service tanks, transfer fuel
between SB and PB storage tanks, and, if necessary,
offload fuel.
4
4.1
OIL AND GARBAGE CONTAINMENT AND

RECOVERY SYSTEM
General description
The Oil and Garbage Recovery System (OGRS) recovers oil and garbage at speeds up to 4 knots. Because of
the high recovery speed the system is able to recover
the largest possible area of an expanding oil slick at
sea. Since it is built into the side of the vessel readyto-go all the time, the system can start skimming oil
within minutes of deployment at the spill site and can
be operated by a small crew. The simple and intuitive
operation of the OGRS requires very little specialized
training. When the system is not in use, it is stored
within the vessel hull and is otherwise independent of
the other functions of the vessel.
The OGRS channel mounted system collects oil of
all types and is unhindered by floating debris or seaweed. Adverse weather conditions and choppy seas
are common when combating oil spills, but the brush
pack can operate in these conditions without losing
performance. Another major advantage of the system
is that the recovered oil contains very little water, less
than 10%, making maximum use of valuable storage
volume.
Description of operation:As shown in Figure 8, with
the OGRS deployed, the entire vessel becomes an oil
slick processing system. The system uses the vessels
forward motion to deflect surface water and oil from
the collection area formed by the jib arm and deflection
boom into the recovery channel for processing.
An effective flow through the oil recovery channels is achieved by the forward speed of the ship,
which induces a pressure difference between the intake
and exit doors. The doors are located so that pressure fluctuations caused by boat motions and waves
are minimized. The flow of water carries oil into the
recovery channel where the oil is efficiently separated
by the brush pack and removed from the flow. The
brush system lifts oil, seaweed and debris from the
water which passes through the channel and returns
to the collection area for further processing. As illustrated in Figure 9, recovered oil and debris ride up the
conveyor, are combed from the bristles and flow by
gravity into holding tanks or into a collection hopper.
From this hopper the recovered material is pumped to
the tanks by a powerful transfer pump.
Figure 8. Description of operation of the OGRS.
Figure 9. The oil recovery process: 1 Brush Conveyor;

2 Brush Cleaner Unit, and 3 Collection hopper/delivery
trough.
4.2 Equipment description

Oil recovery channels: The oil recovery channel is
constructed on both sides of the vessel. The channels
are fitted with guiderails where the OGRS brush packs
are mounted through an opening on the deck fitted with
a hatch cover. Figure 10 provides broad dimensions
and construction of the recovery channel.
Brush packs: For separating oil and debris the
recovery channel is fitted with a standard OGRS
brush pack. The number of brush chains is selected
according to the desired recovery rate and the size of
the oil recovery vessel. As shown in Figure 11, the
brush chains lift the separated material to the specially
designed brush cleaner units which clean the bristles
by a patented combing action.
Hydraulic system: The upper shaft of the brush pack
is driven by a hydraulic motor. The speed of the motor
can be adjusted according to the recovery conditions
in order to minimize the water content. The speed is
104
Figure 13. The side doors.
Figure 10. The oil recovery channels.
Figure 14. Positive displacement Archimedes screw pump.
Figure 11. The brush packs. Clean (left) and utilized (right)
brush pack.
Figure 12. Hydraulic system control panel.
adjusted by a regulator valve installed in the control

panel situated on deck (see Figure 12).
The side doors of the recovery channels are also
operated from the control panel.
Hydraulic power is provided through flexible hoses
from the hydraulic system of the vessel. Hydraulic
hoses and flow control valves should also be provided
by the manufacturer.
Side doors: At each end of the two oil recovery
channels a hydraulically operated side door is fitted.
A crane operated deck cover will be installed over each
channel. The four hydraulically operated side doors are
fitted with oil resistant rubber seals. The side doors

can be closed and locked watertight so that when not
in use, the recovery channels can be emptied of water.
Figure 13, gives a general idea of the dimensions and
construction of the side doors.
Stripping pumps: Each channel is fitted with a
hydraulically driven submersible centrifugal pump for
emptying water. Water is let into the sea or into the
recovered oil tank via a detachable flexible hose. The
stripping pump is placed in a well at the bottom of
both channels.
Oil transfer pump: The cleaners of the brush packs
drop the recovered material into a receiving tank.
A submersible transfer pump is installed on the deep
well to transfer the recovered material outboard. The
receiving tank will be inclined to allow for gravitational flow to the deep well. As shown in Figure 14,
the pump is a hydraulically driven, positive displacement Archimedes screw pump with end suction and
cutting knives. The pump speed can be adjusted freely
between 0100% from the control panel on the deck.
Oil skimming boom: On both sides of the vessel, an
oil skimming boom is fitted. The height of the boom
decreases from the vessel end to the outer end, so that
close to the side the height is sufficient to allow for
boat rolling and heave, while the outer end the boom
height is lower.
As shown in Figure 15, the inner end of the boom
is fastened to the aft port by an aluminum section. The
outer end is similarly fastened in the buoy at the end
of the jib.
The booms are fitted with necessary strengthening
parts, weights and supporting lines.
The skimming booms will be inflatable and made
of nitrile rubber material.
105
Oil containment booms and drum: The vessel will

be fitted with oil containment boom 200 meters long,
NOFI 800S or similar. According to general arrangement, an hydraulic driven drum NOREN NORMAR
8 m3 Oil Recovery Boomreel will be installed on the aft
of the main deck to store 200 meters of rigid containment booms. Hydraulic power to the drum is provided
through flexible hoses from the hydraulic system of
the vessel.
solution presented for an Oil and Garbage Collector

and Boom Layer Vessel is particularly efficient to
combat a moderate spread of oil in harbors and coastal
waters as well as to remove the increasingly abundant
garbage in water surfaces of ports, mainly because
of its readiness and small crew requirements. Having accommodations for 6 men and an endurance of
3 days, allows the vessel to operate without interruption in a harbor, or to attend an oil spillage in certain
areas at some distance from home port.
Besides collecting oil products from the water surface, the vessel is also able to deploy rigid oil containment booms. For this purpose the vessel is equipped
with a hydraulic operated drum to store 200 meters
of this type of rigid booms. Besides this additional
feature, the vessel is quite multifaceted, because the
specific equipments for collecting products are stored
inside the hull, not encumbering the ship deck which
may be used for other harbor services. Moreover, the
deck crane may also be an useful tool for several other
services. The indicative vessel cost is about 2 million
Euros, where the specific oil and garbage recovery
system costs about 30% of this total value. Therefore,
it may be concluded that the OGC&BLV is a fairly
useful and efficient multipurpose vessel for ports and
harbors service worldwide.
4.3
REFERENCES
Figure 15. Arrangement of oil skimming boom during

recovery.
System performance in waves
The announced mechanical performance of the OGRS

is as follows:
Significant wave height = 1.5 m to 2 m (dependent
on vessel seakeeping characteristics);
Recovery speed = up to 4 knots;
Sweeping width = 25 meters (related to the desired
recovery speed, ship size, and recovery rate
required).
Recovered area = With the skimming speed up to 4
knots (even more in good weather conditions) the
size of the skimming system is optimized to clean
up all the oil that is possible to encounter, where
characteristics and speed of the vessel, skimming
width and tank capacity onboard are taken into
consideration.
Recovery capacity = For an advancing skimming
system like this, the theoretical oil recovery
capability is in the range 210350 [m3 /hr].
Recovered water = <10% of total recovered
volume.
Oil to be recovered = All grades and ages, including
debris, seaweed and tar balls.
Min air temperature = 5 C.
Min water temperature = 0 C.
5
CONCLUSIONS
Taking into account the nautical capabilities described

before in terms of crew comfort, structural robustness,
mobility, floatability, stability, seakeeping and operational versatility, it can be concluded that the design
Hogben, N., da Cunha, L.F. and Olliver, H.N., 1986. Global

Wave Statistics, Brown Union London.
Ikeda, Y., Himeno, Y. and Tanaka, N., 1978. A prediction method for ship roll damping. University of Osaka
Prefecture, Re-port No 405.
LAMOR Oil Recovery System (LORS), General Technical Specification, 2009 (electronic brochure available at
www.lamor.com).
Longuet-Higgins, M.S., 1952. On the statistical distribution
of the heights of sea waves. Journal of Marine Research,
Vol. 11, No. 3, pp. 245260.
Odabasi, A.Y., Fitzsimmons, P.A., Ankudinov, V.K. Wiley,
S.A., 1991. Seakeeping Considerations in Ship Design and
their Incorporation in HDDS, BMT. International, Report
no. HDDS.P2SPEC.
Pierson, W.J., Moskowitz, L.A., 1964. A proposed spectral
form for fully developed wind seas based on the similarity theory of S.A. Kitaigorodskii. Journal of Geophysical
Research 69, 51815190.
Ribeiro e Silva, S., Fonseca, N., Guedes Soares, C., 2006.
Performance of a navy ship roll stabilisation system.
In: Proceedings STAB2006, Rio de Janeiro, Brazil,
pp. 669676.
Salvesen, N., Tuck, E.O., Faltinsen, 1970. Ship motion and
sea loads. Transactions of the Society Naval Architects
Marine Engineers (SNAME) 78, 250287.
Sarioz, K., Narli, E., 2005. Effect of criteria on seakeeping performance assessment. Ocean Engineering 32,
pp. 11611173.
St. Dennis, M., Pierson, W.J., 1953. On the motion ships
in confused seas. Transactions of the Society Naval
Architects Marine Engineers (SNAME) 61, 280354.
Stevens, S. C. and Parsons, M. G., 2002. Effects of Motion at
Sea on Crew Performance: A Survey, Marine Technology,
Vol. 39, N 1, pp. 2947.
106

Generation of an adaptive triangular mesh from a parametric surface

J.M. Varela & Manuel Ventura
ABSTRACT: This paper presents a new method for the generation of an adaptive topologically correct triangular mesh on parametric surfaces. The level of approximation is controlled through a given tolerance resulting
into a density of triangles that increases with the local curvature of the surface. The surface is subdivided in
the parametric domain by a recursive algorithm which simultaneously stores the sequence of operations into a
Binary Space Partitioning tree data structure. To each interior edge defined by two or more vertices correspond
always only two faces. After the subdivision, the resulting polygonal faces are triangulated using all the points
generated. Finally the mesh obtained in the parametric domain is mapped back into the Cartesian 3D space. The
generated mesh is topologically correct and provides a basis for a wide range of engineering applications.
INTRODUCTION
Parametric surfaces are the basis of the geometric

models used in several domains of engineering. For
some specific applications, such surfaces can be
replaced by linear approximations assuming that every
point from the resulting mesh will be within a given
tolerance. The main motivation for this work is the
application to the simplified representation of the
ships hull form. In some engineering applications such
as the fast prototyping of the hull structures, it is a relevant factor to speed-up the process the replacement
of the surface by a mesh of planar elements.
The tessellation of parametric surfaces can be carried out in the R2 parametric domain or on simplified
surface approximations, in R3 . Polygonal meshes can
be classified in structured or non-structured. For some
engineering applications such as CFD computations
or ship dynamics, structured meshes are generally
required to obtain results. Non-structured meshes,
namely the triangular ones, have typical application to visualization and finite element computations,
because they are able to be better adapted to the hull
form. For the intended application, the final number of
triangles of the mesh and their topological correctness
is more important that any metric used to qualify the
triangles shape.
Triangular meshes present several advantages: they
allow a simple topological structure, more compact
data storage and are more easily adaptable to more
complex shapes without the use of subdivision or of
block generation methods, which generally require an
interactive definition of the boundaries.
Subdivision methods generate the mesh by subdividing the surface recursively in the parametric domain
into quadrangular faces whose size depends of a
pre-defined level of approximation. This subdivision
generates a topologically incorrect grid in the parametric domain, namely with vertices over the edges of
adjacent faces (the normally called T connections).
As a consequence, the triangular mesh generated
from the grid is also topologically incorrect and may
have undesired gaps between triangles. In order to
avoid this incorrectness, the proposed method identifies and stores adjacent faces and common vertices in
a Binary Space Partitioning (BSP) tree data structure
during the process of subdivision. With this additional
data, the triangulation of the grid in the parametric
domain is executed without Ts or holes and final result
is an adaptive and topologically correct triangular on
a parametric surface.
This work is organized as follows: first it is discussed the state of the art in mesh generation from
parametric surfaces; next, it is presented the adaptive
method proposed. Finally, some results of the application of the method to a sample free surface with
different curvature distribution regions are presented
and discussed.
2
MESH GENERATION
There is some research work done in the mesh generation of surface models without topological data.
Zimmermann (2003) and Bronsart et al. (2004) use
surface reconstruction processes from point clouds to
generate a non-structured mesh. An initial structured
mesh is created for each surface patch, based on an
admissible edge length. Next, each quadrilateral element is subdivided into two triangles by a diagonal.
The resulting triangular mesh is then locally refined,
subdividing triangles in areas where the distance from
the mean points from each side to the surface is
less than a given tolerance. The obtained vertices are
107
processed by a surface reconstruction algorithm that

produces a single mesh. Finally the mesh is simplified
by removing triangles by collapsing edges into a single
vertex. The selection of those edges is based on geometric quality criteria. Sharma and Sha (2003) present
a triangulation method applied to ship hulls, based on
the minimization of the energy of the parametric surface. In Bronsart et al. (2004) the automatic generation
of rectangular meshes from a NURBS surface model
is based on a variant of the advancing front method
in which instead of tessellating the edges of the surface and with those points start creating triangles to
the interior, the process begins from a triangle inside
the surface and is propagated until the boundaries are
reached.
Peterson (1994) presented an adaptive subdivision
method of parametric surfaces that generates a mesh
with a maximum distance to the surface less than
a given tolerance. However, the resulting mesh is
not correct, neither geometrically there are some
gaps between the faces nor topologically there
may appear vertices over the edges resulting into T
connections.
3
PROPOSED METHOD
A topological model is said to be manifold if each edge

is only shared by two faces (or one, if the model is not
closed) and the vertices can not be defined over the
edges of adjacent faces. In some CAD systems the surface models can also contain non-manifold elements,
but that case will not be considered in the present work.
The algorithm that generates the mesh is executed
in the parametric domain of the original surface (2D
coordinate space) and starts by defining a quadrilateral face by four corner vertices that covers the entire
surface. The subdivision of a face in the parametric
domain is based on a subdivision criteria previously
defined and applied to the original surface in the 3D
coordinate space. If the subdivision criteria are verified, then the face is subdivided and replaced by two
new faces. For each of the newly generated faces the
process is repeated until the stop condition is verified
and the final quadrilateral mesh is obtained in the parametric coordinate space. The topology, which in this
case consists in establishing the relation between the
quad faces and the vertices that define them, is generated and updated during the process of subdivision.
Therefore, the algorithm must guarantee that for any
step during the internal subdivision, the topology is
always correct. In the scope of the current study, the
vertices (corner points or not) that compose a face are
called nodes.
After the conclusion of the subdivision process the
result is a mesh of quadrilateral elements which have
a node at each corner and eventually some additional
ones along the edges. Next, a triangulation algorithm is
applied to each quadrilateral face in order to obtain the
final topologically correct triangular mesh, still in the
parametric space. The conversion from the parametric
to the three-dimensional coordinate space uses the
Figure 1. Node entity.
surface equation that establishes the correspondence

between the vertices on both spaces.
4
SUBDIVISION ALGORITHM
4.1 Geometric entities

The two main entities required by the subdivision
algorithm are the node and the face. The subdivision method is based on an axis aligned BSP tree
in the two dimensional parametric space of the surface. In the present case, the BSP tree represents the
recursive, hierarchical subdivision of the parametric
space into convex subspaces. The process of subdivision achieved by partitioning each subspace by an axis
aligned line that intersects the interior of that subspace.
The result is two new subspaces that can be further
partitioned by recursive application of the method.
The tree nodes and faces are stored in two different
lists that are filled during the subdivision process and
compose the output. Each face contains a list of at least
four nodes and each node is associated to a list of one
to a maximum of four faces. If the node is located at the
boundary of the parametric space, then it may have one
face if it is a corner node or two faces otherwise. If it
is an interior node it may have three of four associated
faces. Therefore, before starting the process of subdivision these two lists must be created. The tree node
represents a point in the parametric coordinate space.
Therefore, each node is defined by a position given
by two coordinates (u, v). Since the node is a point
in the surface, the values of u and v in a normalized
domain are always in the interval [0.0, 1.0]. Four nodes
connected by edges generate a face. A node is always
associated to four adjacent zones ordered in counter
clock wise (ccw) direction. Each association covers a
different zone of the space as shown in Figure 1 and
defines connection set to the face located in that zone.
A connection may be set to null if there isnt a
face on the corresponding zone. If the node is at the
boundary of the surface like q2 in Figure 2, then the
connections c1 and c2 of Figure 1 are set to null because
there are no faces on those zones. For the case of the
node q1 , the connections c2 and c3 are both set to the
face F10 , while for the node q0 all the connections are
set to different existing faces. Figure 2 also shows that
every node is always a corner node of at least one face.
108
Figure 2. Consecutive vertical and horizontal subdivisions

of a face.
The face is a four sided entity in the parametric normalized coordinate space. A face is always composed
by four nodes at the corners. Additionally it may also
have nodes along its edges as shown in Figure 2 for
face F10 . During the process of subdivision, each face
is divided into two subspaces (down-top or left-right)
and must be able to identify which node belongs to
each subspace. This can be done by comparing relative
positions of the edge and corner nodes. Connections to
nodes are stored in a list in which the order is irrelevant.
4.2
Initial condition
The initial condition of the recursive algorithm consists

in defining the first bounding face F in the parametric coordinate space. All the faces generated by the
algorithm lie in the interior of this initial one. Face F
is composed by four corner nodes that correspond to
the boundary corner points of the surface in the 3D
coordinate space.
4.3
Subdivision process
The subdivision process presented in (Peterson, 1994)

was adopted. The adaptive subdivision is a recursive
process in which the surface is successively split into
two halves in the parametric direction with the highest
curvature until a patch flat enough is obtained.
The flatness criteria is based on the analysis of the
grid of control points of the surface patch, and checks
four different aspects: if the edges are straight, if the
rows are straight, if the columns are straight and if
the four corners are in same plane, all within a given
tolerance.
When a surface is considered to be flat in a direction,
it is split in the opposite direction and the flatness information is inherited by the two patches resulting from
the subdivision. The surface splitting is implemented
with the Oslo algorithm (Bartels et al., 1987).
4.4
Face subdivision
Let F be the face in the parametric domain that will

be subdivided into two new faces F0 and F1 . Although
in the initial condition, face F is composed only by
Figure 3. Subdivision algorithm.
the four corners, during the subdivision process it

may have additional nodes in the edges as shown in
Figure 3. The subdivision algorithm is divided into
6 main steps which are described in the following
paragraphs.
4.4.1 Identify the nodes that compose the face
Face F is divided into two subspaces S0 and S1 . The
order of these subspaces is bottom-top or left-right
depending is it is and vertical or horizontal subdivision
respectively. The first step is to identify and store the
nodes of face F that belong to each subspace Si .
As result of this step, two lists of nodes Qlist 0 and
Qlist 1 are created corresponding to the two subspaces
of face F.
109
4.4.2
Create the new nodes to perform the

subdivision of the face
Face F is subdivided by the two nodes q0 and q1
specified in Figure 3.
The subdivision may be horizontal or vertical
depending of the subdivision criteria. These nodes may
or may not already exist in the list of nodes; therefore this check must be performed. If the node already
exists, then it only has to be identified. If it doesnt
exist in the list, it must be created, added to the list of
nodes and identified for subsequent operations in the
current iteration. For the case of Figure 3 the node q0
already existed while the node q1 was created.
4.4.3 Create the new faces
The third step is to create the two new faces F0 and
F1 from the two subspaces of F and add them to
the list of faces. An important key point is to maintain the coherence between relative positions of nodes
and faces in both horizontal and vertical subdivisions.
Therefore the new faces should be ordered in the same
way as the subspaces as shown in Figure 3, such
that Si Fi .
4.4.4 Add faces to already existent nodes
As previously mentioned, each node has a list of four
faces corresponding to the four different zones of Figure 1. Let Flist i be the list of four faces of the node
i, in which Flist i [j] is the face corresponding to the
adjacent zone j of the node.
Then, for each of the two Qlist i created in step 1,
go through the nodes of the list and replace the current
face F by the newly created face Fi . Note that for the
nodes of Qlist 0 , the face F is replaced by face F0 and
for the nodes of Qlist 1 is replaced by F1 . Moreover,
before the replacement, face F must always exist in
all the quad nodes of the lists, otherwise the algorithm
already failed somewhere before.
4.4.5 Add faces to the new nodes
The fifth step is to add the faces to the nodes obtained
from step 2. For each node q0 and q1 , the nearest node
before and after qi when rotating ccw, must be identified. Let them be qib and qia as shown in Figure 3 for
the four nodes in different situations.
Let Flist ib and Flist ia be the list of faces of nodes
qib and qia respectively. Then, for each node qi the face
lists are composed by the following elements:
Horizontal subdivision
Flist 0 [0] = Flist 0b [1]
Flist 0 [1] = Flist 0a [0]
Flist 0 [2] = F1
Flist 0 [3] = F0
Flist 1 [0] = F0
Flist 1 [1] = F1
Vertical subdivision
Flist 0 [0] = F0
Flist 0 [3] = F1

Flist 1 [1] = F0
Flist 1 [2] = F1
Figure 4. Space partition in the surface parametric domain.
4.4.6 Add nodes to newly created faces

For each new face the nodes of the corresponding subspace must be added. Therefore, add all the nodes of
Qlist i to the list of nodes of face Fi .
Additionally the new nodes must also be added to
the new faces and to the adjacent faces that already
existed. Using the same notation of step 5, this is
achieved by the following:
Horizontal subdivision
Add q0 to Flist 0b [1] Add q1
Add q0 to Flist 1a [0] Add q1
Add q1
Add q0 to F 1
Add q0 to F 0
Add q1
to F 0
to F 1
to Flist 1b [3]
to Flist 1a [2]
Vertical subdivision
Add q0 to F 0
Add q0 to Flist 0b [2]
Add q0 to Flist 0a [1]
Add q0 to F 1
to Flist 1a [3]
to F 0
to F 1
to Flist 1b [0]
Add q1
Add q1
Add q1
Add q1
Note that each node must be added only once to

each list. Therefore a check must be performed before
adding the node in order to verify if it doesnt already
exist in the list.
At this point it is obtained the complete binary space
partition of the parametric domain corresponding to
the required tolerance, as represented in Figure 4.
Finally, each face is triangulated using all the points of
the polygonal boundary and the resulting triangles are
mapped into the three-dimensional space, producing
the output mesh.
5 VALIDATION TEST
For the validation of the presented algorithm, a
NURBS surface patch from the fore body of a ships
hull was used and the two algorithms, the original
110
density increases directly with the increasing of the

local surface curvature and is topologically correct,
with a reduced increase of the number of triangles.
REFERENCES
Figure 5. Adaptive tesselation method (Peterson, 1994).
Figure 6. Surface tessellated with new method.
and the newly developed were applied, using the same

tolerance (0.05).
Figure 5 presents the tessellation based on the
Peterson method produces 1397 triangles with visible
topological inconsistencies.
The new method shown in Figure 6 produces 1554
triangles, which means that a mesh without any topological errors is obtained with an increase of only about
11% of the number of triangles.
6
CONCLUSIONS
A new method for triangular mesh generation from

parametric surfaces was developed. The method is
an improvement over the one presented by Peterson
(1994) to provide an adaptive tessellation of a parametric surface, which produces a triangle soup and
not a topologically correct mesh. The mesh resulting from the presented new method is adaptive, its
Bartels, R., Beatty, J. and Barsky, B. (1987), An Introduction

to Splines for Use in Computer Graphics and Geometric
Modeling, Morgan Kaufmann, CA.
Bronsart, R., Knieling, G. and Zimmermann, M. (2004),
Automatic Generation of a Panel-Based Representation of Ship Hulls for Wave Resistance Calculations,
Proceedings of PRADS 2004.
Cignoni, P., Rocchini, C. and R. Scopigno, R. (1998),
Metro: measuring error on simplified surfaces, Computer Graphics Forum, Blackwell Publishers, vol. 17(2),
June 1998, pp. 167174.
Cohen et al. (1996), Simplification Envelopes, Computer
Graphics, Proceedings of SIGGRAPH 96, Vol. 30, ACM
Press, New York, pp.163169.
Garland, M. and Heckbert, P. (1997), Surface Simplification Using Quadric Error Metrics, Computer Graphics, SIGGRAPH 97, Vol. 31, ACM Press, New York,
pp. 209216.
Guthe, M. (2005), Fast and Accurate Harsdorff Distance Calculation Between Meshes, Proceedings of the
WSCG 2005, Plzen, Czech Republic.
Klinger, A. (1971), Patterns and Search Statistics, in
Rustagi, J. (Ed.), Optimizing Methods in Statistics,
Academic Press, New York, pp. 303337.
Luebke, D.P. (2001), A Developers Survey of Polygonal
Simplification Algorithms, IEEE Computer Graphics
and Applications.
Owen, S. (1998), A Survey of Unstructured Mesh Generation Topology, proceedings of the 7th International
Meshing Roundtable, Dearborn, MI.
Pbay, P.P. and Baker, T.J. (2003), Analysis of Triangle
Quality Measures, Mathematics of Computation, Vol. 72,
No. 244, pp. 18171839.
Peterson, J.W. (1994), Tessellation of NURBS Surfaces,
Graphics Gems IV, P. S. Heckbert (ed.), Academic Press
Professional, New York, pp. 286320.
Sharma, R. and Sha, O.P. (2003), Generation of Topologically Continuous Smooth and Fair Triangular Elements
for Ship Surfaces, International Shipbuilding Progress,
Vol. 50, No. 3, pp. 209237.
Varela, J., Ventura, M. and Guedes Soares, C. (2008), Fast
Ship Structures Modelling System for Hull Maintenance
Support, Proceedings of the International Symposium
on Ship Repair Technology: Life Cycle Effect of Ship
Repairs, SRT 2008.
Warnock, J. (1969), A Hidden-Surface Algorithm for
Computer Generated Half-Tone Pictures, Technical
Report TR 4-15, NTIS AD-753 671, Computer Science
Department, University of Utah, Salt Lake City.
Zimmermann, M. (2003), Mesh Generation Algorithm for
Arbitrary complex Ship Hull Forms Using Unstructured
Mesh Generation Techniques, MSc Thesis (in German),
University of Rostok.
111
Ship propulsion & control

Controllable pitch propeller control with neural

network algorithm
Fuat Alarcin, Serkan Ekinci & Kayhan Gulez
Yildiz Technical University, Naval Architecture and Maritime Faculty, Istanbul, Turkey
ABSTRACT: In this paper, Controllable Pitch Propeller (CPP) has been used to improve the ship maneuvering
in the fixed shaft speed. The Proportional-Derivative (PD) control method has been applied in order to obtain
high efficiency by changing the pitch angle. These controller results have been evaluated and trained by Fast
Backpropagation Algorithm (FBA). The neural network results show that the training of the pitch behavior
has been verified, and the FBA performance of the CPP controller has a better performance than the PD
controller.
INTRODUCTION
CPP has recently attracted considerable attention on

ships with publicized demonstrations. The background
of CPP for ships was reviewed by Aarseth (2002). By
using this type of propeller, significant achievements
in ship maneuverability can be carried out. Selection of
the control schemes is also important during the design
process. To simplify models, a number of researchers
assumed three main control schemes. The first one
corresponds to a fixed pitch propeller. The second
alternative is to fix the shaft speed. The last alternative is applied by Fukuba et al. (1996), which examines
both pitch and shaft speed simultaneously. In this study
CPP with fixed shaft speed are used. Sorensen et al.
(1997) applied experimental verification on torque and
power control for fixed pitch propellers. Ruth (2004)
considered how to apply controllable pitch thrusters.
A method in this reference for model-based optimal
control of consolidated controlled thrusters, where
both pitch ratio and motor torque was developed.
In the last decade many researchers applied
some propulsion control system. Krger and Haack
(2004) developed a time domain simulation tool for
maneuvering purposes of propulsion control system. Izadi-Zamanabadi and Blanke (1999) presented
fault-tolerant control combined with fault detection
and isolation techniques with supervisory control.
Bonivento et al. (2002) used to develop control reconfiguration structure in order to achieve a fault tolerant
control system.
In the past decades, neural network algorithm has
been successfully applied in a broad spectrum of application areas. The advantages of Neural Network (NN)
controllers is that they are simpler structures and easier
to operate. Haykin (1999) states that a neural network is made up of non-linear neurons, which have
a natural ability to realize (approximate) non-linear

input-output relationship. Recently, NN techniques
have been utilized in many disciplines as well as in
marine fields (Raya et al., 1996; Haddara and Jinsong,
1999). In order to model a ship propulsion system,
backpropagation neural networks are used by Ferreiro
et al. (2005).
This paper presents an approach of controlling the
blade angle of CPP which is fixed shaft speed. Furthermore, the PD and FBA neural networks algorithms
were used to adjust the pitch angle. Simulation results
showed the effectiveness of the FBA neural networks
algorithms.
2
PROPELLER PITCH CONTROL
CPPs are used to improve maneuvering behavior. The

CPP blade pitch is controlled hydraulically through a
system of pitch actuator. Kim et al. (1998) models the
pitch actuator as a first order transfer function, formulated as coefficient terms. The following equation of
motion approximates the ship speed dynamics.
The term Xu represents the added mass in surge which

is negative. Xhydro , Xrudder represents hull and rudder
forces respectively. The thrust deduction (1 t) represents the net thrust lost due to the propeller-generated
flow at the ships stern. t is a constant coefficient taking
values in [0.05, 0.2] (Fossen, 1994).
The pitch controller tunes the blade angle corresponds to the lever. The propeller pitch system, which
is shown in Figure 1, is approximated by a simple integrator modeling the pitch dynamics and proportional
controller. Defining ( min , max ) the rate limits set for
115
Figure 1. Diagram of the pitch controller.
the hydraulic actuator, the overall system is described

by Bonivento et al. (2002).
(min , max ) is the physical interval for propeller blade

travel. The pitch control system is described by the
following equation.
where , d are the measured and desired propeller

pitch respectively. kt the gain of pitch controller.
The resulting constraint on the input and output
variables are described by Kerrigan and Maciejowski
(1999).
In order to design a controller for a CPP, it is essential to

take the effects of variable pitch into consideration. In
this paper, two different methods, PD and FBA neural network algorithms, are applied for the propeller
acting.
Control systems are comprised of the three-stage
four-way electro hydraulic proportional directional
valve, a hydraulic linear actuator, (Figure 2). The
differential equations governing the dynamics of the
hydraulic actuator are given in Merrit (1967).
A mathematical model of the plant can be derived
from the flow equation of the valve, the continuity
equation and balance of forces on the piston. The
valve flow-rate equation is dependent on the valve
displacement from neutral network (Zeb, 2003).
where QL is the load flow (m3 /s), xv is the valve

displacement from neutral position (m),
Kq is the valve flow gain,
Kc is the valve flow pressure co-efficient.
Newtons second law is applied to the forces on the
piston. The force equilibrium can be expressed by,
where F is the actuator force, P1 P2 is load pressure difference, the piston ram area represents A, Mt
represents the total mass of the piston including the
Figure 2. Schematic Diagram of Pitch servo system.
control propeller inertia, Be is the equivalent viscous damping coefficient, K is load spring constant
and Fd represents the external hydrodynamic disturbance force including the frictional force. The angular
position of the propeller is determined by the piston position = xp /R where R represents the effective
torque arm of the linkages. The servo valve flow with
a matched and symmetric orifice can be expressed as
follows.
where Cq is the flow coefficient of the servo valve, w

is the area gradient of the servo valve, xv is the servo
valve spool position, signal (.) is the sign function, is
the density of hydraulic oil, Ps is the supply pressure.
The continuity equation of the cylinder is expressed as
follows.
where Ctp is the total leak coefficient of hydraulic

cylinder is, Vt is the total volume of fluid under compression of two chambers and e is the effective bulk
modulus of the oil. The bulk modulus of hydraulic
oil depends on temperature and pressure. MaCloy and
Martin (1980) gave the variable bulk modulus. Eqns.
(79) are there basic equations of the cylinder controlled by the valve. They determine its dynamic performance. Applying the Laplace Transformation and
116
simplification of Eqns. (79), the following equation

can be obtained (Zeb, 2003).
where
Figure 3. Block diagram with PD controller.
The state space description of a valve cylinder electrohydraulic servo force control system will be derived
dynamic equation of the system. The mathematical
model of hydraulic rudder servo system is obtained by
combining the equation which describes the actuator,
valve and pressure dynamics.
The controller design objective is to control the propeller pitch displacement. Design task is to provide
the stability of a propeller pitch control that has been
shown by the block diagram of the system with a conventional PD controller in Figure 3. Here, d is the
desired pitch value, e (t) = d is error and uPD is
the control signal. The Proportional-Derivative (PD)
controller takes the form as follows:
Derivative (PD) controller takes the form as
follows:
Kp > 0, Kd > 0 where Kp , Kd are the proportional,

derivative gains respectively. The PD control method
is employed in the pitch control. The control is made
by proportioning the pitch angle.
2.2
The linearized model is obtained about the equilibrium as,
where b is the control gain, u, v and f represent the

system control input voltage the system output and
the external disturbances respectively.
2.1
Proportional-derivative (PD)
controller design
The proportional-derivative (PD) method is often used

in classical control design to improve the performance.
Neural network structure for propeller pitch

stabilization
As referred earlier as powerful computational tools,

NN techniques have been utilized in many disciplines
as well as in marine design fields (Raya et al., 1996;
Alkan et al., 2004). The feed-forward NN usually is
trained by a back-propagation algorithm. Here, the
architecture of NN is trained for a set of experimental results with FBA and they describe an efficient
criterion for concrete structures.
As a result, highly accurate predictions based on
a given set of information can be provided. The use
of these predictions can then be used in numerous
applications with the wide aspect design systems to
real-time control ones. The latter are being developed
into controllers to take advantage of this capability.
In Akan et al. (2004), a wide explanation of FBA
is given.
The Neural Network controller uses the pitch actuator for pitch angle. All data sets are established
similarly to apply to the Neural Network controller.
The Neural Network chosen to implement the neural model to the pitch angle is constituted by four
layers as shown in Figure 4. The first hidden layers have 21 neurons and the second hidden layer has
20 neurons.
117
Figure 5. Propeller Pitch comparison results.
Table 2. Some values of the training set used for the

algorithm.
Uncontrolled
PD
CBA
1.0466
3.3261
5.4323
2.7132
3.797
6.6144
0.6968
3.3176
0.81519
3.7611
1.3702
0.5151
0.691
3.38
0.81
1.573
1.053
0.74
Figure 4. Neural network controller for CPP (Alkan et al.,

2004).
4
Table 1.
Description
Param.
Value
Unit
Length
Displacement
Inertia
Weight
Propeller Diameter
Pitch angle control
Im
M
D
kt
147.2
12 840
0.25 106
13 282
6 123
0.15
m
m3
Kgm2
tons
m
m
CONCLUSIONS
Ferry main data.
This article describes research on a ship propulsion

system using Controllable Pitch Propeller. In order to
obtain satisfactory results, a Fast Back-propagation
Algorithm and PD control were applied for a CPP.
In comparison, FBA provided effective controlling in
the pitch angle in the constant of the shaft speed and
achieved a better performance than the open loop system and PD control results. In addition, the propeller
pitch angle control with neural network algorithm
has been determined to be more significant for ferry
propulsion.
SIMULATION RESULTS
REFERENCES
In order to demonstrate effectiveness of the proposed controllers a series of simulations have been
performed on the ferry propeller propulsion system.
Table 1 summarizes the main data of the ferry as
reported by Izadi-Zamanabadi and Blanke (1999).
The Neural Network structure for the input and
output functions for the CPP are given in Figure 5.
A conventional PD controller technique in a propulsion system is compared with intelligent controller
technique based in Fast Back-propagation neural
Networks.
Figure 5 shows that the CPP produces better pitch
angle by means of Fast Back-propagation Algorithm
than the PD controller. The actuator torque controller
is just a feed-through of the desired stable pitch angle.
The stability of the controller depends on the properties
of the PD controller. Some values of the training set
used for the algorithm is given in Table 2.
Aarseth, L. (2002), Analysis and control of propulsion systems subject to thrust losses and power limitations in
extreme seas. Thesis (pre-project). Norwegian University
of Science and Technology.
Alkan, A.D., Gulez, K. andYilmaz, H., (2004), A knowledge
based computational design tool for determining preliminary stability particulars of naval ships, NEJ Naval
Engineers Journal, 116, No. 4, pp. 3751.
Bonivento, C., Paoli, A. and Marconi, L., (2002), Faulttolerant control of the ship propulsion system benchmark
Control Engineering Practice Vol. 11, pp. 227239.
Ferreiro, R., Haro, M., and Velasco, F.J., (2005), Trends on
modeling techniques applied on ships propulsion system
monitoring, Journal of Maritime Research, Vol. II. No. 1,
pp. 87104.
Fossen, T. I., (1994), Guidance and control of ocean vehicles, pp. 246257. New York: Wiley.
Fukuba, H., Morita, S. and Maeda, T., (1996), Simulation
of nox reduction by consolidated control main engine and
118
cpp, Bulletin of Marine Engineering Society in Japan,

24(1), 2127.
Haddara, M.R. and Jinsong, X., (1999), On the identification of ship coupled heave-pitch motions using neural
networks. Ocean Engineering, Vol. 26, pp. 381400.
Haykin, S., (1999), Neural Networks a Comprehensive
Foundation. Prentice Hall International, Inc..
Izadi-zamanabadi, R., Blanke, M., (1998), A ship propulsion
system model for fault tolerant control Technical Report.
Control Engineering.
Izadi-zamanabadi, R., Blanke, M., (1999), A ship propulsion
system as a benchmark for fault-tolerant control, Control
Engineering Practice Vol. 7, pp. 227239.
Jarzyna, H., Koronowicz, T. and Szantyr, J., (1996), Design
of marine propellers selected problems Gdansk, Poland.
Kerrigan, E. and Maciejowski, J., (1999), Fault-Tolerant
Control of a Ship Propulsion System Using Model Predictive Control, Proc. of ECC, Karlsruhe, Germany.
Kim, Y.-B., Byun, J.-H., Jung, B.-G., and Yang, J.-H., (1998),
Optimization control for fuel consumption a ship propulsion system with cpp using decoupling and lmi approach,
Control Applications in Marine Systems, pp. 315320.
Krger, S. and Haack, T., (2004), Design of propulsion control system based on the simulation of nautical
maneuvers. 9th Symp. On Practical of Ships and Other
Floating Structures, Germany.
Lewis, E.W., (1981), Principles of naval architecture
Society of Naval Architecture and Marine Engineers.
New York.
Raya T., R. Gokarna P., and Shaa, O.P., (1996), Neural
network applications in naval architecture and marine
engineering, Artificial Intelligence in Engineering,
Vol. 10(3), pp. 213226.
Ruth, E., (2004), Control of thruster with variable pitch,
Project Thesis, Norwegian University of Science and
Technology.
Sorensen, A.J., Adnanes, A.K., Fossen, T.I., and Strand, J-P.,
(1997), A new method of thruster control positioning of
ships based on power control, 4th IFAC Conference on
Maneuvering and Control of Marine Craft.
119

Conversion of large-bore diesel engines for heavy fuel oil

and natural gas dual fuel operation
J.M. Gomes Antunes
TecnoVeritas, Mafra, Portugal
Sir Joseph Swan Institute for Energy Research, Newcastle University, Newcastle upon Tyne, UK
R. Mikalsen & A.P. Roskilly

Sir Joseph Swan Institute for Energy Research, Newcastle University, Newcastle upon Tyne, UK
ABSTRACT: Due to potential fuel efficiency and exhaust gas emissions advantages, interest in dual fuel
operation of internal combustion engines is increasing. This paper describes the conversion of two marine
4.5 MW diesel engines, modified to use a combination of natural gas and heavy fuel oil. The development of
the engine monitoring and management system is described, including the use of knock measurement to control
the amount of natural gas used. Experimental results are presented, showing the performance of the engines
under different modes of operation.
INTRODUCTION
Increasing fuel costs and tightening environmental legislation drive an interest among users of existing diesel
engines to look at options to improve fuel efficiency
and reduce exhaust gas emissions formation. In most
places of the world, natural gas is continuously available at a reasonable price, and its properties make it an
interesting alternative fuel for use in internal combustion engines. The option of converting existing diesel
engines to dual fuel capability allows flexibility in the
fuel supply, but also potential reductions in operational
costs and exhaust gas emissions. However, the conversion of existing diesel engines to dual fuel operation
implies possible combustion problems that need to be
addressed.
Natural gas as a fuel exhibits a number of interesting advantages such as a reasonable net calorific value
(approximately 38,500 kJ/Nm3 ), low emissions of
CO2 , NOx and particulate matter, and generally a good
knock resistance (which is necessary to allow use in
diesel engines without engine modification). From the
point of view of availability, where available, its supply
is continuous and its price is commercially appealing. When converting a standard diesel engine to dual
fuel operation, engine operational characteristics can
remain largely unchanged, with the liquid fuel acting as
a pilot fuel to ignite the cylinder charge. Therefore the
quantity of pilot fuel must be set to release the minimum ignition energy required to ignite the gaseous
fuel, which constitutes the main energy source of the
working cycle. Due to the pre-mixed charge, problems
of pre-ignition and detonating combustion (knock)
may, however, occur, and the use of an appropriate
engine control system is therefore essential.
Numerous researchers have studied the use of natural gas as a second fuel in diesel engines [see reference
list]. Generally, fuel efficiencies comparable to those
under pure diesel engine operation are achieved at high
engine loads; at low loads dual fuel operation gives
poorer fuel efficiency. Papagiannakis and Hountalas,
(2003 and 2004) attribute this to the poor combustion of the gaseous fuel at part load operation. With a
reduction in the use of liquid diesel fuel, the particulate matter (PM) emissions are significantly reduced;
reductions of above 50% are frequently reported.
Also nitrogen oxides (NOx ) can be reduced in dual
fuel mode, however increases are usually seen in the
emissions of carbon monoxide (CO) and unburned
hydrocarbons (HC).
2
DUAL FUEL OPERATION
The cylinder charge of dual fuel engines (converted

existing diesel engines or simply diesel derived
engines) is usually made up of two fuels with distinct
ignition temperatures and different physical states,
typically one liquid and one gaseous. This gives a combustion process different from those of conventional
diesel or spark ignition engines, and gives the dual fuel
engine some particular operating characteristics which
need to be taken into account when working with such
engines.
2.1 The dual fuel combustion process
The combustion process in dual fuel engines is somewhat more complicated than that of conventional
engines since a combination of premixed and diffusion
combustion occur in this mode of engine operation.
121
The contribution and characteristics of each type of

combustion depends on several parameters, including
fuel properties, injector characteristics, and combustion chamber design, as well as operational variables
such as the engine load, speed, manifold air pressure
and temperature, and the amount of each fuel present
in the combustion chamber.
The combustion process in a dual fuel engine can
be divided into three distinct sub-processes:
ignition of the pilot fuel;

combustion of the gaseous fuel which is in the
vicinity of the pilot fuel cores; and
combustion of the gaseous fuel due to flame
propagation into the premixed lean charge.
The equivalence ratio of the cylinder charge varies
spatially from point to point within the combustion chamber, in particular considering that a fast
homogenisation of the mixture of natural gas and
liquid fuel in most cases has no time to be ideal. At
low loads, the air-gas mixtures may be so lean that
it causes flame propagation interruptions, resulting
in incomplete combustion. This will lead to loss of
fuel through the exhaust, contributing to a reduction in fuel efficiency and high levels of unburned
hydrocarbons in the exhaust.
2.2
Detonation
Detonation occurs during the combustion process

when the burnt gas zone, which is pressurising and
heating the unburned part of the cylinder charge ahead
of the flame front, does so at a such rate that the
unburned fuel achieves its auto-ignition temperature
before the arrival of the actual flame front. The result
is that the unburned charge volume ignites spontaneously over the entire zone where the auto-ignition
conditions have been achieved. The apparent flame
speed in this zone is many orders of magnitude faster
than that of conventional combustion initiated by the
normal flame front, resulting in higher rates of pressure
and temperature rise. Knocking is usually associated
to small portions of volume of the combustion chamber, whereas detonation is associated to the entire
compressed cylinder charge.
Engine knock is directly related to the compression ratio, because the higher the compression, the
closer the charge will be to its auto-ignition conditions. Moreover, the knock intensity will depend on
the intake air conditions, combustion chamber design
and turbulence levels within the cylinder, as well as
the flame speed of the fuel-air mixture. Therefore,
for turbocharged dual fuel engines the temperature of
the air entering the cylinder and its pressure are factors that influence detonation sensitivity. Detonation
is accompanied by a drastic increase in temperature
and pressure within the cylinder, resulting in some
cases in serious engine damage.
In converted diesel engines, the quality of the diesel
fuel oil used is of utmost importance to attain a
smooth and clean combustion. Heavy fuel oils with
bad ignition quality have long ignition delays, and
therefore by the time ignition takes place, the piston

is too close to the TDC, leading to a combustion of
the gaseous part of the cylinder charge with faster
flame speed reaching higher combustion pressures.
The temperature of the heavy fuel is also an important
factor to consider, as viscosity and therefore atomisation of the fuel play an fundamental role in the
homogenisation of the cylinder charge.
2.3 Pre-ignition
Another dangerous and damaging phenomenon of
pre-mixed engine operation is the pre-ignition of the
cylinder charge. This phenomenon results from the
igniting of the cylinder charge through contact with a
high temperature surface prior to the desired ignition
timing (i.e. at the time of pilot fuel injection, or spark
discharge in a spark ignited engine). Pre-ignition is
particularly dangerous since ignition can occur during
the compression stroke, leading to excessive mechanical stress and damages in the crank system, piston
rings, bearings, etc., as well as increased thermal
stress and detrimental effects on cylinder lubrication
due to very high in-cylinder gas temperatures.
The main cause of such a phenomenon is the presence of carbonaceous residues on the combustion
chamber surface. With the use of low-quality fuels
such as heavy fuel oils, the amount of residues on
the combustion chamber components are likely to
increase, and this must be taken into consideration
when converting an engine to dual fuel operation. The
ignition temperature of a natural gas-air mixture will
depend on the fuel-air ratio, and the fuel substitution
ratio and natural gas concentration may therefore be
limited by the pre-ignition of the cylinder charge.
Pre-ignition tendencies in a cylinder will also
have a self-enhancing effect in that the increased incylinder gas temperatures and pressures resulting from
such an occurrence will increase the temperatures of
the combustion chamber walls and thereby increase
the probability of pre-ignition in the next cycle.
With the use of low-quality fuels in a dual fuel engine,
the monitoring and control of the combustion process
to avoid pre-ignition is therefore essential in order to
maintain engine integrity and operational stability.
3
ENGINE CONVERSION
TecnoVeritas has performed engine conversions

worldwide since 1999, including a range of engine
makes, sizes, and different fuels including dual fuel
and multi fuel using diesel oil, heavy fuel oil, natural
gas, producer gas, and hydrogen. A range of in-house
developed technology solutions for engine monitoring,
control, and operational optimisation have been developed. The following sections present a case study of
two Wrtsil 9L32 diesel engines converted to dual
fuel operation on heavy fuel oil and natural gas, and
describe technological solutions chosen as well as
operational data from the engines.
122
Figure 1. The two Wrtsil 9L32 diesel engines converted

at ECE, Portugal.
Table 1.
Engine data.
Engine model
Number of cylinders
Cylinder bore
Stroke length
Speed
Mean effective pressure
Mean position speed
Rated power
Boost pressure
Compressure ratio
Wrtsil 9L32
9
320 mm
350 mm
750 rpm (12.5 s1 )
21.3 bar (2130 kPa)
8.75 m/s
4,450 kW
2.4 bar (240 kPa)
12
Figure 2. Injection controller.
In December 2008, TecnoVeritas finished the commissioning of two nine-cylinder Wrtsil type 9L32
diesel engines, each with approximately 96,000 hours
of operation on heavy fuel oil (HFO). The objective
was to convert these two HFO engines to use as much
natural gas as possible, without creating mechanical
stress higher than that created under normal HFO
operation. Simultaneously, the development of operation software with an appropriate human interface
was required. Figure 1 shows a photograph of the
two engines, and main engine design data are listed
in Table 1.
adjusted via the injection angle and gas pressure, therefore allowing a fine adjustment of the gas energy
per stroke. The system allowed the control of the gas
quantity individually therefore offering a high flexibility in what concerns the engine operation. The
gas fuel injection system was continually monitored
via the KDS (knock Detection System developed
by TecnoVeritas). The KDS system includes one
accelerometer per cylinder identifying in each cycle
the knocking frequencies, or the absence of combustion (misfiring). In anyone of these conditions alarms
and commands are originated to reduce the load to a
particular cylinder or to all of them or even transferring
the engine operation from Dual-fuel to diesel.
3.1
3.2 Performance monitoring and control
Fuel injection system
Gas fuel port injection valves were controlled by an

injection controller receiving information from various sensors and systems installed on the engine, from
the gas regulating unit and from the process controller.
The original mechanical controller was replaced by
an hydraulic actuator controlled by the main injection
controller allowing in this way the transfer of diesel to
dual fuel mode and vice versa. The engine control was
changed to a gas control for a certain percentage of
gas, while remained on fuel oil below that percentage.
The injection of the correct quantity of gas during the
induction stroke, after the closing of the exhaust valve
and before the closing of the inlet valve is continuously
Performance monitoring and engine control equipment has been developed by TecnoVeritas, including
dedicated hardware and software such as the VTec
multi-point port injection system for the conversion
of existing diesel engines into true dual fuel engines,
the KDS Knocking Detection System, and the EDS
Engine Diagnosis System.
Prior to conversion, the engines were surveyed and
operating variables such as pressures, temperatures,
and specific fuel consumptions were logged to be
used as a reference. Based on the operational values,
the engineering team proceeded with the customised
design of the system.
123
Figure 4. Graphical user interface of the VTec performance

monitoring system.
Figure 3. Gas injection valve.
As a consequence of such a demanding contract,

a comprehensive cylinder pressure monitoring system
was fitted to monitor the combustion cycles during
system tuning and normal operation. The VTec
system is based on individual cylinder gas port injection, therefore allowing the correction of cylinder
parameters such as temperature, maximum combustion pressure and knocking, by controlling the quantity
of gas being injected in each cylinder per cycle.
The KDS Knocking Detection System, connected
via a CANbus network, comprise acceleration sensors
incorporated into the cylinder head, giving information on the vibration levels in the proximity of each
cylinder. The KDS system provides information on
the knock or pre-ignition intensity to the injection
valves controller, in order to change the injection timing and/or gas injection pressure for optimal engine
operation. Another function of the KDS system is to
determine if any misfiring occurs, for example in the
case of a gas injector malfunctioning. In such case,
an alarm with indication of the defective cylinder is
set and the engine immediately transferred from dual
fuel to HFO operation. The KDS system has two position transducers one, on the flywheel and a second
on the camshaft, both signal are acquired to exactly
determine the position and phase of the TDC of each
cylinder. The KDS uses an algorithm to perform a Fast
Fourrier Analysis at each cycle, therefore allowing
the identification of knocking frequencies an their
maximum amplitude before rising an alarm and output
an order to the injection controller for the reduction of
gas to a determined cylinder or even the transfer of
operation from Dual-fuel to HFO.
As the objective is to maximise the use of natural
gas, the engine management system will seek the highest possible substitution ratio of natural gas to HFO.
If the knocking or pre-ignition intensity increases, the
gas flow is reduced after a number of cycles (typically 5), and after a third reduction the gas is turned
off, while the engine operation is transferred seamless to HFO or diesel. Hence, the use of natural gas
is maximised, while taking into account the varying
conditions in the engine, in particular in relation to

carbon deposits which promote pre-ignition of the
cylinder charge.
The use of a dual fuel system operated with HFO
and natural gas requires particular attention on the
injection equipment condition. This was the reason
why both engines were equipped with the EDS Engine
Diagnosis System, which allows the monitoring of
the combustion process in each individual cylinder.
In addition to detecting any injection valve malfunction, the injection controller is capable of making
corrections and optimising engine operation based on
the engine manifold pressure, exhaust gas temperatures, etc.
Figure 4 shows the graphical user interface of the
engine management system. It provides the operator with vital engine performance variables, including
boost air pressure and temperature, natural gas properties, exhaust gas temperatures, knock intensity, and
electric output, as well as fuel consumption and substitution ratio. The operator can decide the substitution
ratio set point, as well as switching the engine back to
conventional diesel or HFO operation.
4
EXPERIMENTAL RESULTS AND PLANT

PERFORMANCE
After successful conversion, some performance tests

were done to study engine fuel consumption and
exhaust gas emissions under different modes of operation. As described above, the engines were run with
the highest possible substitution ratio of natural gas
to heavy fuel oil, limited by the knocking intensity in
the cylinders.
4.1 The combustion process
Figure 5 shows in-cylinder pressure traces obtained
during the tests. The graphs show pressure plots for
operation on heavy fuel oil only, normal dual fuel
operation, and for dual fuel operation with knocking due to too high natural gas substitution ratio.
124
Figure 7. Accumulated heat release in dual fuel mode with

and without knock.
Figure 5. In-cylinder pressure plots for different engine

operating modes.
Figure 6. Net heat release rates.
At normal dual fuel operation, the fuel composition

is approximately 70% natural gas and 30% HFO (on
an energy basis), and it can be seen that the performance closely resembles that of normal, HFO-fuelled
operation. For higher substitution ratios, a significantly faster pressure rise and higher peak pressure can
be seen, illustrating the need for the knock detection
system and appropriate injection control.
Figures 6 and 7 further illustrate this, showing the
calculated net heat release rates from combustion,
derived from the pressure plots. Again, with an appropriate substitution ratio, the operation on dual fuel
differs only very little from that on heavy fuel oil.
However, with too high levels of natural gas, characteristic knocking behaviour is observed, with detonating
combustion at a crank angle of around 190. The knock
behaviour leads to pressure waves in the combustion
chamber, compromising the measurements; since the
heat release rate is calculate from the pressure measurements, one gets oscillations as can be seen in the
graph.
4.2 Exhaust gas emissions formation
Regarding emissions formation, natural gas has a number of advantageous features compared with diesel oil
or heavy fuel oil. First, the amount of CO2 produced
per unit energy delivered is lower than that of more
complex hydrocarbons. In the tests, a reduction in CO2
emissions of approximately 16% was obtained in dual
Figure 8. Exhaust emissions in Dual Fuel operation

compared with Diesel HFO operation.
fuel mode compared to that of HFO operation. Second,

particulate matter emissions, an issue of great concern
in diesel engines, dropped by 50% compared to the
exhaust emissions at the same load on HFO, due to the
replacing of a large fraction of the fuel with pre-mixed
natural gas. Finally, NOx emissions dropped by 10%
which is somewhat lower than those reported by other
authors such as Mustafi and Raine (2008).
4.3
Fuel cost and financial viability
The pay-back period of the present conversion project,

based on the achieved substitution ratio and the price
of natural gas and HFO at the time of commissioning
(Oct. 2008), was estimated to be less than one year.
For diesel oil operated engines, the payback period
can be even shorter; diesel oil is more expensive than
HFO and higher substitution ratios can be achieved
(above 90% has been demonstrated), while maintaining the same power and slightly reducing the exhaust
gas temperatures.
5
CONCLUSIONS
The conversion of two marine, large-bore diesel

engines for operation on heavy fuel oil and natural
gas was described. The technical solutions for engine
125
conversion were described, including the fuel injection

system, performance monitoring system, and knock
detection system. Experimental results were presented
showing the performance under standard heavy fuel oil
operation and dual fuel operation.
It was shown that a high substitution ratio could be
achieved with natural gas supplying 70% of the fuel
energy under dual fuel operation. Knocking behaviour
was observed with too high substitution ratios, demonstrating the need for a knock detection system to
allow optimised engine operation and maximised
substitution ratio. Significant reduction in exhaust
gas emissions, including NOx , particulates, and CO2
were found under dual fuel mode compared with
conventional heavy fuel oil based operation.
The conversion of Diesel generating sets to dual fuel
operation, can be considered a reliable and effective
way of reducing the ship emissions while in port.
REFERENCES
Carlucci, A.P., de Risi, A., Laforgia, D., Naccarato, F.
Experimental investigation and combustion analysis of
a direct injection dual-fuel dieselnatural gas engine.
Energy, Volume 33, Issue 2, February 2008, Pages
256263.
Balasubramanian, V., Sridhara, K., Ganesan, V. Performance
Evaluation of a Small Agricultural Engine Operated on
Dual Fuel (Diesel + Natural Gas) System. SAE Paper

951777, 1995.
Mbarawa, M., Milton, B.E., Casey, R.T. Experiments and
modelling of natural gas combustion ignited by a
pilot diesel fuel spray. International Journal of Thermal
Sciences, Volume 40, Issue 10, 2001, Pages 927936.
Mustafi, N. N., Raine, R. R. A Study of the Emissions of
a Dual Fuel Engine Operating with Alternative Gaseous
Fuels. SAE Paper 2008-01-1394, 2008.
Papagiannakis, R.G., Hountalas, D.T. Experimental investigation concerning the effect of natural gas percentage on performance and emissions of a DI dual fuel
diesel engine. Applied Thermal Engineering, Volume 23,
Issue 3, February 2003, Pages 353365.
Papagiannakis, R.G., Hountalas, D.T. Combustion and
exhaust emission characteristics of a dual fuel compression ignition engine operated with pilot Diesel fuel and
natural gas. Energy Conversion and Management, Volume
45, Issues 1819, November 2004, Pages 29712987.
Papagiannakis, R.G., Hountalas, D.T., Kotsiopoulos, P. N.
Experimental and Theoretical Analysis of the Combustion and Pollutants Formation Mechanisms in Dual Fuel
DI Diesel Engines. SAE Paper 2005-01-1726, 2005.
Papagiannakis, R., Hountalas, D., Rakopoulos, C. Combustion and Performance Characteristics of a HSDI
Diesel Engine Operating from Low to High Natural Gas
Supplement Ratios at Various Operating Conditions. SAE
Paper 2008-01-1392, 2008.
Selim, M.Y.E. Pressuretime characteristics in diesel engine
fueled with natural gas. Renewable Energy, Volume 22,
Issue 4, April 2001, Pages 473489.
126

Conditioned maintenance predictive analysis in maritime

propulsion engines
S.S. Lampreia
Escola Naval, Department of Mechanical Engineering and CINAV, Portugal
V. Vairinhos
CINAV, CENTEC (IST) and ISLA Santarm, Portugal
A.S. Matos, J.G. Requeijo & J.M. Dias

UNIDEMI, Department of Mechanical and Industrial Engineering, Faculdade de
Cincias e Tecnologia (FCT/UNL), Universidade Nova de Lisboa, Caparica, Portugal
ABSTRACT: Condition Based Maintenance (CBM) finds today ideal conditions for its implementation:
current development of sensor networks, low-cost wireless data transmission and sensors with local processing
capacity; all this coupled with the need for more efficient and effective maintenance policies. The method we
propose is based on statistical processing of signals and consists in applying control charts to monitor system
condition in the context of CBM, aiming the reduction of failure rate, increasing both reliability and availability
of maritime equipment. Control charts, being a primary tool for control and monitoring of processes, can be used
to control variables such as oil pressure and temperature. This will allow us to distinguish between common and
special causes of data variation. With this methodology we intend, initially, to estimate the statistical parameters
of signals, by applying the Standard Deviation Chart. In a second stage, for monitoring and modeling the behavior
of equipment, we applied modified CUSUM (Cumulative Sum) and EWMA (Exponentially Weighted Moving
Average) charts.
MAINTENANCE METHODS
Today lots of methods and techniques are applied in

condition monitoring of equipments and systems. In
this article, for the equipment object of study, we
found a failure in the monitoring methodology: a continuous analysis of data collected systematically was
lacking, preventing a timely detection of trends, eventually avoiding catastrophic failures such as those that
happened in the past. There is then a gap between the
real occurrence of trend in the collected characteristic values, and the reaction time where it would be
possible to avoid a catastrophic failure.
In the engine under investigation, there was a catastrophic failure leading to total crankshaft fissure,
Figure 1. Since it was not the first time that such catastrophic failures occurred, it was decided to develop
and implement a methodology aiming to prevent
the recurrence of such kind of damage or, at least,
minimize damage from future breakdowns.
With CBM a variant of preventive maintenance
only maintenance actions effectively required by the
state of machine, supposedly known, is performed.
Let us recall various methods commonly used in
monitoring systems with CBM: Parametric Regression Models, AFTM (Accelerated Failure Time
Models) and the nonparametric PHM (Proportional
Figure 1. Damaged Engine.
Hazards Models). These methodologies allow modelling situations in function of variables that represent
the performance of machines, Dias (2002).
Given the nature of the data collected by Nautos
(Naval Automation System), we believe that SPC
Statistical Process Control can be an appropriate
methodology to evaluate the condition of the machinery in study.
SPC is a methodology to estimate the parameters of
a process and monitor its behaviour through statistical
control charts. With SPC, the decision maker knows
127
estimate the parameters of process control; Phase II

monitor the equipment condition applying SPC, Dias,
Requeijo and Pereira (2009).
When the sample size is higher than 10, it is not
advisable to use the range to estimate the dispersion of
the process, so we use the average and standard deviation control charts. We estimate the standard deviation
of oil pressures of a diesel engine defined by manufacturer. In this case, not knowing the parameters of
the first phase, we use global average (X ) and average
as estimates, respectively, for
standard deviation (S),
the mean and standard deviation of the process.
Figure 2. Study Objectives.
the process is able to produce according predefined

technical specification Requeijo (2010). It is believed
that with the application of this method not in an
environment of production control, but in analysis of
diesel engine operation we can identify the need of
an intervention and the cause of the variation of the
selected characteristic.
In synthesis: the object of this study is SPC used
to model the operation of maritime propulsion diesel
engines, as an instrument to implement a CBM system
that reduces the occurrence of failures and associated
costs, initiating a process of continuous improvement
of the maintenance and the operating performance of
the systems, Figure 2.
Data for this work was collected by the NAUTOS
system of one of the ships. We considered 40 days
of data collection. The original data is collected as
TEXT data, transformed to an applicable form for
this research and then converted to EXCEL. This
operation was supported by the program MECPABLERNAUTOS (developed by the Naval Research
Center-CINAV).As a complement to data organization
and analysis we also used STATISTICA.
2 APPLIED METHODS
Shewhart control charts are the most commonly used
form of control charts. These charts use only the most
recent data and therefore are slow to detect early
changes in a production process. Despite this handicap, they have other advantages, so lets apply them
to define the parameters to be used in modelling the
equipment control chart, Pereira and Requeijo (2008).
2.1 Average and standard deviation control charts
Traditionally, two control charts are simultaneously
applied: one for monitoring the mean and another
for monitoring the process dispersion. This process
has two phases: Phase I, assuming that the equipment
has a stabilized operation, m samples are collected to
The control limits of the S Chart, Pereira and

Requeijo (2008), were calculated as follows:
In the previous expressions, B3 and B4 only depend

on sample size and can be read from tables for control
charts (Pereira and Requeijo, 2008).
The standard deviation chart was proposed by Dias,
Requeijo, Pereira, 2009, to monitor the equipment
vibration of repairable systems. Given the lack of data,
that variable was not used in this paper; however, we
recommend its use in the future since we believe it
can contribute valuably to quality control of equipment
operation.
2.2 Special charts
Condition control sensitivity can increase using the
accumulation of information from the samples,
employing CUSUM and EWMA (Exponentially
Weighted Moving Average) control charts.
2.3 Modified CUSUM chart
CUSUM charts, applied since the 50s, have a graphical form of control that can represent the performance
of the process, providing estimations of increases and
decreases for the values previously recorded. This
chart is regarded as a memory chart as opposed to
traditional charts, because it uses statistics for the
whole history of the process until present moment,
Dias (2009).
In this article we intend to detect the moment when
the system reaches a value of oil pressure that can
affect its future behaviour. Therefore the construction of the modified CUSUM chart will be used for
a minimum cumulative sum, T, because it allows the
128
detection of the decrease of the mean and the time t

when this occurred. This chart is defined by:
Where:
Where
In expressions 6 to 10, X is the sample mean at time t,

TL the minimum allowable oil pressure, is standard
deviation, n the sample size, the variable Zt is consider
to be the reduced variable corresponding to X , k is the
reference value expressed as a function of the variation
or . The operation of the equipment is considered
out of statistical control when the following condition
is verified: T < h. Here, h is the decision interval for
CUSUM control charts, Pereira and Requeijo (2008).
The value of control limits can be read from the
abacus presented by Pereira and Requeijo (2008). For
the pair (ARLIn Control , k), we get a h value that is an
adaptation of Gan (1991). The rules that define the
equipment or components needing an intervention are
discussed later in this paper.
2.4 Modified EWMA chart
The EWMA chart is also a sensitive chart to small
changes.
In the modified EWMA chart, the parameter studied is the average of the process estimated by a sample
average through the variable E, which is a weighted
average of all data. The chart is based on negative
deviations from the minimum allowable pressure, TL ,
at time t:
To determine the best values of K and L, depending

on the specified ARLIn Control values and the change in
average = X , an abacus developed by a researcher
in this area, Crowder (1989), is used. The abacus can
be seen on Pereira and Requeijo (2008).
Zou & Tsung (2011) recently reported and showed
that applying the EWMA charts to industrial systems
by transforming the collected data, its modelling is
facilitated. And with application of a multivariate chart
we can determine abnormal behaviour in the systems
in time and improve their performance.
The propulsion system of the ship used in this study

is a CODOG (Combined Diesel or Gas) system. We
decided to study only data from Diesel propulsion
engines. For definite conclusions about common and
special causes of variation in the average, we would
have to consider the influence of other components
of the propulsion system (Gas Turbines for example),
weather and sea condition.
The sensors from NAUTOS we are going to consider are: OP0155 (LUB OIL PRESSURE DE1) and
OP0157 (DE1 PISTON COOLING OIL PRESS).
Note that for the data measured by the OP0155 and
OP0157, the lower limit alarm set by the manufacturer is 4.8 bar and the limit at which the engine
performs an emergency stop is at 3.8 bar oil pressure.
The data collected from OP0155 OP0157 correspond
to variables we name Var1 and Var2 in what follows.
For the CUSUM and EWMA charts we present here
only the results for Var1 are shown, the procedure for
var2 being similar.
4
where 0 < 1.
By applying this technique we consider only the
negative deviations for the minimum allowable pressure (TL ). If there is no alarm limit set by the manufacturer; this value can be estimated through TL
(Dias et al., 2009).
The chart is built with the ordered pairs (t, E ),
considering that there is a situation of out of control,
reflected in a change in the average process, when E
does not belong to the range defined by the limits of
control. The limits of control are defined by:
CONTEXT OF EQUIPMENT ANALYSIS
PROPOSED METHODOLOGY
The methodology proposed here is based on the data

collection of a selected equipment operation and subsequent application of statistical process control, via
control charts of the standard deviation, the CUSUM
charts and EWMA charts, Figure 3.
The methodology:
129
Collect operating data.

Build Shewhart control chart for the standard
deviation.
Monitor the equipment to see if it operates in a stable
mode.
Application of CUSUM and EWMA charts to study
selected data.
Figure 5. S Control Chart for Var2.
Figure 6. CUSUM charts for Var1.
Figure 3. Methodology.
of the engine was also verified, with UCLs = 0,463,

CLs = 0.222 and LCLs = 0, Figure 5.
5.2 Application of modified CUSUM controls
charts
Figure 4. S Control Chart for Var1.
Definition of the limits of the data, according to the

limits set by the manufacturer or by International
Standardization or others.
Establish rules for fault detection and identification
of the need for an intervention:
a. Undertake an intervention to find an anomaly
due to a changing process whenever there are two
consecutive data points below the Lower Control
(LCL).
b. Undertake an intervention to find an anomaly due
to a changing process, whenever four consecutive
data points occur below the alert level (AL).
5
5.1
DIESEL ENGINE APLICATION

Standard deviation control chart application
50 samples are considered (m), each with a size of

5 (n), although there are 1020 individual observations
that for analysis. Figure 4 presents only a part of the
chart.
So it is possible to estimate
= 5.804 and
= 0.222
for Var1. Applying the S chart to the samples considered, we observed the stability of the selected data of an
engine for Var1, as depicted below. Here control limits were calculated using expressions 3 to 5, obtaining
UCLs = 0.436, CLs = 0.209 and LCLs = 0.
For Var2, the estimated average and standard deviation are respectively:
= 5.763 and
= 0.236. For
the S chart the stability of the selected parameter
With the limits of such alarm and emergency stop

engine, set by the manufacturer, the value TL = 4.8
was use. It is considered important to detect the
oil pressure on = 0.5, = 1.12, which corresponds to a reference value of k = 0.56. The values = 1% (ARL = 100) in the definition of AL
and = 0.2% (ARL = 500) in the definition of the
UCL were used. Using the abacus of Gan, Dias
(2009), and with k = 0.56, ARLIn Control = 100 and with
ARLIn Control = 500 respectively we will got AL = 3.2
e UCL = 4.6. The T values are calculated using the
formula (6).
Analyzing the CUSUM chart for Var1, using the
observed data, the average change is considered to be
under control condition.
To validate the proposed methodology we performed a simulation with lower values of the oil
pressure data, which correspond to anomalies in lubricant Fluid. Figure 6 presents this simulated situation
where we can see that the charts are effective enough
to detect changes of equipment condition.
The values were simulated by artificially lowering
the pressure values in the samples No. 40, 68 to 73,
88, 95 to 98, 116, 134-135 and 152. We confirm that
this chart is sensitive to variations. It would then
be possible to monitor the equipment and to define
possible interventions according to the defined rules.
Analyzing the above chart it is clear that there are
four consecutive points (7073) below the alarm so
that, according to defined rules, the engine should
have an intervention in order to determine the cause
of this drop in pressure. Points 97 to 98 do not violate any of the rules, do not reveal any abnormality.
However, if there is no action after the first occurrence, it is expected that the system is degraded, and
one or two of the previously defined rules can occur.
130
With this method we can, hopefully, realize a CBM

system that can reduce the number of failures, reduce
costs and make a definite commitment to continuous
improvement in equipment performance.
7
It is suggested that this methodology should be applied

to other selected equipment for monitoring condition. We believe that, in future, it will be possible to
combine data from vibrations with the data provided
by the automatic control of the platform, in order to
obtain early detection of 100% of possible faults.
Set rules for the collection of operating data
systematically.
Update the rules in accordance with the results
and knowledge of equipment and its operational
requirements.
Develop EWMA and CUSUM charts, adapting
them to the requirements of specific data.
Figure 7. EWMA chat for Var1.
5.3 Application of modified EWMA control chart

Just as in CUSUM charts, we assumed TL = 4.8. It
is considered important to detect the oil pressure for
= 0.5, = 1.12. Setting, = 1% (ARL = 100) in
the definition of AL and = 0.2% (ARL = 500) in
the definition of UCL, using the abacus Crowder
(Dias et al, 2009), we obtained = 0.24 and = 0.17,
respectively, and K = 2.35 and L = 2.95. The values
of AL and UCL are calculated using the formulas 12
and 13 obtaining AL and UCL = 0.457 = 0.6979.
E-values are calculated using the formula 11.
In the representation of the modified EWMA chart
for Var1, we do not detect any values out of control,
so the process is under statistical control.
However, as for the modified CUSUM chart we
wanted to test its sensitivity, Fig. 7. The sensitivity
simulation, only with the reducing of pressure values
of CUSUM charts were not meaningful, so we reduced
the values again to EWMA Modified chart.
So we identified a rule from the methodology that
sets a possible maintenance to the engine.This happens
because the samples No. 70, 71, 73 are below the limit
of control. For points No. 40, 95, 97 and 98 would
not be necessary to take action once that satisfied the
established rules.
CONCLUSIONS
Throughout this article we named the equipment

monitoring system as SPC; however it seems that,
given the context, the concept of Statistical Equipment
Control (SEC) is more appropriate. That concept refers
the application of quality control charts to mechanical
systems, but it should be extended to other kind of
systems or equipments.
When applying the standard deviation control chart
to Var1 and Var2, we found that they were under control for the estimated parameters of the equipment
operation.
Using modified EWMA and CUSUM charts for
statistical control of Var1 showed that it is possible to
do such monitoring and to act preventively, solving the
CBM problem in this context.
Synthetically: it is believed that with a good data
collection system for the engine and with a correct
implementation of this method, we can perform, in
a continuous base, the statistical monitoring of the
equipment operation.
RECOMMENDATIONS
REFERENCES
Dias, J. A. M. (2002). Fiabilidade em Redes de Distribuio
de Energia Elctrica. Tese de Doutoramento, FCT-UNL,
Lisboa.
Dias, J.M., Requeijo, J.G. and Pereira, Z.L (2009). Monitorizao do Estado de Condio dos Sistemas Reparveis,
em Riscos Industriais e Emergentes, Volume 1, C. Guedes
Soares, A. P. Teixeira e P. Anto (Eds), Edies
Salamandra, Lisboa, 2009 (ISBN: 978-972-689-233-5),
pp. 501511
Eyvazian, M., Naini, S. and Vaghefi, A. (2008). Monitoring
process variability using exponentially weighted moving
sample variance control charts. International Journal of
Advanced Manufacturing Technology.
Gan, F.F. (1991). An Optimal Design of CUSUM Quality
Control Charts, Journal of Quality Technology, vol. 23,
pp. 279286.
Marinha, E. (2010). PEETNA2091: MANUTENO. Alfeite.
Montgomery, D.C., (2005). Introduction to Statistical Quality
Control, 5th Edition, John Wiley & Sons, New York.
Parreira, R., Lobo, V. and Vairinhos, V., (2006). Anlise de
parmetros de operao de mquinas martimas, Escola
Naval.
Pereira, Z.L. and Requeijo, J.G., (2008). Qualidade: Planeamento e Controlo Estatstico de Processos, Caparica,
FCT-UNL, Prefcio.
Requeijo, J.G., (2010). Monitorizao e Melhoria dos Processos Usando Metodologias Estatsticas de Excelncia,
Seminrio Data Mining, ISLA.
Sibanda, T., (2007). The CUSUM chart method as a
tool for continuous monitoring of clinical outcomes
using routinely collected data, Bmc Medical Research
Methodology.
Zou, C.L., (2010). Likelihood Ratio-Based Distribution-Free
EWMA Control Charts, Journal of Quality Technology.
Zou, C. and Tsung, F. (2011). A Multivariate Sign EWMA
Control Chart. Technometrics, 8497.
Wu, Z., Jiao, J., Yang, M., Liu, Y. and Wang, Z. (2009).
An enhanced adaptive CUSUM control chart. Iie Transactions, 642653.
131

A nonlinear control approach for a vessel steering system

Lokukaluge P. Perera & C. Guedes Soares
ABSTRACT: This paper proposes a nonlinear control approach for a vessel steering system. A nonlinear vessel steering system is derived considering the second order linear Nomoto model. The sliding mode approach
is proposed as the nonlinear controller. The sliding mode controller is able to deal with parameter uncertainties
and unmodeled dynamics, it quantifies the model response and the performance tradeoffs, and it rejects external
disturbances in vessel steering. The stability conditions of the sliding mode controller are analyzed considering
a Lyapunov candidate function. Finally, the proposed control algorithm is simulated and successful results are
reported for a nonlinear vessel steering system in this study.
INTRODUCTION
The developments of computational technology, satellite communication, and electronic devices (i.e. sensors and actuators) have facilitated the introduction
of autopilot features into modern ship navigation systems. However, most of the autopilot systems in ship
navigation are mainly formulated to follow a desired
course under constant speed conditions. Therefore, to
facilitate present ship navigation requirements various
linear and nonlinear controllers have been proposed
in recent literature. However, most of the linear controllers are based on the assumptions of linear state and
parameter conditions that may not always be realistic
under various maneuvering conditions.
Furthermore, under the assumptions of nonlinear
state and parameter conditions, several nonlinear control approaches are also proposed in recent literature.
However, the nonlinear controllers that are based on
accurate mathematical models of ocean going vessels
have not been successful because those models are still
inadequate.
Therefore, the model based nonlinear controllers
always had performance deficiencies in experimental
implementations. Furthermore, the nonlinear controllers that are based on the assumptions of deterministic state and parameter conditions in vessel steering
may not always be realistic because the state and
parameter conditions are changing under various sea
conditions and the controllers would not be able to
preserve their robustness.
Therefore, a nonlinear controller that does not
require an accurate mathematical model of an ocean
going vessel and is robust against disturbances; is
proposed in this study. Furthermore, the proposed
system model considers the nonlinear maneuvering
conditions of an ocean going vessel.
The sliding mode control approach, one of the nonlinear control approaches, to control nonlinear vessel
steering is adopted in this study. The selection of the
sliding mode controller is based on several reasons
(Slotine and Li 1991): The first is that the sliding mode
control law can include system parameter uncertainties and un-modeled dynamics; the second is that it can
quantify the response of modeling and performance
trade-offs; the third, it can reject bounded external
disturbances.
1.1
Sliding mode control
There are many control approaches of ocean going

vessels that can be observed in recent literature. The
sliding mode controller for ship course control is proposed and simulation results are illustrated by Tomera
(2010). Yaozhen et al. (2010) presented fuzzy sliding model controller for ship course control system.
However, these studies are limited to linearized vessel
steering models.
The terminal sliding mode fuzzy control approach
based on multiple sliding surfaces for a nonlinear ship
autopilot system is proposed by Yuan and Wu (2010).
Similarly, pi-type sliding mode controller is proposed
for tracking control of a ship by Yu and Wu (2004).
However, the rudder rate effects on the vessel steering
are ignored on these studies.
The theoretical and experimental results for path
control of a surface ship in restricted waters are presented by Zhang et al. (2000). The control approach is
based on input-output linearization and sliding mode
controller. However, the study is limited to course
keeping maneuvers.
Several sliding mode control approaches in different applications are presented in the studies of Kizmaz
et al. (2010), Wnag (2006), Qiang et al. (2008), Yeh
133
Figure 1. The proposed control System.
and Huang (2010), Rao et al. (2006), and Lauvdal and

Fossen (1996).
1.2
Figure 2. The coordinate system for vessel steering.
refers by Davidson and Schiff (1946) and can be

written as:
Control system
The proposed control system is presented in Figure 1. The system consists of three modules: vessel
steering system, sliding mode controller and rudder
control system. A nonlinear mathematical formulation
is presented in the vessel steering system. A control
algorithm is presented in the sliding mode controller.
The rudder control system consists of a mathematical
model for rudder actuation that acts upon the rudder angle and the rudder rate. Further details on these
systems are presented in following sections.
The organization of this paper is as follows. Section
2 contains a vessel steering system that is considered
for this study. Section 3 contains a formulation of the
sliding mode controller and its stability conditions.
Section 4 contains a detailed description of the computational results. Finally the conclusion is presented
in section 5.
where the system parameters are m: mass of the vessel,

u0 : surge velocity, xG : distance to the center of gravity, R : rudder angle and Iz : inertia of the vessel alone
the z-axis. The respective hydrodynamic forces can be
written as:
where Yv , Yr , Y , Yv , Yr are the hydrodynamic coefficients of sway motion and Nv , Nr , N , Nv , Nr are the
hydrodynamic coefficients of yaw motion.
The state space model describing the vessel linear steering system, introduced in equation (1) can be
re-written as:
2 VESSEL STEERING SYSTEM

A general mathematical model for ocean going vessel
steering is discussed in this section and the respective
coordinate systems are presented in Figure 2. The vessel steering system is associated with two coordinate
systems: Xn Yn Zn : the earth fixed coordinate system
and Xb Yb Zb : the vessel body fixed coordinate system.
Several simplified mathematical models have been
proposed in the recent literature based on vessel kinematics and dynamics: Surge model (u), maneuvering
model (v, r), horizontal motion model (u, v, r), longitudinal motion model (u, w, q) and lateral motion model
(v, p, r); where u: surge linear velocity, v: sway linear
velocity, w: heave linear velocity, p: roll angular velocity, q: pitch angular velocity, r: yaw angular velocity,
X : surge force, Y : sway force, Z: heave force, K: roll
moment, M : pitch moment, and N : yaw moment (see
Figure 2).
where the system states are = [v r]T , and the

system matrices MR , NR (u0 ), and BR are:
Because the matrix, MR , is positive definite, the vessel

linear steering system presented in Equation (3) can be
further written as:
where the system matrices A and B of Equation (5) are:

2.1
Sway-yaw sub-systems
Assuming a constant linear surge velocity the sway

and yaw sub-systems can be coupled. This approach is
categorized as the vessel linear steering model that
134
where the initial parameters d1 = 1/T1 + 1/T2 ,

d2 = KR /T1 T2 and d3 = T3 , are straightforward to be
defined. Hence, Equation (12) can be rewritten as:
and the respective coefficients are:
where the final parameters can be defined as:

1 = a2 d2 , 2 = a1 d2 , 3 = d1 , 1 = d2 , and
2 = d2 d3 . This nonlinear vessel steering model in
Equation (15) has been used in this study as the system
model for controller design.
3
SLIDING MODE CONTROLLER
3.1 Dynamic systems model

2.2 The second-order Nomoto model
The second-order linear Nomoto model (1957), can
be derived by eliminating the sway velocity, v, from
Equations (5) resulting in:
where the coefficients are:
= x2 ,
where the system states are defined as: = x1 ,
= x3 and the system functions can be written:
and
The second-order linear Nomoto model in Equation

(8), can be written considering the heading angle, ,
of the vessel:
Some features of this model have been analyzed by

Sutulo and Guedes Soares (2005) for directionally
unstable ships.
2.3 The modified Nomoto model
The second-order Nomoto model is often adequate for
course keeping maneuvers, but this model may not
sufficient for course changing maneuvers because it
does not represent the nonlinear vessel steering conditions. Therefore, the model presented in Equation (10)
is modified to include nonlinear steering conditions, as
is
proposed in Amerongen and Cate (1975), where
and the nonreplaced by a nonlinear function KR H ()
= a1
+ a2
3 . Thus,
linear function is given by H ()
equation (10) can be written as:
Equation (11) can then be written as:
The controller design process for the nonlinear vessel

steering system is discussed in this section. To derive
the sliding mode controller, the mathematical model
presented in Equation (13) can be written as:
where ux : the control input to the rudder control system. The control input is defined as the input voltage/
current to the rudder control system and 1 = 0 and
2 = 0 are parameters of the rudder control system.
3.2 Sliding mode controller
In the sliding mode control approach, a feedback control law that constrains the motions of vessel steering
to a sliding manifold (i.e. surface) is defined. The
sliding manifold is a hyper surface that is inserted
in the state space of vessel steering. One should note
that the sliding manifold is independent of the vessel
steering system structure, therefore sliding mode control inherently robust to parameter uncertainties and
un-modeled dynamics. Therefore, the nonlinearities
associated with vessel steering near the zero rudder
angle can be reduced. This phenomenon can also be
seen as a model reduction in vessel steering due to the
sliding manifold.
The vessel steering system behavior under sliding
mode control can be divided into two phases: reaching phase and sliding phase. In the reaching phase, the
vessel steering trajectory starting at the initial conditions moves toward the sliding manifold. In the sliding
phase, the trajectory moves toward the origin of the
equilibrium point.
In the sliding phase, after the system trajectory
reached the sliding manifold, the dynamics of the
135
system is constrained to a reduced-order model of

the system. Therefore, the system nonlinearities associated with vessel steering can be reduced by the
manifold during the sliding phase. The sliding manifold, s = 0, that constrains the motions of the system
can be written as:
where the constants a1 > 0 and a2 > 0 and the respective errors e1 , e2 and e3 can be written as:
and the vessel reference heading is presented by

xr = r . The motion on the manifold, s = 0, satisfies
stability conditions and the differentiation of Equation
(16) leads to:
Hence, the feedback control law can be written as:
where 0 is a constant. One should note that the equation (24) achieves the negative semidefinite condition
under the feedback control law in (26). Therefore,
the proposed feedback controller law in (26) guarantees the robustness of the controller even under the
uncertainties in the functions of f (x) and g(x).
There are some practical challenges with the sliding
mode controller; the control inputs can be affected by
chattering, which can lead to high temperature generation, high wear of actuator moving parts, or can
excite unmodeled high frequency dynamics, eventually degrades the controller performance. However,
adjustments of the inclination of the sliding surface
and the function, (x), can overcome some of those
challenges.
3.3 Rudder control system
Applying Equations (14) and (17) into (18) results in:
where the following vessel state conditions are

assumed:
Considering a Lyapunov candidate function:
It is not possible to control rudder angle and rudder rate simultaneously; therefore the input voltage/
current of the rudder control system is considered to
control vessel steering in this study. However, one
should note that the accurate control of the rudder
angle and the rudder rates are not required by this
approach.
its differentiation leads to:
Assuming a known function (x) that satisfies the

conditions of:
One should note that the function, (x), can be

seen as the upper bound of the function, (a1 x2 + a2
x3 + f (x))/g(x), which can be calculated and/or estimated by the system identification techniques (Perera
et al. (2011)). However, an accurate model of the function, (a1 x2 + a2 x3 + f (x))/g(x), is not required in this
approach, where an approximation is adequate for the
control design. Applying, the proposed function, (x),
into Equation (22) can be illustrated as:
where the switching function, sgn(s), is given as:
The main objective of the rudder control system is to

convert the control input, ux , which is generated by
the sliding mode controller into the rudder angle and
the rudder rate. The mathematical model for rudder
control systems can be written as:
COMPUTATIONAL RESULTS
The proposed sliding model controller is implemented on the MATLAB software platform. The
vessel parameters in the steering model are assumed
as 1 = 0.3710 (1/rad2 ), 2 = 0.4340 (1/s2 ),
3 = 3.4000 (1/s), 1 = 0.3500 (1/s3 ), and 2 =
0.1225 (1/s2 ). Some of these parameter values are
extracted from the study of Amerongen and Cate
(1975), and others are generated by trial and error calculations considering the vessel response under stable
response conditions. The sliding surface parameters
are considered as: a1 = 0.8 and a2 = 3. The parameter
of the feedback control law constant is considered as:
0 = 0.75.
The computational simulations of the sliding mode
controller in vessel steering are presented in Figures 3, 4 and 5. Figure 3 consists of the vessel reference heading and the actual heading with the sliding
mode controller. As presented in the figure the vessel
actual heading is converged to the reference heading
136
Figure 5. Zoomed rudder angle and rudder rate responses.
Figure 3. Vessel heading and error conditions.
The zoomed versions of the rudder angle and rate are

presented in Figure 5. As presented in the zoomed
simulations the rudder angle and rudder rate values
are accumulated with small chattering effects that are
well known for the sliding mode controllers.
CONCLUSIONS
This paper introduces a sliding mode control approach

for nonlinear vessel steering. As presented in the
computational results, the controller successfully converges the vessel actual heading into the reference
heading. The successful results obtained in this study
show the control system capabilities to overcome
the challenges faced by vessel steering: parameter
uncertainties and un-modeled dynamics. The main
contribution in this study is that even without accurate knowledge of the nonlinear vessel steering system,
that controller can be implemented.
ACKNOWLEDGEMENTS
Figure 4. Rudder angle and rudder rate responses.
conditions. Furthermore, heading error converges to

zero as presented in the same figure.
Figure 4, consists of the rudder angle and rudder rate values with respect to the controller output.
The first author has been supported by the Doctoral

Fellowship of the Portuguese Foundation for Science
and Technology (Fundao para a Cincia e a Tecnologia) under contract no. SFRH/BD/46270/ 2008.
Furthermore, this work contributes to the project of
self-propelled models, which is being funded by
137
the Portuguese Foundation for Science and Technology (Fundao para a Cincia e a Tecnologia) under
contract no. PTDC/TRA/74332/2006.
REFERENCES
Amerongen, J. V., Cate, A. J. U. T., 1975. Model reference
adaptive autopilots for ship. Automatica 11, 441449.
Astrom, K. J., 1980. Why use adaptive technique for steering large tankers, International Journal Control 32 (4),
689708.
Astrom, K. J., Kalstrom, C. G., 1976. Identification of ship
steering dynamics. Automatica 12, 912.
Davidson, K. S. M., Schiff, L. I., 1946. Turning and course
keeping qualities. Transactions of Society of Naval
Architectures Marine Engineers 55.
Khalil, H. K., 2002. Nonlinear Systems. Prentice Hall,
New Jersey, USA.
Kizmaz, H., Aksoy, S., Muhurcu, A., 2010. Sliding mode
control of suspended pendulum. In: Modern Electric
Systems. Wroxlaw, Poland, pp. 1419.
Lauvdal, T., Fossen, T. I., 1996. Nonlinear non-minimum
phase rudder-roll damping systems for ships using
sliding mode control. In: Proceedings of the IFAC
World Congress, San Francisco, USA.
Nomoto, K., Taguchi, T., Honda, K. and Hirano, S., 1957. On
the Steering Qualities of Ships, International Shipbuilding
Progress 4, 354370.
Perera, L.P., Oliveira, P., and Guedes Soares, C., 2011.
Dynamic Parameter Estimation of a Nonlinear Vessel
Steering Model of Ocean Navigation., In Proc. 30th
Int. Conf. on Ocean, Offshore and Arctic Engineering,
Rotterdam, The Netherlands, ASME paper OMAE201150249.
Qiang, W., Juan, C. and Zhiyong, T., 2008. Study of sliding
mode control for stewart platform based on simplified dynamics. In: Proceedings of IEEE International
Conference on Industrial Informatics. Daejeon, Korea,
899892.
Rao, S., Brandtstadter, H., Buss, M. and Utkin, V., 2006.

Sliding mode control in mechanical systems with electric actuators. In: Proceedings of International Workshop
on Variable Structure Systems, pp. 14.
Slotine, J. J. E., Li, W., 1991. Applied Nonlinear Control.
Prentice Hall, New Jersey, USA.
Sutulo, S., and Guedes Soares, C., 2005. Numerical study
of some properties of generic mathematical models of
directionally unstable ships. Ocean Engineering 32 (34),
485497.
Tomera, M., 2010. Nonlinear controller design of a ship
autopilot. Internal Journal of Applied Mathematics and
Computer Science 20 (2), 271280.
Triantafyllou, M. S., Hover, F. S., 2003. Maneuvering
and Control of Marine Vehicles. MIT Cambridge,
Massachusetts, USA.
Wang, W., 2006. Adaptive fuzzy sliding mode control
for inverted pendulum. In: Proceedings of the Second
Symposium International Computer Science and Computational Technology. Huangshan, China, pp. 231234.
Yaozhen, H., Hairong, X., Weigang, P. and Changshun, W.,
2010. A fuzzy sliding mode controller and its application
on ship course control. In Proceedings 7th International
Conference on Fuzzy Systems and Knowledge Discovery.
pp. 635638.
Yeh, F. K., Huang, J. J., 2010. Fuzzy sliding mode control for
a mini AUV. In: Proceedings of SICE Annual Conference.
pp. 33173323.
Yu, K. W., Wu, C. E., 2004. Tracking control of a ship by
pi-type sliding controller. Journal of Marine Science and
Technology 12(3), 183188.
Yuan, L., Wu, H. S., 2010. Terminal sliding mode fuzzy
control based on multiple sliding surface for nonlinear
ship autopilot systems. J. Marine Sci. Appl. 9, 425430.
Zhang, R., Chen, Y., Sun, Z., Sun, F., and Xu, H., 2000. Path
control of a surface ship in restricted waters using sliding
mode. IEEE Transactions on Control Systems Technology
8 (4), 722732.
138
Onboard systems

Broadband communications with adhoc networks case study

with simulated scenarios
Fernando Correia
CINAV/Inesc-ID, Alfeite, Portugal
Jorge Santos & Paulo Mesquita

Escola Naval/CINAV, Alfeite, Portugal
ABSTRACT: The design of a naval communication system can be done using the Mobile Ad hoc Network
working concept (MANET), enhanced with the capability to communicate with the command and control centres
ashore.Actually the Portuguese Navy has the capability to communicate between the land based command centres
and the ships, and between ships. However, the communication technology used in ship-to-ship link set bounds
to the services which are required for the command and control procedures. The use of a naval MANET enhanced
with the capacity to link the ships at sea with the command and control centres ashore will extend the network
services to a greater number of ships. This document presents a case study about the usability of a MANET
by a naval force. In this study was considered different classes of routing protocols and several models for the
Transmission Control Protocol (TCP). In this scope, it is done a performance comparison between the routing
protocols with the TCP models under study. The results were achieved through simulation using different naval
scenarios.
INTRODUCTION
Taking into account the advances in the computing and

wireless technologies, the Mobile Ad hoc networks
(MANETS) have grown as a part of the computational
environment. This fact has been a start point to develop
new communication systems dedicated to transport
specific data formats. Thus, the data rate transmission
increases with the technology evolution.
The use of wireless technology allows on a simple
and fast way to install and deploy a dynamic variable
size mobile network.
In this scope, we can include tactical communication networks. A tactical communication network
has assigned several mobile units, where the link
bandwidth between network nodes is related with
the communication technology used (HF, VHF, UHF,
Wi-Fi, SATCOM). At military operation, there are
several domains where these kinds of networks can
be used, like the deployment of a naval force or a
marine land force.
Considering the data transmission in a naval communication environment, the use of satellite communication equipment combined with a MANET,
supported on UHF or WiFi (802) technology, presents
the advantage to bring together the simplicity to
create a broadband tactical wireless network with
Quality of Service (QoS) link connection to the command and control centers ashore, independently the
network size.
This paper is organized as follows: in Section 2 are

presented some operational requirements and technological solutions used in the Portuguese Navy; the
Section 3 presents a scenario for naval mobile communications with which were done several simulations;
finally, in Sec.4 are presented some conclusions.
2
2.1
OPERATIONAL REQUIREMENTS
SATCOM satellite communications
In the last years, the Portuguese Navy has shown an

evolution in naval communications with the use of
satellite communication system SCOT 3P. This evolution comes from operational requirements associated
to command and control capabilities necessary for a
ship to be the naval force flagship.
Since 1994, when the first SATCOM system was
installed, the importance of this kind of communications and the services available for this physical
communication support has been increased. This situation can be observed through the constant increase of
the bandwidth assigned to the communication channel ship-shore-ship, which increased from 64 Kbps in
1995, to 1 Mbps in 2009.
Actually, the SATCOM system allows the use a set
of communication services. These services are available on board of Portuguese Navy frigates through
the SATCOM communication system. These services
141
share the same communication link using multiplexing

equipments.
The services can be divided in two major categories
in accordance with the data traffic nature:
Constant Bit Rate services (CBR): voice

Available Bit Rate services (ABR): web, mail,
NSWAN, etc.
The voice service uses a direct connection to the

PSTN network located ashore. The terminal equipment is located on board of the FFGH ships.
The voice communication is a service that requires
Quality of Service (QoS) through an effective bandwidth reservation. This service is sensitive to network
traffic latency and should present a high priority in
the network routing equipment.
The military message handling system is also
a service that requires QoS. The required network
resource reservation is related with the cipher
machines synchronism alignment procedure.This way,
the terminal equipment located on board of the FFGH
ships can be connected to the MMHS network through
dedicated communication channels.
The other services that use the SATCOM channel share the remaining bandwidth between them. For
each service is assigned a minimum bandwidth value
to ensure a correct working mode. It is used the TCP/IP
stack to transfer data traffic in the network. A network
that supports TCP/IP, its channel transport capacity is
shared by all the applications and services provided
there are sufficient resources, in accordance with the
defined traffic management rules. These rules can
present a static or dynamic behaviour. In the static
mode, each service has a bandwidth value assigned.
This resource can only be used by that service. In
the dynamic mode, when the resources reserved are
not being used by the service, they are shared by the
other active services. With this procedure, the network
transmission resources can be optimized.
In a network with limited resources, there must be
traffic priority management procedures. As the network load increases, the network traffic latency also
increases, and as consequence, the time for a user to
access information also increases.
In the actual system configuration, the connection
to the navy wide area network (NavyWAN) through
SATCOM is peer-to-peer type. This connection is for
the ships exclusive use. Due this, if there is a naval
force at sea, only the ships with SATCOM capacity
may have broadband communications and access to
the services available in the NavyWAN.
2.2
Mobile ad hoc networks
In the mobile communications, the wireless networks

are an emergent technology. The wireless networks
can be personnel, like the Wireless Personnel Area
Network (WPAN), or wide and for generic use, like
the MANET (Macker and Carson, 2009; Ding, 2008;
Pullin et al., 2008), or for more specific application,
the Vehicular Ad hoc Networks (VANET) (Saha and
Figure 1. MANET topology.
Johnson, 2004) which is a specific type of MANET

used for inter-vehicle communication.
The MANET is a self-configurable network which
doesnt require a physical communication infrastructure to connect all terminal communication
equipments.
Each network node is simultaneous terminal
equipment and a router (Gupta and Kumar, 2000; Li
et al., 2001; Toh, 2002). Due the mobile nature of the
network nodes, the architecture for the routing traffic
procedures must be able to adapt themselves to the
constant change of the network topology.
The MANET coverage area depends of the number
of network nodes, their geographical position, their
transmission power and the radio reception sensibility.
Figure 2 presents two possible topologies that can be
observed in a MANET.
In Figure 1a, the network nodes are all concentrated in the same geographical zone. In this situation,
every node can communicate directly with any other
node. However, due to the network nodes mobility, the
coverage area tends to increase as shown in Figure 1b.
In this scenario, although node A has no direct radio
link with node F, they communicate using the other
network nodes.
The MANET has a self-configuration property
and can optimize the network available resources
through the use of different communication layers
(hierarchies), as presented in Figure 2.
2.3 Network routing
The traffic is routed in accordance with the technology
used to construct the network. The SATCOM, UHF or
142
Figure 5. Future communications concept.
The transmission capacity associated to the

SATCOM link must be adequate to accommodate
different services available in the NavyWAN.
STUDY OF SIMULATED SCENARIOS
Figure 2. MANET with two hierarchies.
Figure 3. Actual concept for satellite communications.
Figure 4. MANET communications concept.
WiFi technologies have different modes to route the

traffic.
SATCOM has a peer-to-peer working mode. The
data transfer is done through a dedicated link between
the ship and the ashore station. Only the ships equipped
with SATCOM systems are able to connect to the
NavyWAN, as shown in Figure 3.
However, a network working in ad hoc mode allows
that every node (ship) to establish communications
with any other nodes, independently of their geographic position (see Figure 4). Due to the radio
transmission range, a connection to the NavyWAN
requires at least one node with ashore connection.
Thus, one way to optimize the communication
resources, is to use together both SATCOM and
MANET technologies to create one mobile network.
This way, it is possible to connect all the ships to the
NavyWAN through one ship-shore connection (see
Figure 5).
The wireless transmission technologies are available

in different standards, in which working characteristics can be use in different scenarios. Given the
characteristics for each technology, it is possible to
verify a boundary overlap between some of them.
The use of SATCOM and MANET communication systems gives connectivity to variable size
geographic areas (with UHF or even with WiFi) with
connectivity to ashore services (NavyWAN).
In this scope, it can be possible to identify several
scenarios where a mobile network of this type can
be used. The scenarios can include naval forces and
land forces supported by naval forces in a humanitarian mission among other missions. Thus, a study on
routing protocols for MANET and on the transport
protocols was conducted, to evaluate the network performance to transport specific data formats currently
used in the NavyWAN. The network simulator NS2
version 2.31 (ns2, 2007) was used. For each proposed
test, 30 simulations were performed. The 802.11b
standard at the Medium Access Control (MAC) layer
was considered and the radios were simulated with
the Two Ray Ground propagation model (Bertoni,
1999). With the specifications, a 200 meters link range
with a transmission rate of 2Mbps was achieved. In
wireless communications, finding the path differences
between the line-of-sight propagation and the nonline-of-sight propagation is essential. The Two Ray
Ground model considers both the direct path and the
indirect path. In (Rappaport, 1996), it is shown this is
an accurate free space propagation model.
This study was carried out in two phases. Each
one was related with routing protocols and the transport protocols: the User Datagram Protocol (UDP)
(Postel, 1980) and the Transmission Control Protocol (TCP) (Todorovic and Lopez-Benitez, 2006; Adya
et al., 2004: Barakat, 2001).
The aim of each simulation phase was to observe
the routing protocols behaviour considering the two
transport protocols. It was simulated the following
routing protocols: the Adhoc On-demand Distance
Vector (AODV) (Perkins et al., 2003), the Dynamic
Source Rounting (DSR) (Johnson and Maltz, 1996;
Meng and Wu, 2008), the Destination-Sequenced
Distance Vector Protocol (DSDV) (He, 2002), and
143
Figure 6. Network overhead.
Figure 7. Volume of bytes delivered at the destination node.
the Optimized Link State Routing Protocol (OLSR)

(Clausen and Jacquet, 2003). These protocols were
chose because they can be representative of two major
routing classes: pro-active protocols like the DSDV
and the OLSR, and on-demand protocols like the
AODV and the DSR.
To evaluate performance of each routing and transport protocol pair, the overhead generated by the
routing protocols and the volume of data delivered
at the destination node through the metric goodput
were measured.
3.1
UDP protocol behaviour
The UDP is a simple connectionless transport protocol.

It cannot realize the make traffic flow control. So, in
the presence of a broken link or a congestion situation,
the source node continues to generate traffic until it
receives an explicit indication that the traffic cannot
reach the destination node. The simulation results are
presented in Figure 6 and Figure 7.
Figure 7 represents the overhead present in the network due the node mobility. It was considered an
average speed of 10 Miph. In this figure it is possible
to observe the overhead is related with the routing protocol class. The pro-active routing protocols
DSDV and OLSR generate more overhead than the
reactive protocols as it can be verified for the AODV
and the DSR. In this scenario the reactive protocols
present values lower than 1%.
The data delivery capacity at the destination node
is related with the network overhead. As the overhead
is lower, more network resources became available for
data transfer. Thus, the on-demand routing protocols
present a better behaviour than the pro-active routing
protocols, as it can be seen in Figure 8. In this scenario,
the AODV is the protocol with the best performance.
It can deliver data at the destination node with a
network rate greater than 100 kbps.
3.2 TCP protocol behaviour
The performance evaluation for the TCP protocol took
in consideration different TCP models. This evaluation
procedure not only allowed verifying which routing
protocol presented the best performance, but also,
what was the best TCP model. Figure 8 presents the
simulation result.
Figure 8. Performance of the TCP models.
Once again, it is possible to observe the on-demand

routing protocols presenting the best performance.
The performance for the AODV is similar to the performance for the DSR, independent of the TCP model
that was used. However, the TCP models can affect the
performance of the pro-active routing protocols.
CONCLUSIONS
Integration of different wireless communication

technologies allows the construction of a variable
dimension broadband wireless network MANET.
A MANET is easy to deploy and can be can be
tailored to different scenarios that fall within the
missions of the Portuguese Navy.
The use of SATCOM systems, enhanced with
MANET networks, ensures optimization of the
resources and gives access to a set of services that
are available in the navy WAN. These services can be
accessed not only by the ships with SATCOM systems
but also by the other ships within the naval group.
Due the results obtained through simulation, it is
possible to conclude that on-demand protocols routing
class are a better solution than the pro-active class of
routing protocols.
Considering the services available in the Portuguese
Navy INTRANET, the network performance results
observed with simulation indicate that it is possible to create a naval communications network using
MANET.
144
A MANET to be used in the same scenarios

that has been simulated is currently being created at
the Naval Academy using commercial off the shelf
equipment.
REFERENCES
Homepage of IEEE 802.11, http://www.ieee802.org/11.
Ns-2 simulator. http://nsnam.isi.edu/nsnam, 2007.
Adya, A., Bahl, P., Padhye, J., Wolman, A. and Zhou, L.
A multi-radio unification protocol for ieee 802.11 wireless networks. In Broadband Networks, 2004. BroadNets
2004. Proceedings. First International Conference on,
p. 344354.
Barakat, C. TCP/IP modeling and validation. IEEE Network,
15(3):3847, May, 2001.
Bertoni, H.L. Radio Propagation for Modern Wireless
Systems. Prentice Hall Professional Technical Reference,
1999.
Clausen, T. and Jacquet, P. Optimized Link State Routing
Protocol (OLSR) - RFC3626. IETF Network Working
Group, October 2003.
Ding, S. A survey on integrating manets with the
internet: Challenges and designs. Comput. Commun.,
31(14):35373551, 2008.
Gupta, P. and Kumar, P. The capacity of wireless networks. Information Theory, IEEE Transactions on, 46(2):
388404, mar 2000.
He, G. Destination-sequenced distance vector (DSDV)
protocol, May 06 2002. http://citeseer.ist.psu.edu/531710.
html; http:// keskus.hut.fi/opetus/s38030/ k02/Papers/
03-Guoyou.pdf.
Johnson, D.B. and Maltz, D.A. Dynamic source routing in
ad hoc wireless networks. In Mobile Computing, pages
153181. Kluwer Academic Publishers, 1996.
Li, J., Blake, C., De Couto, D.S., Lee, H.I. and Morris,
R. Capacity of ad hoc wireless networks. In MobiCom
01: Proceedings of the 7th annual international conference on Mobile computing and networking, pages 6169,
New York, NY, USA, 2001. ACM. ISBN 1-58113-422-3.
doi: http://doi.acm.org/10.1145/381677.381684.
Macker, J. and Corsen, S. IETF Mobile adhoc networks
MANET working group charter. http://www.ietf.org/
html.charters/manet-charter.html, 2009.
Meng, L. and Wu, W. Dynamic source routing protocol
based on link stability arithmetic. In Information Science
and Engieering, 2008. ISISE 08. International Symposium on, volume 2, pages 730733, Dec. 2008. doi:
10.1109/ISISE.2008.168.
Perkins, C.E., Royer, E. and Das, S.R. Ad hoc On-Demand
Distance Vector (AODV) Routing RFC3561. IETF
NetworkWorking Group, July 2003.
Postel, J. RFC 768: User Datagram Protocol, Aug. 1980.
Pullin, A.J., Presland, S. and Pattinson, C. Using ship
movement in the Irish Sea for MANET evaluation. In
Computer Modeling and Simulation, 2008. EMS 08.
Second UKSIM European Symposium on, pages 394
399, Sept. 2008. doi: 10.1109/EMS.2008.50.
Rappaport, T.S. Wireless Communications Principles and
Practice. Prentice Hall, Upper Saddle River, 1996. ISBN
0-13-375536-3.
Saha, A.K. and Johnson, D.B. Modeling mobility for
vehicular ad-hoc networks. In VANET 04: Proceedings
of the 1st ACM international workshop on Vehicular ad
hoc networks, pages 9192, New York, NY, USA, 2004.
ACM. ISBN 1-58113-922-5. doi: http://doi.acm.org/
10.1145/1023875.1023892.
Todorovic, M. and Lopez-Benitez, N. Efficiency study of
tcp protocols in infrastructured wireless networks. In
Networking and Services, 2006. ICNS 06. International conference on, pages 103103, july 2006. doi:
10.1109/ICNS.2006.42.
Toh, C.-K. Ad Hoc Mobile Wireless Networks: Protocols and
Systems. Prentice-Hall, 2002. ISBN 0-13-007817-4. URL
http://www.phptr.com/ptrbooks/ptr0130078174.html.
145

Bridge officer decision support tool

CINAV Centro de Investigao Naval, Escola Naval, Alfeite, Almada
ABSTRACT: A vessels stability, operation performance and safety depend on her load condition. During the
design phase only a few load conditions are verified and therefore much responsibility relays on who is operating
the vessel and therefore is prone to human error. This paper intends to present a software tool that aims to increase
safety during operation by monitoring load condition, checking stability and strength criteria, and estimating
the vessels performance in regular intervals to provide a clarified scenario. Additionally, if flooding occurs, the
software includes a decision support tool that advises on how to manage fluids in order to increase the vessels
stability depending on the load condition previous to damage.
INTRODUCTION
Handling a ship in transit, while manoeuvring in shallow waters or channels, during operations or in case
of damage, requires a solid knowledge of the platform, her capabilities, professional experience and
specific training. Today, amongst shipping companies
and navies all around the world, crew numbers are
becoming smaller, they are less experienced, and crew
replacements occur more often.
On the other hand, todays technology allows information to be assessed more easily, so such problems
can be minimized if these tools are smartly used. Following this reasoning, this paper presents a Decision
Support Tool (DST) that is currently being developed
for the Bridge Officer as the responsible for the vessel operation, safety of navigation and monitoring the
vessels stability, i.e. how the vessel relates with the
surrounding medium.
The proposed tool is based on the fact that the
weight distribution influences the vessels stability,
structural integrity, performance, and overall safety.
Therefore, by monitoring the vessels load condition, it
makes possible to check her stability criteria and determine up-to-date operational characteristics in frequent
intervals, turning the vessels operation simpler and
less prone to human error.
In case of damage, it becomes even more significant
to have up to date knowledge of the ships load condition since every action done to increase her stability
must have it into account, or her ability to survive can
be drastically reduced.
In the next sections, a small review of the influences taken into consideration to develop this work
are presented, followed by the DST concept analysis
and finally the presentation of its implementation.
STATE OF THE ART
Since the first Safety of Life at Sea convention of

1914 (SOLAS), several deterministic and probabilistic criteria for stability have arisen, such as the the
International Maritime Organization (IMO) code of
1993 IMO (1993), or the new probabilistic approach
of SOLAS 2009 IMO (2009), and also military criteria,
most of them developed from Sarchin and Goldberg
(1962).
During ship design and construction only a few load
conditions are checked against the criteria, since it is
virtually impossible to check all possible load conditions and liquid cargo distributions in the tanks. Therefore, during operation, the responsibility to check
intermediate load conditions is the masters responsibility. For that purpose, stability calculators that
simulate conditions input by the user are available
aboard most cargo vessels
Among these calculators, the United States
Navy produced Flooding Casualty-Control Software
(FCCS) (Plumley, 2009). It is probably one of the most
complete ones, is able to:
evaluate the ship attitude, hydrostatic, stability and
strength status;
model the ships loads;
evaluate the effects of flooding on stability;
evaluate the ships strength in case of flooding or
structural damage;
send out warnings and advise if stability or strength
criteria fail;
evaluate stability and bottom reaction forces
during vessel stranding.
Nevertheless, the safety of the ships operation is
not only concern with stability, but also depends on
147
how the vessel is operated by the crew (House, 2007),

and on her manoeuvrability, seakeeping and interaction characteristics (Barras, 2004). Particularly, only
in recent years has manoeuvrability become subject to
IMO resolutions as in IMO (1987) and IMO (2002).
Further, for seakeeping and interaction we have no
knowledge of any existing safety criteria, though there
are several references suggesting ship motion limits, such as Graham (1990), and on how to calculate
interaction effects, as in Barrass (2004) and Millward
(1990). For the purpose of monitoring the ships operational characteristics we have no knowledge of any
available tool.
DECISION SUPPORT TOOL MODEL
This work, as previously mentioned, intends to propose

a decision support tool (DST) to support vessels operation and emergency reaction, and its basic structure
is presented in figure 1. This tool is composed of four
separate modules with the following purposes:
monitor the vessels loading condition by automatically measuring tanks levels;
verify at regular intervals, or as demanded, stability
and structural criteria;
provide decision support for reaction in case of
damage;
estimate the vessels operational performance,
limitations and manoeuvring characteristics.
This software follows a similar approach of the
FCCS software, but differs from it by being able to
estimate the load condition automatically, by being
able to provide some data on the ships performance,
and by having implemented an optimization model
using genetic algorithms to manage ballasting and
fluid transfer correctly upon damage occurrence, as

discussed in Martins and Lobo (2011).
In order to better explain the software, in the next
sections, each software task is going to be dealt with
separately.
3.1 Load condition estimate

The load condition monitoring is done by visual
reading of the draught marks at the begging of the
journey and then by monitoring automatically at regular intervals the tanks fluid capacity sensors using an
OPC standard interface (Schleipen (2008)) available
through the platform management system. The time
interval introduced is 5 minutes. Further there is also
the possibility for the user to introduce his own load
condition in a separate routine and then go ahead with
any analysis he wishes to do.
3.2 Stability criteria and strength criteria

monitoring
Stability and strength criteria are checked every time
load condition is monitored, by stand alone routines
that can be replaced by any other implementation of
different criteria as required for the vessel.
At the moment, the stability criteria for warships
presented in Sarchin and Goldberg (1962) are implemented, which include intact stability criteria and
damaged stability criteria for a damage length of
any two consecutive watertight compartments with no
regards for decks.
As far as structural criteria is concerned, for the
load condition in question, still water bending moment
and shear forces are compared with the ones used for
the ships design, which depend upon which classification society rules were used, or upon the International Association of Classification Societies Unified
Requirement (IACS, 2007).
3.3 Operational analysis

The load condition also affects vessels behaviour, seaworthiness, and manoeuvring as mentioned in several
references such as Inoue, Hirano et al. (1981). Our
aim is to present up-to-date values of the manoeuvring
data mentioned in IMO (1987), complemented with
interaction data.
Based on sea trials data required by IMO (2002),
complemented with some other trials (most of them
turning trials) it is possible to provide some evaluation
about:
manoeuvring characteristics with the present load
conditions;
travelled distances by the ship at current speed up to
full stop following emergency stop or coastal stop
procedure;
autonomy at current speed and using economic
speed from that time onwards;
Figure 1. Bridge officer DST fluxogram.
148
maximum squat values and canal width and depth

limits from which interaction may be expected,
using Barrass (2004).
In order to estimate hydrostatic coefficients from sea
trials data, an optimization process similar to the one
mentioned in Viviani et al (2007) was used.
Nevertheless, in order to provide the data required
for navigation planning (figure 2) we intend to implement a neural network architecture that will be trained
with data collected while the ship manoeuvres under
different weather conditions, with different load conditions, and with different depth limitations (open
waters, restricted waters or confined channels). This
prediction method has already been validated, e.g.
Moreira and Soares (2003) and Martins and Lobo
(2007).
3.4
Damage input
The objective of the damage mode capability

(implemented in separate routines) is to support the
users decision on how to manage fluid loads in case
of flooding and to advise on how to attribute drainage
equipment to the damaged areas.
Obviously, the decision must take into account all
tanks levels at that time, the damage location and its
dimensions in order to calculate the sea water inflow
rate, and how the flood spreads.
Knowing all of these, of which only the damage
location and dimensions is required to be a users input,
a Decision Support Tool (DST) can be built upon the
software that is being discussed.
This DST is basically an optimization model with
tank levels and watertight compartments flooding level
as input variables, aiming to maximise stability and
minimize response time, while complying with several
constraints, such as damage stability criteria and limit
bending moments to the ones supported by the actually
built structure. Next section will be dedicated only to
this matter.
Figure 2. Navigation plan of turning in restricted waterways (adapted from reference House, 2007).
SHIP SURVIVABILITY UNDER DAMAGE

AS AN OPTIMIZATION PROBLEM
Finding a course of action to increase stability of a

damaged ship by managing fluids, by flooding compartments, pumping water out, etc., can be seen as
a complex optimization problem whose fluxogram
is presented in figure 3 and whose solution can be
achieved using genetic algorithms as is explained in
more detailed in Martins and Lobo (2011).
From now on we are going to present the mathematical formulation that characterizes the optimization
problem.
4.1 Input variables
As mentioned before, the input variables are the ones
that characterise the problem, i.e. vessels characteristics and damage characteristics. So, the following
variables can be defined:
LCG TCG VCG
HY
LK
TK
HY
ships light displacement, and

correspondent centre of
gravity position;
hull form;
ships cross curves;
internal tank subdivision
and capacity tables;
watertight compartments
subdivision, dimensions
and characteristics;
Figure 3. Flooding response optimization problem.
149
SK
4. Fuel can be changed between tanks but it cant be

discharged overboard
solid weights and

correspondent centre of
gravity position;
damage inflow rate calculated
from its position and
dimensions.
Qi
4.2
5. Lub oil and miscellaneous tanks are not changed
Decision variables
The fluid level (weight) that is inside each tank and

watertight compartment, as well as drainage equipment attribution, are going to be the decision variables.
Therefore:
Ct (TKi )
Ct (WCi )
WQi
fluid tanks levels, where i identifies the

different tanks;
watertight compartment flood levels
where i identifies the different
compartments;
watertight compartment drainage flow
rate dependent upon the attributed
equipment.
6. When one of two compartments with a crossflooding pipe connecting each other is flooded, so
is the other
where j and j are two compartments connected by
a cross-flooding pipe
7. No watertight compartments that were not flooded
due to damage are going to be flooded
4.3 Auxiliary variables

From the previous variables it is possible to calculate
the initial load condition (LC0 ) and, as the journey
continues, the different load conditions by the change
of fluids within the tanks change Ct (TK) while the
solid cargo is considered to be unaltered, so that for a
given instant in time:
where t1 is the time when damage occurs and t2 is

some time after.
8. Only if drainage flow of the available equipment
is larger than the flow of water ingress, is the
equipment used to drain the compartment
9. Only if damage stability criteria are met after

drainage, is equipment used to drain the compartment
where f is a known function.

4.4
where j identifies the equipment and WCk corresponds to a set of flooded compartments that dont
include WCi .
Constraints
The first major constraint is to ensure that stability

and strength criteria are met, whose values can be calculated using the current load condition and flooded
watertight compartments data, that are complemented
with some other:
1. Stability criteria mathematical representation:
4.5
Cost function
As mentioned before, the aim of the problem is to

increase vessels survivability rapidly in case of damage. So the cost function has to be developed taking
into account:
2. Tanks and watertight compartments are filled/

flooded from 0 to 98% of their total capacity
how far the results of the damage stability criteria

calculations were from their limits;
how much time does it take to drain the flooded
compartments;
how much time does it take to ballast tanks or to go
ahead with the fluid change between tanks;
3. Fresh water tanks must have the same amount of

water or less (sea discharge) since no ballast was
considered to fill them up
each one of these are measured by a value which is

then multiplied by a scale factor, in order to make their
magnitudes comparable, and finally weighted according to the importance the user gives to each one. This
can be translated into expression (11) that is intended
to maximize:
where t1 is the time when damage occurs and t2 is

some time after.
150
where k1 , k2 , k3 are scale/weight factors, t is the time

corresponding to the solution implementation, and
t 1 is the time corresponding to damage.
BRIDGE OFFICER SUPPORT SYSTEM

IMPLEMENTATION
At that moment in time (t = 0) the load condition

previous to damage is read, and how the flood is going
to spread is checked by identifying all accesses (doors,
hatches) and ventilation conducts that are opened from
the flooding compartment.
The optimization process is done using genetic
algorithms in two steps. In the first step 20 initial candidate solutions are built up changing ballast tank levels
and after only three iterations (80 damage stability calculations done in about 1 minute) a first proposal for
immediate action, including the attribution of drainage
equipment is presented.
After that the best obtained solution, together with
the initial set of candidate solutions, are once again
used as an input for the genetic algorithm, in order to
refine the results presented.
Figure 6 presents the improvement on the purposed solution given by the GA implementation for
a scenario where a given area of the ship was flooded.
As we can see, in this case, the genetic algorithm
converged very rapidly to the final solution (we formalized the problem as a minimization problem so
lower fitness means better stability). It is also clear
that the two step generations method is significantly
better than the other.
The decision support tool software was developed

using MATLAB, which not only enables the user to
work in a friendly environment (see figures 4 to 6),
but also includes several libraries of which the OPC
library, the genetic algorithms library and the neural
networks library were used.
Having used a modular and flexible implementation, there are several routines that can work as
standalone, or can be called at different stages of the
program. This allows easy adaptation to other ships,
simplified debugging, and makes the addition of different methods of calculation and further development
easier.
Ships data are stored in a structure type variable,
and the program starts by reading the input file of arbitrary loads, and begins reading sensor data via OPC.
Once the interaction starts, the user is presented with
figure 4, where we can see several ship parameters
(upper left), weight distribution and still water bending moment (upper right), manoeuvring data (lower
left), righting arms curves (lower center), and the state
of the alarms (lower right).
Additional information can be accessed from the
menu on the top of the window, although in normal operation that is not required and criteria will be
checked autonomously.
The system is also capable to run simulations for
a given condition once the user introduces the tank
filling percentage (%) and solid cargo distribution.
The methods used are the same as in the automatic
mode. However, once the simulator is turned off, all
simulation related data are lost (figure 5).
In case of damage, in order for the optimization process to start, the user has to select the damage position
and characteristics making use of the interface window
presented in figure 6.
Figure 5. Simulation interface window (watertight subdivision and tank identification was erased).
Figure 4. Main interface window.
Figure 6. Genetic algorithms progress comparison of two

methods for a given damage situation.
151
CONCLUSIONS
A decision support tool for the bridge officer was

presented, which provides up-to-date information on
stability and strength, manoeuvring data and other
performance characteristics, during current operation.
On the other hand, the system recommends the
best course of action in case of damage, by suggesting which tanks to flood or empty. Simulation tests
show that our approach to this problem provides good
solutions in a very short time.
We feel that the system developed can significantly
improve safety at sea.
REFERENCES
Barrass, D. C. B. 2004. Ship Design and Performance for
Masters and Mates.
Graham, R. 1990. Motion-Induced Interruptions as Ship
Operability Criteria. Naval Engineers Journal.
House, D. J. 2007. Ship Handling, Theory and practice.
IACS. 2007. Requirements concerning Strength of Ships.
UR S.
IMO. 1987. Resolution A.601(15). Provision and display of
manoeuvring information on board ships
IMO. 1993. Resolution A.748(18). Code on Intact Stability
for all types of ships covered by IMO Instruments.
IMO. 2002. Resolution A.137(76). Standards for ship
manoeuvrability
IMO. 2009. SOLAS Consolidated Edition 2009 Consolidated text of the international Convention for the Safety
of Life at Sea London.
Inoue, S., M. Hirano, et al. 1981. Hydrodynamic Derivatives of Ship Manoeuvring International Shipbuilding
Progress 28.
Martins, T. and S. Lobo. 2011. A GA based decision support
tool for stability and structural viability under damage
Damaged Ship, London.
Martins, T. and V. Lobo. 2007. Estimating Maneuvring and
Seakeeping Characteristics with Neural Networks. IEEE
Conference Oceans 07. Abardeen.
Millward, A. 1990. A preliminary design method for the
prediction of squat in shallow water Marine Technology
(27(1)): 1019.
Moreira, L. and G. Soares. 2003. Training Recurrent Neural
Networks with Noisy Data for Manoeuvring Simulation.
Computer Applications and Information Technology in the
Maritime Industries (COMPIT03).
Plumley. 2009. Damage-Control Tactics in Extreme Stability Situations. Sea and Shore.
Sarchin, T. H. and L. L. Goldberg. 1962. Stability and
Buoyancy Criteria for U.S. Naval Surface Ships. SNAME.
New York.
Schleipen, M. 2008. OPC UA supporting the automated
engineering of production monitoring and control systems. 13th IEEE International Conference on Emerging
Technologies and Factory Automation.
Viviani, M., C. P. Bonvino, et al. 2007. Identification Of
Hydrodynamic Coefficients From Standard Manoeuvres
For A Series Of Twin-Screw Ships. 2nd International
Conference on Marine Research and Transportation
Naples.
152

Decision support system for the safe operation of fishing

vessels in waves
J.M. Rodrigues, L.P. Perera & C. Guedes Soares
ABSTRACT: The paper describes an onboard decision support system that is being developed to support the
operation of fishing vessels in waves, in particular for the decisions about ship handling in waves that have
to do with vessel safety. The system includes a component of hardware that monitors several motion related
parameters, and a component of data processing that uses those measurements and provides the ship master
with the consequences of the different ship handling decisions. The paper describes the decision criteria and
the approaches adopted for the calculation of the parameters that govern the masters decisions. It describes the
software that was developed to perform those calculations and to display in a user interface the advice to the
master as well as the data acquisition and processing hardware that has been considered and tested to be installed
on board.
INTRODUCTION
The fishing industry is one of the professional activities most prone to labor accidents. Petursdottir et al.
(2001) state that an average of 24000 fishermen loose
their lives every year. As Anto and Guedes Soares
(2004) show in their study regarding the Portuguese
fishing fleet, 89% of the accidents studied are related
to operational safety issues. In Mantari et al. (2009),
a study has been made regarding the stability of fishing vessels in waves when gear loads are introduced.
Here it has been concluded that fishing gear heeling moments are more important than the heeling
moments produced by the IMO weather criterion, and
that a combination of these two leads to capsizing
even in regular load conditions.
Complementary to this, Tello Ruiz et al. (2009)
made a seakeeping analysis of a set of fishing vessels
performance under different operational conditions.
The results have shown that large lateral accelerations
have a strong impact on the crew and on the machinery
systems performance, even when the ships survivability is not at risk. Both these works utterly propose the
solution of a decision support system to overcome the
simplification of the IMO stability standards, particularly regarding the operability conditions, adding to
the stability monitoring.
Up until Koyama et al. (1982), the development of
onboard systems to aid the navigation in rough weather
had been mainly concerned with structural integrity
and equipment safety, and only proposed for application on large ships. Lindemann and Nordenstrom
(1975) developed one such system by measuring the
accelerations in six degrees of freedom and the stresses
at a cross section. Hoffman (1976) suggested using

ship to shore communications along with charts for
routing in heavy weather. Later in Hoffman (1977) and
(1980), results of the installation of such a system on
board of a vessel in operation were published. However, its application seems to have served primarily
as a way of establishing a performance index for the
vessel in rough sea. At the same time, similar projects
were under way such as Lindemann et al. (1977) in
cooperation with DNV and Lloyds classification societies. Also a monitoring system with the objective of
avoiding excessive angular motions was introduced
in offshore industry in the North Sea (Schofield,
1977).
The work of Koyama et al. (1982) consisted of
a computer based system capable of computing the
mean period and the root mean square prediction of
roll motion. The input component was a pendulum
for measuring the ship motions and, given a predetermined criterion, an alarm would fire in case of
danger. Unfortunately the pendulum system proved
unsatisfactory especially for high speeds so the results
were shown to be unreliable.
Kse (1994 and 1995) proposed a scheme based on
low cost equipment which, with special purpose developed software, resulted in an encompassing reliable
system for stability monitoring and advising applicable to small ships. Varela et al. (2010) developed a
computer application to be part of an onboard system specifically tailored for fishing vessels, which is
similar to the ones depicted in Benedict et al. (2006).
However, the application developed strictly considers the IMO (2007) guidance formulae without any
calculated prediction of the response of the ship.
153
Table 1.
a period of two hours. It is governed by the following

expressions:
RMS criterion.
Description of work
x 3,P
m/s2
x 2,P
m/s2
x4,P
deg
Light manual work

Heavy manual work
Intellectual work
Transit Passengers
Cruise Liner
0.20 g
0.15 g
0.10 g
0.05 g
0.02 g
0.10 g
0.07 g
0.05 g
0.04 g
0.03 g
6.0
4.0
3.0
2.5
2.0
MSI =
In this paper a system is presented which is based on

the real-time computation of the predicted responses
of the ship considering the local sea state. That work
is here further developed to implement the computation of the predicted motions and accelerations of the
vessel, and confront statistical values of these against
operational criteria. In addition, there is now the possibility to visualize combinations of speed and relative
heading where the probability of occurrence of parametric roll exceeds a certain value. The effect of the
gear is considered as a static suspended weight due
to the very large length of the wires (particularly in
trawlers). The prediction of motions and accelerations
is done through the construction of the directional
JONSWAP scalar spectrum from the wave main characteristics: significant wave height, wave modal period
and wave direction (to be fed by the data acquisition
system once installed). Spreading of the incident wave
field is taken into consideration following the work of
Longuet-Higgins et al. (1963) and Mitsuyasu et al.
(1975).With these values, the criteria for operability
and seasickness in Faltinsen (1990) and in OHalon
and McCauley (1974) are assessed.
A data acquisition system is proposed to be installed
on board and provide the necessary data to the application described above. One of its core functionalities
is to estimate in real-time the sea spectrum from
the measurements of motion, speed, acceleration, etc.
This component has been developed previously and
its development, along with a full description of its
implementation can be found in Pascoal et al. (2005),
Pascoal et al. (2007), Pascoal and Guedes Soares (2008
and 2009).
where az stands for the vertical acceleration at some

point in the ship where one wishes to assess the incidence of motion sickness and, in (Eq. 2), e is the
frequency of encounter.
The spectrum considered is the JONSWAP spectrum with its formulation generated from the significant wave height HS and the average period Tm , with
the following spectral density as a function of the
circular frequency :
where Hasselmann et al. (1973) suggests:
with:
The directional spreading implemented follows the

work of Longuet-Higgins et al. (1963) and Mitsuyasu
et al. (1975). It is given by:
where is the wave incidence angle and A2 may be

calculated with the following expression:
2 CRITERIA FOR SHIP OPERATIONAL
SAFETY
The operability criterion is related with the type of
work to be performed at a given location on the
vessel with its maximum rms (root mean square)
values of lateral acceleration, vertical acceleration
and roll amplitude. The limit values according to
NORDFORSK (1987) are listed in (Table 1).
The OHanlon and McCauley (1974) criterion for
MSI (Motion Sickness Incidence) is defined as the
percentage of people to experience seasickness during
In (Eq. 8), s is the directional spreading factor, with an

infinite value associated to a non-dispersed wave. The
function has a definition valid for complex numbers
z with a positive real part, given by Legendre:
154
For each wave incident angle, ship course and speed,

the directional spectrum comes:
Here, represents the mean wave incident angle and

a specific angle to which the wave disperses.
For each combination of ships speed v and relative
wave incident angle , the motion response spectrum,
as a function of the frequency and wave mean incident
angle, is calculated by:
where Fi,v, stands for the transfer function regarding

the motion mode i.
The amplitude root mean square values of motion
0 and acceleration 4 are given by:
In the expressions above, m0 and m4 are the variances of motion and acceleration respectively. These
are calculated as follows:
In (Eq. 12) and (Eq. 13) the motion, speed and

angle indices are implicit, but have been omitted for
the sake of clarity.
It is crucial to point out that the above defined quantities are evaluated at some point in the structure of
the ship where it is significant to assess the motions.
Therefore, there is a need to expand the linear motions
from the centre of gravity, which are given by the
raw transfer functions, into the absolute motions at
the specific point of interest.
3
DECISION SUPPORT SYSTEM
The aim of the decision support system is to assess

the real time loading situation, the sea state and, based
on the vessels hydrostatic and dynamic characteristics, to evaluate the stability condition and prediction
of the probability of occurrence of several dangerous
phenomena related to the presence of waves. Furthermore, the criteria defined in Section 2 are checked
and countermeasures are suggested to the master if
the ship is in a speed/course combination where the
operational safety may be compromised.
The sea state is inferred from the measurement of
the motions of the ship through the onboard installation of a data acquisition system (DAQ), by using
a Kalman filter type algorithm (see Section 4). By
equating the ships speed and course, and the wave
relative heading, period and amplitude, a simplified
analysis, as presented in the IMO (2007) guidance
formulae, is conducted. When the calculations, for

a given speed and course, result in a high probability of occurrence of synchronous rolling, parametric
rolling, surf riding and broaching, or encounter with
high amplitude system of waves, the corresponding
speed/course combinations are identified and mapped
into a polar plot (the drawn zones are the danger ones
to steer off). Also, the system computes the expected
value of the lateral acceleration, vertical acceleration
at specific important locations, and of the roll angle
amplitude. If these quantities surpass the limits of the
criteria described in Section 2, the speed/course combinations where it occurs are also drawn onto the polar
plot. The user has the possibility to visualize whatever phenomena danger zones he/she wishes to see by
pressing the respective buttons on the user interface.
Next, a description of the architecture and of the data
flow between the various modules that constitute the
implementation of the system is presented.
The system is composed of a Load input module, a Static stability analysis module and a Dynamic
behavior in waves module. The first allows the consideration of the weight distribution through a pre-built
file and/or by entering the values manually.
The second module computes the hydrostatic condition of the vessel by crossing the information in
the previous module with the hydrostatic tables in an
accessible database. Finally it checks for the intact
stability criteria.
The third module has as input the wave system
main parameters and the ships course and speed. This
module implements the computation of the angle and
speed limits for the various dangerous phenomena as
a function of the combination of vessels speed and
incident wave heading. The combinations of speed
and heading for which to expect the occurrence of
parametric roll are also taken, by querying a pre-built
simulation results database with the current sea-state
and cargo condition as the search predicate. The rms
values of the quantities to be checked by the operability and seasickness criteria are computed and the
zones for which these values exceed the tolerable are
identified.
The resulting data are output in the form of a
polar diagram where combinations of speed and relative wave heading to avoid are drawn. If the vessels
current speed and heading is within any zone as
defined in Varela et al. (2010), then countermeasures
are suggested. The main architecture is presented in
(Figure 1).
3.1 Load input module
The load module is divided into four groups: (i) Vessel
identification, (ii) Static weights, (iii) Tanks and suspended load, and (iv) Sensors.
The Vessel Identification data are taken from a
pre-built file. These data include the ships name,
IMO number, Id number, mean displacement, length
between perpendiculars, breadth, draft and the longitudinal location of the draft marks at the bow and stern.
The Static weights group lists the current static weights
155
Table 2.
Intact stability criterion fields.
Criterion
Fields
Intact stability
general
calculations
Displacement (ton)
LCGmin (m)
LCGmax (m)
KG (m)
FS correction (m)
Corrected GMt (m)
Area until 30 (m.rad)
Area until 40 (m.rad)
Area 30 40 (m.rad)
Area A (m.rad)
Area B (m.rad)
Draught fwd. (m)

Draught aft. (m)
Trim (m)
Freeboard stbd (m)
Freeboard port (m)
Maximum righting
arm angle (rad)
Area A (m.rad)
IMO intact
stability
criteria
IMO weather
criterion
USCG
suspended
weight
GZ at 30 (m)
Initial GMt (m)
GZmax (m)
A/B
Figure 1. Main system architecture.
Figure 3. Intact stability module.

Figure 2. Load input module.
and allows for the addition or subtraction of individual weight items by specifying its load and location. It
is possible to reset the list of items to the pre-defined
project values.
The Tanks and suspended load group shows
the current tank filling status and allows for the
modification of the individual fillings in terms of percentage. If the vessel is under hauling operation or
pulling the fishing bag, the strength and spatial position in action may be introduced. The Sensors group
shows the instantaneous values of the instruments.
These represent measurements of longitudinal inclination, transverse inclination, portside and starboard free
board, and suspended load. At the present stage these
are still dummy measurement values. Figure 2 shows
a screenshot of the user interface of this module.
3.2
Intact stability module
The module is divided into three main components:

(i) Stability calculations, (ii) Visual representation,
and (iii) Criteria. Table 2 lists the data that are visualized in the Criteria component, whose fields are
properly described in Varela et al. (2010), and Figure 3
shows a screenshot of the corresponding user interface.

Besides the calculated data based on the cargo condition, the instantaneous values from the instruments
readings are also shown.
The Visual Representation group comprises the
visualization of the percentage of filling of each
individual tank, discretized in four levels of percentage where a color code is applied. The Criteria
group shows the evaluation of three criteria defined
in Varela et al. (2010).
3.3 Dynamic behavior in waves module
In this module the user may see the danger zones to
steer clear off. Figures 5 to 13 show the polar plot
with the possible combinations of speed and course.
The combinations where the phenomena described
in IMO (2007) are probable to occur, or where the
criteria defined in Section 2 are triggered, are plotted,
each one with a specific color code. The current ship
speed and heading is represented by a red arrow, which
is observable in the same figures. The area beneath
the polar plot is the field where countermeasures are
written.
156
Figure 4. Dynamic behavior in waves module.
Figure 6. Parametric roll (IMO).
Figure 5. Synchronous roll (IMO).
Figure 7. Surfing and broaching (IMO).
The module is divided into three main areas:

(i) Wave input, (ii) Criteria output, and (iii) Polar
representation.
In the Wave input group the user is allowed to
fill the wave and routing fields. The first consists on
the wave direction, significant wave height and mean
period. The second relates to the vessel course and
speed and the corresponding resulting governing values. Finally the third is a polar diagram that shows the
zones where the several dangerous phenomena may be
experienced and where the criteria for operability and
seasickness are exceeded.
This module has two distinct parts that consist on a
simplified approach and an advanced approach. The
first regards the dangerous phenomena to be avoided
following the IMO-MSC(2007) criteria, whereas the
second complements this information with the results
of the numerical simulations carried out for the parametric roll occurrence and exceeding of the operability
and seasickness criteria. The transfer functions necessary in (Eq. 11) for the set of speeds and relative
wave incidence angle are computed a priori by using an
in-house developed code. The vertical acceleration and
lateral acceleration is computed for a set of two distinct
locations on the vessel, the criteria are consequently
checked for each of these points.
The data flow of the implementation of the dynamic
behavior module (advanced approach) is represented
Figure 8. High amplitude wave group (IMO)).
in Figure 4. What is shown here is the algorithm that in

reality is made of two dynamic link libraries (as seen in
Figure 1). In Figures 5 to 8 screenshots of the module
are presented showing the dangerous phenomena, as in
Varela et al. (2010). The parametric rolling prediction
and the operability and seasickness criteria assessment
output is in Figures 9 to 11. Finally, the superposition
of these two sets of phenomena are shown in Figure 12
and Figure 13.
157
Figure 9. Parametric roll (from DB).
Figure 12. Superposition (IMO).
Figure 10. Operability (Faltisen 1990).
Figure 13. Superposition (advanced approach).
Figure 11. Seasickness (OHanlon and McCauley 1974).
DIGITAL DATA ACQUISITION SYSTEM
The proposed system has been implemented on a digital data acquisition platform in a prototype scale as
presented in Figure 14. The digital data acquisition
system is divided into two sub-systems: Motion monitoring sub-system and Force monitoring subsystem.
The main objective of the Motion monitoring subsystem is to evaluate the vessel motions under regular
and irregular sea conditions. The Motion monitoring system consists of several sensors: accelerometer,
angular rate sensor, inclinometer and a wave height
measurement sensor. The accelerometer measures the
surge, sway, and heave accelerations, the angular rate
sensor measures the yaw angular velocity and the
inclinometer measures the roll and pitch angles of the
motion of vessel. The wave height measurement sensor is capable of measuring the absolute wave height of
the bow waves and has an internal accelerometer for
compensation of vessel motions during its data collection process. However, the wave height sensor will
be replaced by an estimation process that will evaluate wave height from the vessel motions in future
developments. Therefore the cost of this system will
be minimized.
The main objective of the force monitoring subsystem is to evaluate the fishing-gear load in real-time.
This data is used to assess the suspended weight criteria
described in Varela et al. (2010), and may also be used
for further specific studies on this subject. However,
due to the unavailability of fishing-gear facilities up
to the present date, the force monitoring sub-system
has only been simulated. The proposed digital data
acquisition system will now be further illustrated in the
following sections: Hardware Structure and Software
Architecture.
158
Figure 15. The experimental platform.
Figure 14. Digital Data Acquisition System.
4.1
Hardware structure
Figure 16. The software architecture.
As shown in Figure 14, the hardware structure consists mainly of the following units: laptop computer, Ethernet switch, desktop computer, GPS unit,
Compact-RIO, power supplies, external hard-drive,
and other sensors (accelerometer, angular rate sensor,
inclinometer and wave height measurement sensor).
All units communicate through the Ethernet switch
that is presented in Figure 14. The experimental
platform of the hardware structure is presented in
Figure 15.
The laptop computer that is connected to the
Compact-RIO unit through the Ethernet switch is
the main input/output user interface in this system.
Furthermore, the laptop computer is also connected
to the external hard drive for data storage and to the
GPS unit for vessel positions, course and speed evaluation. The GPS unit consists of an external antenna and
it is capable of accurately estimating the vessel position
(Longitude and Latitude), and the speed and course on
the earths referential. The Compact-RIO main unit is
connected to the sensors previously depicted, for collecting the measurement data as shown in (Figure 14).
Furthermore, a power supply unit is used in this system
to power several main units: GPS unit, Ethernet switch
and Compact-RIO.
The Laptop computer acts as the main control unit
of the prototype system. There are two software components that run on the laptop: Main Program (with
its user interface) and the Ship Operability Assessment Program. The first, coded with LABVIEW, is
related to the digital data acquisition. The second is
responsible for assessing the operational criteria as
presented in section 2. Both these components are
further discussed in section 4.2.
The desktop computer runs the LABVIEW realtime operating system, which enables high-speed
calculation for Kalman filter based wave spectrum

estimation. The inputs to the wave spectrum program are the sensor measurements, the vessel course
and speed. The output of the wave spectrum program
is the estimated wave spectrum including the wave
direction, significant wave height and mean period.
These quantities are then forwarded into the Ship
Operability Assessment Program for further analysis.
4.2
Software architecture
The proposed software architecture is presented in

Figure 16. Within the LABVIEW environment there
are three core modules: The DAQ User Interface,
the Main Program (running on the laptop computer),
and the Wave Spectrum Estimation Program which is
running under desktop computer.
The Main Program collects and stores the realtime sensor data, sends and receives information
to/from the Wave Spectrum Estimation Program and
proceeds to interact with the Operability Assessment
Program through the use of continuously updated temporary data files. These temporary files and the results
from the Operability Assessment Program analysis are
the link between the data acquisition system and the
decision support system described in section 3. The
Operability Assessment Program serves basically as
a wrapper and uses the same dynamic link libraries
shown in (Figure 4).The linking of these libraries to the
Dynamic behavior in waves module (see section 3.3),
is to be replaced with a straightforward querying of the
temporary files depicted in this section. The user interface of the data acquisition system (see section 4.3)
serves mainly as a technical tool for the development
and troubleshooting of the system, while the one in
section 3 is to be used by the bridge officers.
159
by the program and a warning light will also be

displayed accordingly.
Motion signal display: The calibrated ship motions
that are measured by the sensors are displayed
in this section. The ship surge, sway and heave
accelerations, pitch and roll angles and yaw rate
are displayed in the top area of this section. The
measured wave height is displayed on the bottom
area of this section.
Figure 17. User interface in Laptop Computer.
4.3
DAQ user interface
The view of the DAQ user interface on the laptop

computer is presented in Figure 17. The interface
consists of 9 parts: data management, application
management, GPS display, channel test, danger zones
display, statistics data, estimated spectrum display,
loads conditions display, and motion signals display.
These sections are presented by separate tabs:
Data management: The data management unit is

responsible for collecting sensor and GPS data
(accelerometer, wave height measurement sensor,
inclinometer, rate gyro sensor, strain gauge sensors).
The sensor data is then saved to an external hard
drive.
Application management: The application management section enables the operator to stop, reboot
and restart individual processes and programs.
These processes and programs are: User Interface,
CompactRIO and EtherCAT units, and Wave Spectrum Estimator.
GPS display: The GPS display is responsible for
read and display services of data collected from the
GPS sensor.
Channel test: This section is designed to read and
write uncelebrated data from the different channels
in the data acquisition system. Any problems with
read and write operations with DAQ channels can
be evaluated in this unit.
Danger zones: The main objective of the danger
zone section is to display the results calculated by
the Operability Assessment Program. A polar plot
similar to the one described in section 3 is presented.
Statistics display: This unit consists of statistical
information about the accumulated data that was
collected by the ship sensors. However, currently
this unit is under development.
Estimated spectrum display: Relates to the wave
spectrum estimation which has been done by the
Kalman filter algorithm that is running under the
LabVIEW RT platform on the Wave Spectrum
Estimator (PC Computer). An isometric view of
the bi-dimensional spectrum is displayed in this
window.
Force conditions display: The vessel fishing-gear
load conditions that are simulated are displayed in
this component. The overloading is to be monitored
CONCLUSIONS
An existing program, designed to function as the

software component of an onboard decision support system for fishing vessels, has been successfully
extended. Information about the conditions favourable
to parametric roll occurrence has been implemented
based on a pre-built database. Also both Operability Index and Motion Sickness Incidence prediction,
which are inferred from a priori computed transfer functions, the wave spectrum estimation and the
spreading function inclusion, have been introduced. A
prototype model of the proposed data acquisition system has been assembled. This system was subjected to
fundamental operational testing and the results have
been coherent. The authors expect an opportunity to
proceed to ship installation and trials, in view of
validation and troubleshooting.
ACKNOWLEDGEMENTS
This work has been developed in the scope of the
project Decision Support System for the Safety
of the Fishing Vessels Subjected to Waves (SADEP)
funded by the Portuguese Foundation for Science
and Technology under contract n.o PTDC/EME-MFE/
75233/2006
The initial version of this system, which was a precursor to the present system, has been developed by S.
Varela and T. Santos, as indicated in the text.
REFERENCES
Benedict, K., Kirchhoff, M. and Baldauf, M. (2006),
Decision Support for Avoiding Resonance and Wave
Impact for Ship Operation in Heavy Seas, Safety
and Reliability for Managing Risk, Guedes Soares, C.,
Zio, E. (Eds.), Francis & Taylor, pp: 27432750.
Datta, R., Rodrigues, J. M. and Guedes Soares, C. (2011),
Study of the motions of fishing vessels by a time domain
panel method, Ocean Engineering 38 (2011), 782792.
Faltinsen, O. M. (1990), Sea Loads on Ships and Offshore
Structures, Cambridge University Press.
Francescutto A. (2007), The intact ship stability code:
Present status and future developments, Proceedings of
the 2nd International Conference on Marine Research and
Transportation, Naples, Italy, Session A, 199208.
Gefaell G. (2005), Some consideration about the stability
and safety of fishing vessels less than 24 m of length,
Naval Engineering, April, 4658.
160
Graham, R., Baitis,A. E., Meyers W. G. (1992), On the development of seakeeping criteria, Naval Engineers Journal,
Vol. 104, Issue 3, pp: 259275.
Hasselmann, K., Barnett, T. P., Bouws E., Carlson H.,
Cartwright D.E., Enke K., Ewing J.A., Gienapp H., Hasselmann D.E., Kruseman P., Meerburg A., Mller P.,
Olbers D.J., Richter K., Sell W. & Walden H. 1973. Measurements of wind-wave growth and swell decay during
the Joint North Sea Wave Project (JONSWAP). Dtsch.
Hydrogr. Z. Suppl. A 8(12): 95.
Hoffman, D. (1976), The impact of seakeeping on ship
operations. Marine Technology, 13: 241292.
Hoffman, D. (1977), Wave measurements and implementation on board ship, Technical report, National Maritime
Research Center, NMRC-KP-141.
Hoffman, D. (1980), The integration of shipboard and shore
based systems for operation in heavy weather, Ship
Operation Automation, pp: 159168.
IMO-MSC (2007), MSC Circ. 1228 Revised Guidance to
the Master for Avoiding Dangerous Situations in Adverse
Weather and Sea Conditions.
Kose, E. (1994), Dynamic monitoring of ship stability,
Ph.D. thesis, University of British Columbia.
Kose E., Gosine R.G., Dunwoody A.B., Calisal S.M. (1995),
An expert system for monitoring dynamic stability
of small craft, Ocean Engineering, Vol. 20, No. 1,
pp: 1322.
Koyama, T., Susumu H. and Okumoto, K. (1982), On a
micro-computer based capsize alarm system, Second
International Conference on Stability of Ship and Ocean
Vehicles, pp: 329339.
Lindemann, K. and Nordenstrm, N. (1975), A system for
ship handling in rough weather. Fourth Ship Control
System Symposium.
Lindemann, K., Odland, J. and Strenghangen, J. (1977),
On the application of the hull surveillance system
for increased safety and improved utilization in rough
weather, SNAME, Vol. 85.
Longuet-Higgins M. S. (1963), The effect of nonlinearities
on statistical distribution in the theory of waves, J. Fluid
Mech., 17, 459480.
Mantari J.L., Ribeiro e Silva S., Guedes Soares C. (2009),
Intact stability of fishing vessels under operational
loading conditions, Proceedings of the 13th Congress
of International Maritime Association of the Mediterranean (IMAM 2009), 1215 October, Istanbul, Turkey,
pp: 185193.
Mantari J.L., Ribeiro e Silva S., Guedes Soares C. (2009),
Variations on Transverse Stability of Fishing Vessels
due to Fishing Gear Pull and Waves, Proceedings
of the XXI Naval Architecture Pan-American Conference (COPINAVAL 2009), 1822 October, Montevideo,
Uruguay.
Mitsuyasu, H., Tasai, F., Suhara, T., Mizuno S., Ohkusu, M.,
Honda, T. and Rikiishi, K. (1975).
NORDFORSK. 1987. Seakeeping performance of ships,
Assessment of a ship performance in a seaway. Nordic
Cooperative Organization for Applied Research.
OHanlon, J. F. and M. E. McCauley (1974), Motion
Sickness Incidence as a Function of Vertical Sinusoidal
Motion, Aerospace Medicine, AM-45(4), 366369.
Pascoal, R., Guedes Soares C. and Srensen, A.J. (2005),
Ocean Wave Spectral Estimation Using Vessel Wave Frequency Motions, Proceedings of the 24th International
Conference on Offshore Mechanics and Arctic Engineering (OMAE 2005), 1217 June, 2005, Halkidiki, Greece,
Paper: OMAE2005-67584.
Pascoal, R., Guedes Soares, C., Srensen,A.J. (2007), Ocean
Wave Spectral Estimation Using Vessel Wave Frequency
Motions, JOMAE, 129(2), pp. 9096.
Pascoal, R. and Guedes Soares, C. (2008), Non-parametric
wave spectral estimation using vessel motions, Appl.
Ocean Res, 30(1), pp. 4653.
Pascoal, R. and Guedes Soares, C. (2009), Kalman Filtering
of Vessel Motions for Ocean Wave Directional Spectrum
Estimation, Ocean Engineering, 36(67), 477488.
Ribeiro e Silva, S. Tello Ruiz, M. And Guedes Soares,
C. (2009), Seakeeping Performance of Fishing Vessels
under Operational Conditions, Proceedings of the 13th
Congress of International Maritime Association of the
Mediterranean (IMAM 2009), 1215 October, Istanbul,
Turkey, pp: 401412.
Santos T., Villavicencio R. and Guedes Soares C. (2010),
Safety and stability of fishing vessels on waves, The
Portuguese Maritime Sector (in Portuguese), C. Guedes
Soares & V. Gonalves de Brito (Eds.), Edies
Salamandra, Lda., Lisbon.
Schofield, M. (1977), Unique heavy lift monitoring
system computerizes offshore conctruction. Offshore
Magazine.
Tello Ruiz, M., Ribeiro e Silva, S. and Guedes
Soares, C. (2009), Fishing Vessels Responses in
Waves under Operational Conditions, Proceedings of
the XXI Naval Architecture Pan-American Conference
(COPINAVAL09), 1822 October, Montevideo, Uruguay.
Varela, S.M., Santos, T. and Guedes Soares, C. (2010),
Monitoring system for the safety of fishing vessels subject to wave loads, The Portuguese Maritime Sector (in
Portuguese), C. Guedes Soares & V. Gonalves de Brito
(Eds.), Edies Salamandra, Lda., Lisbon.
161
Ship dynamics and hydrodynamics

Prediction of the motions of fishing vessels using time

domain 3D panel method
R. Datta, J.M. Rodrigues & C. Guedes Soares
ABSTRACT: A linear time domain 3D panel method is used to determine a variety of motion results for
three different fishing vessels. In the present investigation, some modifications are also prescribed in order to
increase the robustness and efficiency of the existing linear time domain panel code. Initially a rigorous study is
performed to check the efficiency and accuracy of the time domain code after incorporating such modifications,
then computed results are compared with other established theories and commercially accepted codes to analyze
the motion responses, while in a previous paper, most of the results are obtained only for the head waves,
in the present study, results are computed for head seas, quarterly head waves and following waves with various
speeds, to capture the effect of the speed and also the effect due to the changing of heading angles of waves.
It is observed that the present time domain panel code shows a consistent and good agreement between other
computed motion results.
1
INTRODUCTION
The seakeeping or ship motion problem is one of the

most complicated and challenging topics in the field
of numerical marine hydrodynamic. The study of ship
motions and wave loads of a ship is very important for
ship design as large motions affect largely and that can
lead to structural damages. Therefore it is necessary to
develop a practical and reliable tool for prediction of
ship motions and sea loads. Due to this need, prediction
of the motions of a ship when it traverses through a
wave field is still a challenging and interesting topic for
the researchers model with conventional strip theories.
For the hydrodynamic analysis of an advancing
ship, initially strip theories in various versions were
developed, followed by a series of developments in
slender-body theory. Among all other two- dimensional strip theories developed during the 1960s and
1970s, the Salvesen et al. (1970) version of strip theory
(STF) is still the one of the most popular and widely
used by the industry for the practical ship motion
and wave load calculations. Since early 80s, due to
the advances in available computing power the focus
gradually shifted towards development of 3D panel
methods.
Some of important developments in this context
are as follows. Liapis and Beck (1985) introduced the
time-domain Green function based solution method
for the 3D linear forward speed problem. Later Lin and
Yue (1990), Korsemeyer and Bingham (1998) among
others pursued variants of the same method for different class of 3D forward speed problems. Nakos and
Sclavounos (1990) proposed Rankine panel method
in frequency domain. Rankine source method was
also applied in the time domain (Nakos et al. 1993).

Solution methods using the frequency-domain forward
speed Green function are also pursued by several
researchers, and some relevant works here are by Wu
and Eatock Taylor (1987). All these are based on the
potential flow theory. But most of the work is done for
linear theory.
Among many different sophisticated theories, the
3D time domain panel method is one of the most common methods for analysing the ship motion problem as
it has many advantages over other frequency domain
methods. Time domain formulation allows some of
the nonlinear effects to be investigated using the bodynonlinear approach such as non-linear hydrostatic,
nonlinear Froude-Krylov force.
Hull forms of fishing vessels are significantly
different from those of merchant ships such as a container ship or large cargo ship. They have different
Length/Beam ratios, block coefficient, etc. also cross
coupling is very significant because of the large asymmetry in bow and stern part. Therefore it is interesting
to compare the results of a time domain code with
other presently available methods for these particular
hull geometries. The computational scheme proposed
by Datta and Sen (2007) shows good agreement with
other published results for large ships. But where it
has been applied to fishing vessels, it has been found
to give inadequate results. Therefore, there is a need
to introduce some modifications in the scheme, to go
beyond such restriction.
Datta et al. (2011) studied this problem and suggested some modification in terms of inclusion of
cross coupling term in force equation, solution for
equation of motion and refinement of mesh. These
165
changes result in more robust and efficient solution for

the motion problem even for fishing vessels but still
more extensive study is needed to check the robustness
and efficiency of the proposed scheme. In Datta et al.
(2011), most of the computations are carried out for
head waves and only for zero and service speed of the
vessel.
Hence in the present paper, more extensive study
is made for three fishing vessels and compared with
other available tool such as WAMIT and strip theory.
In this paper, results are obtained for head waves, following waves and for quarterly head waves. Also one
intermediate speed close to service speed of the vessel
is considered to check the efficiency of the proposed
method more rigorously. The results shows overall
good correlation with WAMIT and strip theory for
head waves, but one can notice some difference when
the computation are carried out for the other heading
especially when forward speed is considered. However, the overall behaviour of the motion results is very
similar for all type of theories applied here.
2
2.1
BRIEF MATHEMATICAL FORMULATION

AND NUMERICAL IMPLEMENTATION
the waterline at z = 0, VN represents the two dimensional normal velocity in the z = 0 plane of a point
on .
Differentiating with respect to the normal and
applying body boundary condition gives the following:
By using the body boundary condition, a solution for

(q, t) can be obtained from equation (3). Once the
solution for (q, t) is calculated, from equation (2),
the expressions for the disturbed potential can be
found out.
Once is known, the hydrodynamic pressure
can be calculated using the Bernoulli equation and
finally by integrating the pressure, one can get the
hydrodynamic forces and moments acting on CG.
2.2 Numerical implementation
Solution for hydrodynamic problem
The detailed formulation is available in many sources

(Lin and Yue, 1990), and here very brief derivation is
described for completeness.
For the linear formulation, a three-dimensional
floating body is considered advancing with steady forward speed U under waves. Let Oxyz be the Earthfixed and O x y z be the body fixed co-ordinate
system where O placed on the mean water surface,
where as O is placed on the waterline above CG of the
body. These two systems are related by the following
relation:
Let T be the total potential, then = T I is the

total disturb potential, where I is linear incident wave
potential. The velocity potential satisfies the Laplace
equation, free surface boundary condition, linearized
body boundary conditions and far field radiation
condition. Using time domain Greens function,
following integral equation can be obtained:
For the solution purpose, the zero order panel method

is used. Let N be the total number of panels which is
used to define the wetted body surface and Nw be the
total number of line segments representing the waterline (t). It is assumed that the source strength over
the waterline segments is the same as that of the corresponding adjacent body panels. The discretized form
of (3) is given by:
Using the collocation method, equation (4) can be

written in the form of a system of linear equations:
This system is then solved for the unknown source

strengths. Once these are evaluated, the velocity potential distribution can be solved similarly by discretizing
equation (2).
In the above, q = q(, , ) is the location of a point

source on the wetted body surface, (q, t) is the associated source strength at retarded time , G o (p, q) and
G f (p, q; t ) are the Rankine and regular parts of the
transient free surface Greens function, () represents
2.3 Coupling between force and motion

Let FT be the total force. Then FT consists of the three
following parts:
166
The static force FStatic can be calculated as follows:
where
Figure 1. Mesh of Fishing Vessel 2 (FV2). Left: Mesh of
original geometry, Right: Mesh modified at the water line.
Table 1.
where AWP , Mwp and correspond to the waterplane area, the moment about the waterplane area and
the displaced volume at the mean draft respectively.
GMT and GML are the transverse and longitudinal
metacentric heights.
The FK force is obtained analytically from the
incident wave potential and Fdynamic is calculated by
integrating the hydrodynamic pressure.
Once the total force is obtained in the present time
step t, one needs to use some explicit or implicit
scheme to predict the velocity and motion for the
advanced time step t + t. In order to do that, an
explicit rule can also be used for the integration which
results in the value of the displacement. Detailed
description of the scheme is presented in Datta et al
(2011), which is a reformulation of the method proposed by Datta and Sen (2007). The reformulation has
been done to avoid the numerical instability occuring when fishing vessels are considered. The detailed
discussion is given in Datta et al. (2011) and hence
not repeated here.
Main characteristics of the set of fishing vessels.
Designation
[m]
LBP
[m]
B (moulded)
[m]
D (moulded)
[mT]
FV1
FV2
FV3
15
22
20
5.4
5.8
7.4
2.4
1.6
3.4
104.9
94.7
302.9
Figure 2. Mesh of FV1.
2.4 Dealing with the free surface Green function

In the integral equations (2) and (3), numerical computations for free surface Green function G f (p, q; t )
and its derivatives demands more attention because
free surface Green function is very sensitive near free
surface. A detailed presentation can be found in Sen
(2002). It may be noted that the behaviour of the
free surface Green function near the free surface is
extremely diverging as discussed by Clement (1998).
It is observed that this nature of the free surface
Greens function leads to some numerical instability
for the seakeeping problem if the vessels hull includes
horizontal panels at the waterline.
To avoid such numerical instability, the body geometry of such vessels is modified near the waterline
by introducing some small vertical panels there. It is
assumed that, due to such small changes, the computed response of the ship is not affected significantly.
Figures 1 illustrate this procedure, where the panel
mesh has been modified accordingly. The detailed
discussion regarding the refinement of mesh is given
in Datta et al. (2011) and hence not repeated here.
Figure 3. Mesh of FV3.
RESULTS AND DISCUSSIONS
In the present study, three fishing vessel are considered, the principal characteristic of three vessels are
given in Table 1. FV2 is the only ship which presents
nearly horizontal panels at the free-surface vicinity.
Its original geometry, as well as the modified mesh,
can be seen in Figures 1a and 1b. Figures 2 and 3,
show the meshes corresponding to FV1 and FV3
respectively.
From Figure 4, it may be noted that due to the
modification of meshes near the free surface of FV2,
numerical results improves significantly. In Figure 4
one may notice the numerical instability occurs in
the motion history computed for an incident wave
with /Lbp = 6.28 for zero speed case with respect
to the mesh given in Figure 1a. On the other hand,
Figure 5 shows the same computation, but with the
167
Figure 4. Force calculation of FV2 vessel, Left: Original

Geometry, Right: Modified Geometry.
modified geometry given in Figure 1b, where a regular

sinusoidal stable solution is obtained.
Next, motion results for unrestrained hull are presented. Results are presented in form of response
amplitude operators (RAO). In the figure captions the
acronyms W, ST and TD are defined as the results
obtained using WAMIT (only for zero speed), Strip
theory and the present Time domain scheme respectively. The time domain transfer functions are obtained
by averaging the peak to peak value of the steady
sinusoidal part of the motion history. The frequency,
heave and pitch motion are non-dimensionalized by
where is the encounter frequency, k is wave number,

x3 and x5 are heave and pitch motion amplitude, a is
wave amplitude.
The first set of results is presented for the FV1 vessel. The results are computed for zero speeds, then for
7.5 and 10 knots speed. In all figures, the legends W,
ST, TD refer to WAMIT, strip theory and time domain
results, respectively. For FV1, results are shown for
= 180 , 135 , 0 . i.e. for head, quarterly head and
following waves. To study the co-relation between
strip theory and time domain results more carefully as
speed of the vessel increases; the results of FV1 and
FV2 vessel are grouped in such a way so that effect of
speed can be observed more clearly.
Figure 56 represent the heave and pitch RAO
(Response Amplitude Operator) for the vessel FV1
in head waves. From the figures, it can be noted that
WAMIT and Time domain results compare excellently,
where a slight deviation can be noticed for strip theory
results. But all the results are within acceptable range.
Also it is interesting to see that although there is a small
deviation with strip theory results for the zero speed;
agreement becomes better for the higher speed.
For the pitch results also, time domain results are
slightly higher in the peak frequency range and also in
the low frequency range for the zero speed case. But as
speed increases, the correlation between strip theory
and time domain results becomes better. Only significant difference can be noticed at the peak frequency,
otherwise results are excellently matched.
Similar features can be noted for other heading
waves too. Figures 7 10 represent the motion RAO
for heave and pitch for the wave heading of 135 and
Figure 5. Heave response of FV1 vessel for different.

Wave heading: 180 degree. Top: 0 speed, middle: 7.5 knots,
bottom: 10 knots.
Figure 6. Pitch response of FV1 vessel for different speeds.

bottom: 10 knots.
for the following waves, i.e. for 0 . From the figures

it may be noted that, for the zero speed case, WAMIT
and time domain results agree very well for both heave
and pitch but there are some differences with strip
theory results. However in the case of 7.5 knots and
10 knot, some difference can be observed near peak
frequency range, otherwise both time domain and strip
theory results co-relate excellently for heave. In case
of pitch, some difference can be seen at the lower
frequency range and in peak frequency for the zero
speed case, but shows very good agreement for the
forward speed cases.
From the Figures 611, it can be noted for the FV1
vessel, all results match excellently for heave motion.
But for pitch motion, time domain results agree very
well with WAMIT for high frequency range, however
168
Figure 7. Heave response of FV1 vessel for different speeds.

bottom: 10 knots.
Figure 9. Heave response of FV1 vessel for different speeds.

bottom: 10 knots.

bottom: 10 knots.

bottom: 10 knots.
small deviation can be found near pick frequency and

low frequency limit. But the overall behaviour of the
motion results is similar. Also the agreement with strip
theory results improves with speed.
Next set of results is shown for the vessel FV2. It
may be noted that all the computation for FV2 vessel is carried out using the modified mesh, therefore
from the figures one can conclude that small geometric
changes dont affect to the final response. Figure 11
14 represents the heave and pitch RAO results for the
FV2 vessel. The results are computed for zero speed,
7.5 knots speed and 10 knot speed which is service
speed for the vessel. The computations are carried out
for head waves and quarterly head waves. Comparisons with WAMIT and strip theory results are also
presented for different speed and angles respectively.
From the plot, it can be observed that time domain
results gives higher value in comparison to the strip

theory results. Also unlike with FV1, the agreement
between these two theories is better for the zero speed
case in comparison to forward speed case.
Figures 1112 have the results for heave and pitch
response function for wave heading 180 . Again good
agreement between all theories can be observed for
the zero speed case. Also from figures 11 and 12, (top
most figures) it may be noted that, for heave, strip
theory results are very close to the other two results.
The correlation between strip theory results with other
two results is much better in comparison to the zero
speed results of FV1 vessel. But as speed increases,
the agreement between strip theory results and time
domain results deteriorate.
The difference is more acute in case of pitch. In
Figure 12, it can be noted that in the lower frequency
169
Figure 11. Heave response of FV2 vessel for different

speeds. Wave heading: 180 degree. Top: 0 speed, middle:
7.5 knots, bottom: 10 knots.

bottom: 10 knots.
zone, the agreement between strip theory results and

time domain results is comparatively poor; also time
domain results shows a clear peak, but strip theory
results doesnt show such clear peak; but needless
to say, the overall behaviour is similar. Also the
agreement is better in the high frequency range.
The difference between strip theory and time
domain is further increased for the quarterly head wave
results. From Figure 1314, it can be noted that even
for the heave results, the difference is larger. It is interesting to note that for the FV2 vessel, strip theory
compares very well for the zero speed case (top figure
of 14 and 15) with the time domain code, but not for
the forward speed case, especially for heave motion.
It means that the nature of agreement is completely
Figure 13. Heave response of FV2 vessel for different

speeds. Wave heading: 135 degree. Top: 0 speed, middle:
7.5 knots, bottom: 10 knots.

bottom: 10 knots.
different from that of the FV1 vessel. In case of the

FV1 vessel, time domain and strip theory results are
comparing well for the forward speed especially for
heave. From this observation, it may be noted that for
strip theory computations, both slenderness property
and forward speed effect play significant role in heave
response. The vessel FV2 is relatively slender than that
of FV1 and therefore motion results of FV2 for zero
speed case agrees relatively well with other two codes.
On the other hand, forward speed results obtained by
strip theory agree well with time domain results may
be because of forward speed effect. Therefore it may
be concluded that for the strip theory computations,
slenderness property is important when speed of the
vessel is low or zero, but forward speed effect is more
important when high speed is considered.
170
Figure 15. Heave response function for FV3, Zero Speed

Top: heading 180 , bottom: heading 135 .
Figure 19. Heave response function for FV3, 10 knot Speed

Figure 16. Pitch response function for FV3, Zero Speed

Figure 20. Pitch response function for FV3, 10 knot Speed

behaviour is similar and consistent throughout all the

computations.
From all the figures, one can notice that like the FV1
vessel, strip theory results compare relatively well with
time domain code for forward speed, but have some
deviation for zero speed results. From the Table 1, it
may be noted that the Length/Beam ratios of the FV3
vessel is very similar to FV1, therefore motion results
also showing similar kind of agreement between all
three. Looking at the results, it may be conclude that
the slenderness of the ship is important for the heave
motions when zero speed is considered, but forward
speed effect dominates over the slenderness property.
Figure 17. Heave response function for FV3, 7.5 knot Speed
Figure 18. Pitch response function for FV3, 7.5 knot Speed
The same kind of characteristic of the motion results

can be seen for the FV3 vessel. The same set of results
are obtained for the FV3 vessel as well. The geometry
of FV3 is more complicated than that of the other two
ships; also it has a large bulbous bow. Figures 1516
show results for zero speed case and Figures 1720
again for the forward speed case. Again, for the FV3
vessel, excellent agreement between the time domain
and WAMIT is found out for the zero speed case,
but for the forward speed, the correlation with strip
theory results is good and most importantly overall
CONCLUSIONS
In the present study, a time domain panel code is used

to study the seakeeping performance of fishing vessels and to compare with well validated WAMIT code
and strip theory code for zero speed and with strip
theory code for forward speed. From the results, it is
observed that the present method agrees very well with
WAMIT for zero speed which indicates the efficiency
and robustness of the proposed scheme. It also shows
good agreement with strip theory code in most cases.
Moreover, large deviations are observed near the peak
frequency range, although the overall behaviour is similar and consistent. Computations were carried out
considering non-simplistic hull forms and the results
demonstrate the robustness of the present code.
From the results presented, it may be concluded
that for heave motions, the slenderness property is
important for the strip theory in low speed situations,
but the forward speed effect becomes the dominating
factor as the speed increased.
171
ACKNOWLEDGEMENTS
This work has been performed within the project
SADEP-Decision support system for the safety of
fishing vessels subjected to waves, which has been
financed by the Foundation for Science and Technology (Fundao para a Cincia e a Tecnologia),
from the Portuguese Ministry of Science and Technology, under contract PTDC/EME-MFE/75233/2006.
The first and second authors have been funded by the
under the contracts SFRH/BPD/66845/2009 and
SFRH/BD/64242/2009.
REFERENCES
Clement, A.H., 1998, An ordinary differential equation for
green function of time-domain free-surface hydrodynamics, Journal of Engineering mathematics, vol. 33,
pp. 201217.
Datta, R. and Sen, D., 2007, The simulation of ship motion
using a B-spline based panel method in time domain,
Journal of Ship Research, vol. 51(3), pp. 267284
Datta R, Rodrigues J. M, Guedes Soares C. 2011, Study of
the Motions of Fishing Vessels by a Time Domain Panel
Method, Ocean Engineering, vol. 38, pp. 782792
Liapis, S. J. and Beck, R. F., 1985, Seakeeping computations using time domain analysis, Proceedings, 4th
International Conference on Numerical Ship Hydrodynamics, National Academy of Sciences, Washington, D. C,
pp. 3454.
Lin, W. M. and Yue, D., 1990, Numerical solution for large
amplitude ship motions in the time domain. Proceedings of 18th ONR Symposium on Naval Hydrodynamics,
National Academy Press, Washington D.C, pp. 4166.
Nakos, D. E., and Sclavounos, P. D., 1990, Ship motion by a
three dimensional Rankine panel method, Proceedings of
8th ONR Symposium on Naval Hydrodynamics, National
Academy Press, Washington D.C, pp. 2140.
Nakos, D. E., Kring, D. E. and Sclavounos, P. D., 1993,
Rankine panel method for time-domain free surface
flows., Proceedings, 6th International Conference on
Numerical Ship Hydrodynamics., Iowa City, Iowa.,
University of Iowa.
Salvesen, N., Tuck, E. O. and Faltinsen, O., 1970, Ship
motions and sea loads. Trans. SNAME, vol. 78, 250287.
Sen, D., 2002, Time domain computations of large amplitude
3D ship motions with forward speed, Ocean Engineering,
vol. 29, pp. 9731002.
Wu, G.-X and Eatock Taylor, R. 1987, A Greens function form for ship motion at forward speed. International
Shipbuilding Progress, vol. 34 (398), pp. 189196.
172

Influence of sea current on manoeuvring of a surface autonomous model

V. Ferrari, L. Moreira, S. Sutulo & C. Guedes Soares
ABSTRACT: A navigation and control algorithm for the path following task using heading and speed controls
is described. The navigation algorithm is based on the waypoint scheme using the line-of-sight projection method
and is applied to simulating the manoeuvring motion of an autonomous ship model. A set of simulation tests was
realized in order to evaluate the influence of the sea current on the path tracking process varying the model speed,
and the current speed and direction. It is demonstrated how the sea current may interfere with the performance
of the vehicle.
INTRODUCTION
In recent years there has been a growing interest in

the automated maritime navigation in particular in
application to autonomous marine vehicles (AV), both
surface (ASV) and underwater (AUV). Such vehicles allow access to otherwise unreachable regions
and can, in principle, simplify the task of acquiring
hydrographical data fast and cost-effectively without placing human lives at risk. Envisioned missions
include environmental monitoring, inspection of estuaries and harbours, pipeline inspections, geological
and biological surveys, marine habitat mapping and
military applications.
Some specific missions require the vehicle to follow the desired path, with minimal deviations. These
cases include, for instance, the bathymetric survey and
mine hunting where it is required that the followed
path reliably cover a pre-defined area. The vessel has
to follow a fixed trajectory that must entirely cover
a predefined area. Even minor deformations of the
preset path can significantly compromise the mission
leaving non-inspected areas.
Deviations from the desired trajectory may be
caused by exogenous factors like the current, wind and
sea waves but also additional errors can arise from
the non-perfect steering. The planned desired path
cannot be always followed precisely due to physical
constraints. For instance, a polygonal trajectory with
angular points is often planned in navigation but such
path cannot be followed exactly by any real craft with
finite inertia.
The present paper is concentrating on numerical
study of the path-tracking algorithm developed earlier
by Moreira et al. (2005, 2007) but here in presence of
a steady uniform current.
The navigational software was devised on the basis
of the waypoint navigation scheme using the Line-ofSight (LOS) projection algorithm. Relatively recently
Figure 1. The Esso Osaka ship model.
this concept was applied to automation of surface

ships navigation tasks by Healey and Lienard (1993),
McGookin et al. (1998) and Fossen et al. (2003) to
name just a few. Using the LOS algorithm it was possible to calculate the desired heading angle serving as
input to a PID controller while the speed of the ship can
be regulated independently. The navigation and control
code was developed using the Matlab application.
The numerical simulations were performed for a
scaled model of the well-known benchmark tanker
Esso Osaka. This ship is characterised by a great
amount of available information on the hull shape,
manoeuvring mathematical models and manoeuvring
performance and the simulated scaled model (Moreira
and Guedes Soares, 2011) is actually available for
experimental work (Figure 1) which influenced the
choice of the modelled object.
The manoeuvring mathematical model used in simulation is that developed by Abkowitz (1980) on the
basis of full-scale trials of Esso Osaka followed by
the system identification procedure.The scale factor of
173
Table 1.
Esso Osaka model main particulars.
Length overall
Breadth
Depth
Draught (during the tests)
Block coefficient
Displacement
3.430 m
3.250 m
0.530 m
0.283 m
0.063 m
0.831
0.0902 m3
the physical model is = 100 and its main particulars

are given in Table 1.
2
MATHEMATICAL MODEL
Figure 2. Definition of LOS vector.
A somewhat more detailed description of the used

mathematical model presented here in brief can be also
found in Moreira et al. (2005, 2007).
2.1
Equations of motion
It is supposed that the ship has 3 degrees of freedom

and her instantaneous position in the horizontal plane
is described by her advance x and transfer y defined
in the Earth-fixed frame shown in Figure 2 and by
the heading angle . A steady homogenous current
is present with the velocity Uc characterized by the
magnitude Uc and the direction . Then, if u and v are
respectively the velocities of surge and sway over the
ground defined as projections of the ship velocity U
on the body-fixed axes and ur , vr are their analogues
with respect to water, the following relations are valid:
The speed
over the ground (SOG) is then defined as
U = u2 + v2 and the equations describing the ship

kinematics are:
u , v and r are the so-called acceleration derivatives

and f1,2,3,4 are originally multivariate cubic polynomials depending on ur , vr , r and on the rudder angle
with constant coefficients estimated for the Esso
Osaka ship by means of an identification procedure
after full scale test. Numerical values of these coefficients and also of the acceleration derivatives are
given in (Abkowitz 1980) in the non-dimensional form
and they were used in the present study for simulation of the manoeuvring motion of the scaled model
of the same ship. A possible scale effect was considered insignificant for the purpose of the present
investigation.
Dynamics of the steering gear is basically described
by the standard first-order aperiodic plant i.e.
where is the rudder angle, c is the rudder order, and

TL is the time lag typically varying in the interval 35
seconds for a full-scale ship while for the model it was
assumed TL = 0.5 s. The mathematical model for the
gear included also two saturation-type nonlinearities:
the absolute value of the rudder angle was limited by
35 and the model deflection rateby 23.3 /s which
corresponds to 2.33 /s in full scale.
2.2 Line-of-sight algorithm
where r is the angular velocity of yaw.

The equations for ship dynamics were written by
Abkowitz (1980) in the form:
where A = Iz Nr , B = mxG Yr , C = m Yv , D =
mxG Nv and E = AC BD are the inertial aggregates, Iz is the ship moment of inertia, m is the ship
mass, xG is position of the centre of mass with respect
to the midship plane, X , Y and N with the subscripts
The control strategy does not take into account presence of wind and/or current and is based on the
waypoint guidance algorithm by LOS presented by
Fossen et al. (2003) and modified by Moreira et al.
(2005, 2007). Figure 2 illustrates the LOS guidance
principle and defines the LOS vector as traced from
the vessel coordinate origin (x, y) to the intersecting
point on the path (xlos , ylos ) at the distance of n ship
lengths Lpp ahead of the vehicle.
In many applications the LOS vector is taken as
a vector from the body-fixed origin (x, y) to the next
way-point (xk , yk ). This suggests that the set point for
the heading controller should be chosen as:
174
where (x, y) are co-ordinates of the observed ship

position and
which produces the standard heading lying in the

interval [0, 2].
A 3rd order reference model generates the necessary
signals required by the heading controller as well as is
smoothing the discontinuous way-point switchings to
prevent rapid changes in the desired yaw angle fed to
the controller.
The drawback of a LOS vector pointing to the next
way-point is that a way-point located far away from
the ship will result in large cross-track errors in the
presence of wind, current and wave disturbances.Thus,
the desired yaw angle can be computed as:
where the LOS coordinates (xlos , ylos ) satisfy:
The term rmin represents the radius of the optimal circle around the ship introduced for improving
the convergence of the algorithm. It is calculated
dynamically as a function of the ship and waypoints
coordinates.
The pair (xlos , ylos ) can be obtained from (8). When
the vehicle is moving along the path, a switching mechanism for selecting the next way-point is needed. The
waypoint (xk+1 , yk+1 ) can be selected depending on
whether the ship lies within a circle of acceptance with
radius R0 around waypoint (xk , yk ), as illustrated in
Figure 2. Moreover, if the vehicle positions (x(t), y(t))
at the time t satisfies the condition
the next way point (xk+1 , yk+1 ) should be selected, i.e.,

k should be incremented to k: = k + 1. This radius can
be static i.e. the same for all waypoints, or may vary
dynamically between them and be stored together with
the way points in the waypoint structure. A guideline can be to choose R0 = 2Lpp (Fossen, 2002). In
this approach both the ship enclosing circle radius
rmin + Lpp and the radius of the circle acceptance R0
can be viewed as additional design parameters to be
used when optimizing the path following behaviour of
a ship using LOS guidance.
Figure 3 shows the situation at a given time t. The
LOS point (xlos , ylos ) is the intersection of the desired
Figure 3. Illustration to LOS Algorithm.
path between the waypoints and of the circle centered

at the model position (x(t), y(t)) and with the radius
rmin + Lpp . The resulting LOS vector, represented in
the figure by the arrow, has the components x and
y and is then used to calculate the new reference
heading angle. This procedure is calculated at each
time-step, therefore for every ship position there is a
corresponding LOS point lying on the desired path.
Once ref is calculated, the third-order filter is
used for obtaining the actual desired heading d . The
autopilot must have both course-keeping and turning
capabilities which can be obtained in one design by
using a reference model to compute the desired states
d , rd , and rd needed for course-changing (turning),
whereas course-keeping, that is the straight run regime
d = const can be treated as a special case of turning.
The transfer function of the mentioned filter is
where and n are the filter parameters whose values

are chosen as: n = 0.45 rad/s and = 1.
d and
d
Notice that lim d (t) = ref and that
t
remain smooth and bounded even for a discontinuous

ref .
2.3
Heading controller
Assuming that both and r are known, for instance

measured by a compass and a rate gyro, the PID heading controller for full state feedback is given by Fossen
(2002):
where
175
Table 2.
Heading PID constants.
KN (Nomoto gain constant)

TN (Nomoto time constant)
KP
KI
KD
of the effective thrust and the contribution of the hull

and rudder is given by
0.1705 s1
7.1167 s
41.7422
0.0587 s1
77.6189 s
is a feed-forward term based on the first-order Nomoto

model and
where CT is the total ship drag coefficient; S is the wetted surface area of the ship, c is the weighted average
flow speed over the rudder given by
is the PID regulator per se. The first and second time
derivatives of the desired heading are estimated as
with t time step used in the simulation.

The Nomoto parameters were found using Journe
(2001) and Clarke (2003) method of analysing zig-zag
manoeuvres. The PID gains were estimated using the
pole-placement method applied to the same Nomoto
model:
where n = 1 rad/s and = 1.

More details on the determination of the gains
and time lags can be found, for instance, in Moreira
et al. (2007) and the resulting values are reported in
Table 2:
The PID constants presented in Table 2 were used
in all simulations independently of the speed of the
model.
2.4
c0 is the value of c corresponding to the straight run

propulsion point; Ap is the propeller area; AR is the
rudder area; w is the wake fraction; uA is the induced
axial velocity at infinity behind the propeller disk
given by
and where e is the geometric rudder attack angle (i.e.

without hull and propeller influence); KT is the thrust
coefficient; D is the propeller diameter.
To obtain the axial induced velocity at the rudder,
the second term in (21) must be multiplied by the factor
k depending on the distance between the propeller and
the rudder (Abkowitz, 1980).
The ordered acceleration is governed by the PID
control law (Fossen, 2002)
and the required thrust increment is then
Speed controller
The surge equation from the set (2) using (1) can be
re-written as
where the used gains are: Kp = 0.15, Ki = 1.0 105 .

A 2nd-order filter with the transfer function
where the meaning of the function f1 () is somewhat

different from that originally introduced by Abkowitz
(1980):
where n = 0.25 rad/s and = 1, was applied to the

desired speed signal to make the transitions smoother.
where the effective equilibrium thrust is
1 , 2 , 3 are constant parameters, n is the equilibrium

propeller rotation frequency, T is the controlled part
The first set of simulation runs presumed three different desired speeds Ud with zero speed of the current. The model always starts its manoeuvre with the
initial speed of 0.5 kn and is ordered to reach at the
176
NUMERICAL STUDY OF THE INFLUENCE

OF MODEL SPEED
Table 3.
Model speed simulations.
Simulation
n
Model
speed
[m/s]
Model
speed
[kn]
Equivalent
ship speed
[kn]
Froude
number
1
2
3
0.26
0.31
0.36
0.5
0.6
0.7
5
6
7
0.046
0.055
0.065
Figure 6. Heading angle and speed errors.
Figure 4. Model trajectories with different speeds.
Figure 7. Time history of desired heading.
Figure 5. Zoom of the models trajectories
first waypoint the desired speed, indicated in Table 3,

and to further maintain it. The initial heading angle is
arbitrarily set to 200 in order to test robustness of the
LOS algorithm and the PID controller.
The resulting trajectories are presented in Figure 4.
It is seen that at the lowest speed the model reaches
the desired path with delay. On the other hand, at 0.7 kn
the model comes closer to waypoints but shows overshoots, The best quality of tracking was obtained at the
intermediate model speed 0.6 kn (Fig. 5).
The time histories of the heading angle and speed
errors err = d and Uerr = U Ud are shown in
Figure 6. Except for the initial settling interval, the
speed error never reaches significant values proving
high quality of the speed controller. Behaviour of

the heading error is more complicated. This error
increases every time the model approaches a waypoint as it has to turn to the next one. Thus, analysing
peaks of the heading error it is possible to detect
when the vehicle reaches a waypoint. Naturally, these
moments corresponding to different speed values are
diverging with the advance of the model as is also
seen in Figures 7 and 8, where time histories for the
desired heading and the yaw angular velocity and
acceleration are presented. It is seen that when the
model reaches a waypoint, the desired heading does
not change abruptly thanks to the applied filter. As a
consequence, the desired rate of yaw and its derivative
are bounded to acceptable values.
Figure 9 represents time histories of the rudder
angle and rudder rate for the 0.6 kn simulation. The
rudder angle is saturated at 35 and reaches these
extremal values when the model has to move towards
the next waypoint. Similarly it can be seen that the
rudder rate is saturated at 23.3 /s. These saturations
do not indicate in this case to an overdesigned controller but, on the contrary, the chosen values of the
control gains lead to an approximate reproduction of
the bang-bang time-optimal control.
177
Figure 8. Time histories for desired rate of yaw and yaw

acceeration.
Figure 10. Model trajectories varying with higher speeds.

Table 5.
Sea current speed simulations.
Sea
current
Simulation speed
No.
[m/s]
Sea
current
speed
[kn]
Sea
current
Model Relative
direction speed current
[ ]
[kn]
speed
7
8
9
10
0.47
0.68
0.87
1.07
0
0
0
0
0.25
0.35
0.45
0.55
0.6
0.6
0.6
0.6
0.78
1.13
1.45
1.78
Figure 9. Rudder and rudder rate time histories for model

speed 0.6 kn.
Table 4.
Higher model speed simulations.
Simulation
No.
Model
speed
[m/s]
Model
speed
[kn]
Equivalent
ship speed
[kn]
Froude
number
4
5
6
0.57
0.67
0.77
1.1
1.3
1.5
11
13
15
0.100
0.118
0.137
Since the maximum speed of the Esso Osaka is

16 kn, corresponding to a model speed of 1.6 kn, three
additional simulations were run in order to see the
behaviour at higher speeds, as shown in Table 4 and
Figure 10. The resulting trajectories here are similar
to those obtained at 0.7 kn case except for expectedly
increased overshoots.
4 NUMERICAL STUDY OF THE INFLUENCE
OF SEA CURRENT AT LOW SHIP SPEED
In all simulations presented in this section the desired
model speed with respect to water Urd was 0.6 kn
Figure 11. Trajectories varying with sea current speed.
as yielding the best results without current while

the current velocity Uc was varying from 0.25 to
0.55 m/s which corresponded to the relative current
speed Uc /Urd varying from 0.78 to 1.78 (Table 5).
The results of simulations are presented in Figures 11
through 13.
Expectedly, the over shoots became much more
pronounced in the presence of the current and at the
highest current speed the vehicle is wobbling on the
178
Table 6.
Sea direction simulations.
Sea
current
Simulation speed
No.
[m/s]
Sea
current
speed
[kn]
Sea
current
Model Relative
[ ]
[kn]
speed
11
12
13
14
1.07
1.07
1.07
1.07
0
90
180
270
0.55
0.55
0.55
0.55
0.6
0.6
0.6
0.6
1.78
1.78
1.78
1.78
Figure 12. Zoom of the model trajectories.
Figure 14. Trajectories varying with sea current direction.
Figure 13.
speed.
Heading angle error varying with sea current
course parallel to the x-axis. With such heavy overshoots the heading error can reach very high values,
sometimes up to 90 (Figure 13).

OF SEA CURRENT DIRECTION
In these sets of simulations the highest speed of the

current magnitude is kept constant and its direction
is varying. The parameters used in the simulations
are listed in Table 6 and the trajectories are plotted
in Figure 14 where = 0 corresponds on the plot to
the current directed from right to left.
The model was having problems in keeping straight
course in sailing against the current. For instance,
when = 90 the model moves upwards (on the
plot) heavily suffering from overshooting while this
was much less pronounced when the vehicle was moving downwards. On the other hand, tracking errors
are larger in the following and lateral current.

OF SEA CURRENT AT HIGH MODEL SPEED
The simulation plan of the present section is similar

to that of Section 4 but the required model speed was
much higher, i.e. 1.3 kn, which corresponded to the
full-scale service speed 13 kn (Table 7). Results of
these simulations are shown in Figures 15 and 16.
In all cases pronounced over shoots happened at all
current velocities near each waypoint but in general
the behaviour of the model was better than at lower
model speed.
In general, lower sea current velocities permit using
lower speed of the vehicle. When the current becomes
faster, the model can partly lose controllability suffering from path overshoots. These overshoots maintain
at higher speed of the vehicle but the used PID controller permits efficient recover resulting in smoother
paths.
Although the obtained trajectories are hardly the
best ones, as the used PID control law was not optimised, they are close enough to desired paths. Obviously, further growth of the current speed would result
in complete loss of controllability even at maximum
model speed.
It is therefore important to know how the model
and the controller behave at different working and
environmental conditions, for instance using the
Uc /Urd ratio. Then, the gain scheduling technique
could be used for varying control gains to adapt them
179
Table 7.
leads to increased operation time and cost, and in some

cases, such as the bathymetry recording or mine detection, even to the mission failure. However, the obtained
results must be treated as preliminary and further
studies including combined influence of the current,
wind and sea waves, as well as working out better
control strategies are envisaged.
High model speed simulation.
Sea
current
Simulation speed
No.
[m/s]
Sea
current
speed
[kn]
Sea
current
Model Relative
[ ]
[kn]
speed
15
16
17
18
0.47
0.68
0.87
1.07
0
0
0
0
0.25
0.35
0.45
0.55
1.3
1.3
1.3
1.3
0.36
0.52
0.67
0.82
ACKNOWLEDGEMENTS
This work has been conducted within the project
self-propelled models, which is being funded by the
(Fundao para a Cincia e a Tecnologia) under contract PTDC/TRA/74332/2006. The work of the second
author has been supported by research fellowships of
the Portuguese Foundation for Science and Technology (Fundao para a Cincia e a Tecnologia) under
contract SFRH/BPD/48088/2008.
REFERENCES
Figure 15. Trajectories varying with sea current speed and

high speed model.
Figure 16. Heading angle error varying with sea current

speed and high model speed.
to altered conditions. Also, changing mission strategy

or even aborting it could be preferred when it is known
in advance that the vehicle will not be controllable.
7
CONCLUSIONS
It was demonstrated that the sea current may influence considerably performance of the navigation and
control system in path-tracking regimes resulting in
parasitic wobbling with respect to the planned path
and large overshoots. Lower path-tracking capability
Abkowitz, M. A. (1980), Measurement of Hydrodynamic Characteristics from Ship Maneuvering Trials by

System Identification, SNAME Transactions, Vol. 88,
pp. 283318.
Clarke, D. (2003), The Foundations of Steering and Ma
noeuvring, Proceedings of 6th Conference on Maneuvring and Control of Marine Crafts (MCMC2003),
Girona, Spain, pp. 216.
Fossen, T. I. (2002). Marine Control Systems: Guidance,
Navigation and Control of Ships, Rigs and Underwater
Vehicles. Trondheim, Norway: Marine Cybernetics AS.
Fossen, T. I., Breivik, M. and Skjetne, R. (2003). Line-OfSight Path Following of Underactuated Marine Craft,
Proceedings of the 6th IFAC Conference on Maneuvering and Control of Marine Crafts (MCMC2003), Girona,
Spain.
Healey, A. J. and Lienard, D. (1993). Multivariable Sliding Mode Control for Autonomous Diving and Steering
of Unmanned Underwater Vehicles.IEEE Journal of
Oceanic Engineering, Vol. 18, Issue 3, pp. 327339.
Journe, J. M. J. (2001). A Simple Method for Determining the Manoeuvring Indices K and T from Zigzag
Trial Data. DUT-SHL Technical. Report 0267. Delft,
Netherlands.
McGookin, E.W., Murray-Smith, D.J., Lin Y. and Fossen,
T.I. (1998). Ship Steering Control System Optimization
using Genetic Algorithms. Journal of Control Engineering Practice, 8, 429443.
Moreira, L. and Guedes Soares, C. (2003). Dynamic Model
of Maneuverability using Recursive Neural Networks,
Ocean Engineering, Vol. 30, Issue 13, pp. 16691697.
Moreira, L. and Guedes Soares, C. (2011) Autonomous
Ship Model to Perform Manoeuvring Tests, Journal of
Maritime Research, Vol. 8, Number 2, pp. 2946.
Moreira, L., Fossen, T. I. and Guedes Soares, C. (2005).
Modeling, Guidance and Control of Esso Osaka
Model. Proceedings of the IFAC World Congress, Prague,
Czech Republic, 48 July.
Moreira, L., Fossen, T.I. and Guedes Soares, C. (2007). Path
Following Control System for a Tanker Ship Model,
Ocean Engineering, Vol. 34, Issues 1415, pp. 2074
2085.
180

Interaction of gravity waves with moored flexible floating membrane

D. Karmakar & C. Guedes Soares
ABSTRACT: The interaction of oblique incident waves with a moored finite floating membrane is analyzed for
finite water depth with changes in bottom topography. The solution for the physical problem is obtained based on
linearized water wave theory using the eigenfunction expansion method and orthogonal mode-coupling relation.
The numerical computations are performed for the reflection coefficient, transmission coefficients and vertical
deflections to understand the effect of mooring and changes in bottom topography on the wave interaction with
the floating membrane. Energy relation for oblique gravity wave scattering by floating membrane where there
are changes in bottom topography is used to check the accuracy of the computation. It is observed that due to
the presence of the moored floating membrane the amplitude of the transmitted wave diminishes significantly.
The changes in the reflection and transmission coefficient are also observed due to the variation in the bottom
topography and also to the changing of the tension parameter of the floating membrane. The present study shows
the effect of spring mooring on the wave propagation due to the presence of floating membrane with changes in
bottom topography.
INTRODUCTION
The flexible floating membrane breakwaters are used

for the protection of offshore facilities near the coasts.
They are found to be effective in reduction of the initial wave attacks inside the harbour area and also on
other coastal structures by attenuating the incoming
incident waves. The floating flexible membrane breakwaters are used for a wide variety of tasks, such as
protecting harbour entrances, attenuating wave height
near the floating offshore windmills, reducing the rate
of littoral drift, protecting the ocean from oil spill
and creating artificial fishing grounds for aquaculture
activities.
In recent decades, a lot of work has been carried out to understand the effect of wave interaction with floating membranes used as breakwaters.
Cho and Kim (1998) analyzed the interaction of
oblique incident waves with horizontal flexible membrane based on two-dimensional linear hydroelastic
theory. They obtained the analytic solution of the
diffraction and radiation of a submerged impermeable
horizontal membrane using both eigenfunction expansion method and by multi-domain boundary element
method. The interaction of incident monochromatic
waves with a tensioned, flexible, circular membrane
submerged horizontally below free surface was analyzed by Cho and Kim (1999). Further, Cho and Kim
(2000) investigated the interaction of waves with horizontal porous flexible membrane using both boundary
element method and eigenfuction expansion method.
They observed that the wave blocking performance of
the horizontal porous membrane can be enhanced by
using proper porous material. Yip et al. (2001) studied

the oblique gravity wave scattering by multiple floating membranes in the presence and absence of end wall
using the matched eigenfunction expansion method.
There has also been considerable progress in the
literature on the propagation of gravity waves over
variable bottom topography. Miles (1967) investigated
the diffraction of gravity waves by a shelf using variational approximations to determine the effects of
trapped modes on the resonant width of a shelf. Smith
(1987) studied the refraction due to the interaction of
surface waves with change in step bottom topography and change in velocity parallel to the step. The
transmission of water waves over step bottom topography using a Greens integral equation was analyzed
by Rhee (1997, 2001). The result obtained using this
integral equation method was observed to be consistent with the result obtained previously using matched
eigenfunction expansion method by Newman (1965)
and variational formulation by Miles (1967). Using
integral transform approach, Ehrenmark (1998) analyzed the classical problem of oblique wave incidence
on a plane beach.
In the literature, very few studies were carried out on
the wave propagation over variable bottom due to the
presence of floating flexible membrane used as breakwater. Recently, Karmakar and Sahoo (2008) analyzed
the scattering of gravity waves with floating membrane
due to changes in step bottom topography.
In the present paper, the oblique wave interaction
with a finite floating membrane connected with spring
moorings is investigated based on small amplitude
water wave theory including the effect of changes in
181
where Re denotes the real part and l is the component of the wave number along z-direction. The spatial
velocity potential j (x, y) for j = 1, 2, 3, satisfies the
Helmholtzs equation given by
Figure 1. Schematic diagram for moored floating

membrane.
bottom topography. The numerical computations are

performed for the reflection coefficient, transmission
coefficient and vertical deflections to understand the
effect of changes in the bottom topography on wave
interaction with the moored floating membrane. A
detailed analysis on the various values of the tension
parameter, spring stiffness, length of membrane and
water depth for wave interaction with moored floating
membrane is performed.
2
In this section, the boundary value problem associated

with wave scattering by a finite floating membrane
connected with spring moorings at its ends including changes in bottom topography is formulated and
described in detail.
2.1
2
where xy
= (x2 + y2 ).
In the open water region (, a) (0, ), the
free surface boundary condition is given by
where = 2 /g. Combining the linearized kinematic

boundary condition and membrane equation (see
Karmakar and Sahoo (2008)), the linearized membrane covered boundary condition is obtained as
where Q = T/(w g ms 2 ), K = w 2/(w g ms 2 ),

T is the tension of the membrane, ms is the mass
per unit area, w is the density of water and g is the
acceleration due to gravity.
It is considered that there is an abrupt change in
bottom topography. So the bottom boundary condition
is given by
General boundary value problem
The problem is analyzed in the three dimensional

Cartesian co-ordinate system with the x z being
taken as horizontal and the y-axis being vertically
downward positive with the fluid occupying the
regions < x, z < and 0 < y < hj . It is assumed
that a finite membrane is floating on the undisturbed
water surface y = 0 and < x, z < .
The fluid domain is divided into three sub-domains,
(0 < x < , 0 < y < h1 , < z < ) referred as
region 1, (a < x < 0, 0 < y < h2 , < z < )
referred as region 2 and ( < x < a, 0 < y < h3 ,
< z < ) referred as region 3 as shown in Fig. 1.
Hereafter, the subscripts will denote the parametric
values in the respective fluid regions. A monochromatic surface gravity wave is obliquely incident at
an angle on the membrane at x = 0, y = h1 which
then propagates through the membrane at a < x < 0,
y = h2 and then gets transmitted at x = a, y = h3 .
Assuming that the fluid is inviscid, incompressible
and the motion is irrotational and simple harmonic
in time with angular frequency , the velocity potential j (x, y, z, t) and surface elevation j (x, z, t) are
expressed as
where n is the outward drawn normal. In addition,

across the interface between the membrane and the
free water surface x = 0, a, 0 < y < h2 , j = 1, 2 the
continuity of velocity and pressure yields
It is assumed that the membrane is anchored with

springs at the ends x = 0, a. Thus, at the edge of
the floating membrane, the restoring forces in the
spring must be equal to the transverse component of
the tension (as in Magrab 1979), which is derived in
terms of the velocity potential after using the kinematic
condition, and is given by
where Tsj is the tension of the membrane and ksj is the

spring constant. Further the velocity potentials satisfy
the far field radiation condition given by
and
182
where R0 and T0 are the complex amplitudes of the

reflected and transmitted waves and j0 for j = 1, 3 are
the wave numbers of the incident and transmitted wave
satisfying the dispersion relation
with l = 10 sin , j02 = kj02 + l 2 . In the next section

the solution procedure for wave scattering by finite
floating membrane will be analyzed in detail.
2.2
with respect to the orthogonal mode-coupling relation

as defined by
where
Method of solution
The velocity potentials j (x, y), j = 1, 2, 3 in each of

the three regions are given by
for m = n = 0 with Cn and Cn for n = 1, 2, . . . are
obtained by substituting jn = ijn . In order to determine the unknown coefficients, the mode-coupling
relation (13b) is applied on the velocity potential
2 (0, y) and the eigenfunction f2m (y) along with the
continuity of pressure as in Eq. (5) across the vertical
interface x = 0, 0 < y < h2 and the edge condition (6)
to obtain
where Rn , n = 0, 1, 2 . . ., An , Bn , n = 0, I , II , 1, 2 . . .
and Tn , n = 0, 1, 2 . . . are the unknown constants to be
determined. The eigenfunctions fjn (y)s for j = 1, 2, 3
are given by
for m = 0, I , II , 1, 2, . . ..
Further, using the orthogonal property of the eigenfunction f2m (y) as in Eq. (12) and the expressions of
the velocity potentials as in Eq. (9) yields
where jn are the eigenvalues and satisfy the dispersion

relation
with jn = ijn for j = 1, 2, 3, n = 1, 2 . . .. In Eq. (11a)

the dispersion relation has one real root j0 and infinite
numbers of purely imaginary roots jn for n = 1, 2 . . ..
In Eq. (11b) the dispersion relation has one real root
j0 and two complex roots jn for n = I , II of the
form i. In addition, there are infinite numbers
2
of purely imaginary roots jn with 2jn = jn
l 2 for
n = 1, 2 . . .. It may be noted that the eigenfunctions
fjn (y)s satisfy the orthogonality relation as given by
for m = 0, I , II , 1, 2, . . .
Again, applying the mode-coupling relation (13a)
on 1x (0, y) and f1m (y) and utilizing the continuity of
horizontal velocity across the vertical interface x = 0,
0 < y < h2 as in Eq. (5) leads to
for m = 0, 1, 2, . . ..
183
Further using the orthogonal property of the eigenfunction f1m (y) as in Eq. (12) and the expressions of
the velocity potentials as in Eq. (9) yields
1 for m = 0,
k = ijm for
0 for m = 1, 2, 3, . . . , jm
j = 1, 2, 3, m = 1, 2 . . . and kjn = ijn for j = 1, 2, 3,
n = 1, 2 . . . Once again, applying the mode-coupling
relation (13b) on 2 (a, y) and f2m (y) and utilizing
the continuity of horizontal velocity across the vertical interface x = a, 0 < y < h2 as in Eq. (5) along
with the edge condition as in Eq. (6) leads to
where m =
for m = 0, I , II , 1, 2, . . ..
Using the orthogonal property of the eigenfunction
f2m (y) as in Eq. (12) and the expressions of the velocity
potentials as in Eq. (9) yields
Figure 2. Reflection and transmission coefficient Kr and

Kt versus angle of incidence for various values of a/h1
with h3 /h1 = 1.0, h2 /h1 = 0.5, T /gh22 = 0.1, ks1 = 0 N m1 ,
ks2 = 103 N m1 , = 5.0 s and ms = 250.0 kg/m2 .
where kjm = ijm for j = 1, 2, 3, m = 1, 2 . . .. The infinite sums present in Eqs. (15), (17), (19) and (21) are
truncated up to finite N (say) terms to facilitate the
numerical calculation which leads to a linear system
of (4N + 8) algebraic equations for the determination
of (4N + 8) unknown constants.
Once, the unknown constants R0 and T0 are determined, the reflection and transmission

coefficients
tanh 30 h3
are obtained as Kr = |R0 | and Kt = 30
T0 .
10 tanh 10 h1
It is also observed that R0 and T0 satisfy the energy
relation Kr2 + Kt2 = 1.
where k2n = i2n for m = 1, 2 . . .. Again, applying the

mode-coupling relation (13a) on 3x (a, y) and f3m (y)
and utilizing the continuity of horizontal velocity
across the vertical interface x = a, 0 < y < h2 as in
Eq. (5) gives
Using the orthogonal property of the eigenfunction

f3m (y) as in Eq. (12) and the expressions of the velocity
potentials as in Eq. (9) yields
The interaction of oblique surface gravity waves with

a floating membrane in water of finite depth is studied
by analyzing the reflection coefficient Kr , transmission coefficients Kt and the vertical deflection of
the membrane j for different values of water depth
ratio h2 /h1 , membrane length a/h1 , tension parameter T /gh22 , angle of incidence , mass per unit area
of the membrane ms and wave period . The values of the physical parameters, which are kept fixed
throughout the numerical computations, are given by
w = 1025 kg m3 and g = 9.8 m s2 . In the computation h2 /h1 is varied but h3 /h1 = 1.0 is always kept
fixed. The accuracy of the numerical results is checked
with the appropriate energy relation.
In Fig. 2, the reflection and transmission coefficient Kr and Kt are plotted versus angle of incidence
for various values of a/h1 considering h3 /h1 = 1.0,
h2 /h1 = 0.5, = 5.0 s, T /gh22 = 0.1, ks1 = 0 N m1 ,
ks2 = 103 N m1 and ms = 250 kg/m2 . It is observed
that with the increase in the membrane length
the resonating pattern in the reflection coefficient
increases and approaches to one when approaches
to 90 . It may be noted that for 40 < < 45 and
184
Figure 3. Reflection and transmission coefficient Kr and Kt

versus angle of incidence for various values of a/h1 with
h3 /h1 = 1.0, h2 /h1 = 0.5, T /gh22 = 0.1, ks1 = 103 N m1 ,
ks2 = 0 N m1 , = 5.0 s and ms = 250.0 kg/m2 .

Kt versus angle of incidence for various values of
h2 /h1 with h3 /h1 = 1.0, a/h1 = 1.0, T /gh22 = 0.1, = 5 s,
ks1 = 0 N m1 , ks2 = 103 N m1 and ms = 250.0 kg/m2 .
Figure 4. Reflection and transmission coefficient Kr and Kt

versus angle of incidence for various values of a/h1 with
h3 /h1 = 1.0, h2 /h1 = 0.5, T /gh22 = 0.1, ks1 = 103 N m1 ,
ks2 = 103 N m1 , = 5.0 s and ms = 250.0 kg/m2 .

Kt versus angle of incidence for various values of ks1 and
ks2 with h2 /h1 = 0.5, a/h1 = 1.0, T /gh22 = 0.1, = 5.0 s and
ms = 250.0 kg/m2 .
70 < < 75 the reflection coefficient approaches to

zero for all values of the membrane length.
In Fig. 3, the reflection and transmission coefficient Kr and Kt are plotted versus angle of incidence for various values of a/h1 with h3 /h1 = 1.0,
h2 /h1 = 0.5, = 5 s, T /gh22 = 0.1, ks1 = 103 N m1 ,
ks2 = 0 N m1 and ms = 250 kg/m2 . It is observed
that when spring mooring in the transmitted wave
region is absent the zeros in the reflection coefficient is observed at 70 < < 75 for all values of the
membrane length.
In Fig. 4, the reflection and transmission coefficient Kr and Kt are plotted versus angle of incidence for various values of a/h1 with h3 /h1 = 1.0,
h2 /h1 = 0.5, = 5 s, T /gh22 = 0.1, ks1 = 103 N m1 ,
ks2 = 103 N m1 and ms = 250 kg/m2 . The reflection
coefficient is observed to be same at 0 < < 45 and
approaches to zero at = 75 for all values of membrane length. This may be due to the presence of spring
moorings at both the ends of the floating membrane.
In Fig. 5, the reflection and transmission coefficient Kr and Kt are plotted versus angle of incidence for various values of h2 /h1 with h3 /h1 = 1.0,
a/h1 = 1.0, = 5.0 s, T /gh22 = 0.1, ks1 = 0 N m1 ,
ks2 = 103 N m1 and ms = 250 kg/m2 . In this case, it
is observed that the reflection coefficient is higher for
h2 /h1 = 0.5 and Kr approaches to one as approaches
to 90 . The absence of spring mooring in the incident region increases the reflection coefficient with
the increase in water depth until h2 /h1 = 0.5 and then
again keeps on decreasing.
In Fig. 6, the reflection and transmission coefficient Kr and Kt are plotted versus angle of incidence
for various values of mooring stiffness ks1 and ks2
with h3 /h1 = 1.0, h2 /h1 = 0.5, a/h1 = 1.0, = 5.0 s,
185

Kt versus angle of incidence for various values of T /gh22
with h3 /h1 = 1.0, h2 /h1 = 0.5, a/h1 = 1.0, ks1 = 0 N m1 ,
ks2 = 103 N m1 , = 5.0 s and ms = 250.0 kg/m2 .
Figure 9. Vertical deflection j (x, z, t) with h2 /h1 = 0.5,

h3 /h1 = 1.0, a/h1 = 2.0, = 45 , T /gh22 = 0.1, = 5.0 s
ks1 = 0 N m1 and ks2 = 103 N m1 .
Figure 10. Vertical deflection j (x, z, t) with h2 /h1 = 0.5,

h3 /h1 = 1.0, a/h1 = 2.0, = 45 , T /gh22 = 0.1, = 5 s,
ks1 = 103 N m1 and ks2 = 103 N m1 .
Kt versus angle of incidence for various values of T /gh22
with h3 /h1 = 1.0, h2 /h1 = 0.5, a/h1 = 1.0, ks1 = 103 N m1 ,
ks2 = 103 N m1 , = 5.0 s and ms = 250.0 kg/m2 .
T /gh22 = 0.1 and ms = 250.0 kg/m2 . It is observed

that with the increase in the spring stiffness in each of
the ends the resonating pattern in the reflection coefficient keeps on decreasing. The zeros in the reflection
coefficient for all values of the spring stiffness is
observed within 70 < < 75 .
In Fig. 7, the reflection and transmission coefficient Kr and Kt are plotted versus angle of
incidence for various values of tension of the
membrane T /gh22 with h3 /h1 = 1.0, h2 /h1 = 0.5,
a/h1 = 1.0, = 5.0 s, ks1 = 0 N m1 , ks2 = 103 N m1
and ms = 250.0 kg/m2 . It is observed that with the
increase in the membrane stiffness parameter the
reflection coefficient keeps on increasing. It may be
noted that for lower values of T /gh22 the zeros in
reflection coefficient is observed whereas the zeros
in reflection coefficient diminishes as the values of
T /gh22 increases in the presence of spring mooring

in the reflected region.
In Fig. 8, the reflection and transmission coefficients Kr and Kt are plotted versus angle of incidence
for various values of tension of the membrane T /gh22
with h3 /h1 = 1.0, h2 /h1 = 0.5, a/h1 = 1.0, = 5.0 s,
ks1 = ks2 = 103 N m1 and ms = 250.0 kg/m2 . In this
case also the reflection coefficient keeps on increasing with the increase in the stiffness constant. It may
be noted that, for T /gh22 = 0.1 the zeros in the reflection coefficient are not observed when both the spring
moorings are present. However, in Fig. 7 the zeros are
observed for T /gh22 = 0.1.
In Fig. 9, the vertical deflections j (x, z, t) are
plotted considering T /gh22 = 0.1, ks1 = 0 N m1 ,
ks2 = 103 N m1 and = 5.0 s. In this case, the deflection in the incident region is observed to be higher than
the membrane covered region and transmitted region.
This suggests that due to the presence of membrane
and the spring mooring in the reflected region the
attenuation of wave height takes place in the transmitted region.
186
In Fig. 10, the vertical deflection j (x, z, t) are

plotted considering h2 /h1 = 0.5, h3 /h1 = 1.0, = 5 s,
a/h1 = 2.0, = 45 , T /gh22 = 0.1, ks1 = 103 N m1
and ks2 = 103 N m1 . It is found that for both ends having stiffness constant the deflection in the transmitted
region is lower than the incident region. The present
observation is similar to the result observed in Fig. 9.
However, due to presence of spring moorings at both
the ends the deflection in the incident region is lower
than the deflection in the absence of spring mooring
in the incident region.
4
CONCLUSIONS
In the present paper, oblique wave interaction with

finite floating membrane connected with spring moorings at its ends is analyzed in water of finite depth
including the effect of changes in bottom topography.
The analysis is done using eigen-function expansion
method with the application of developed orthogonal mode-coupling relation. It is observed that with
the increase in membrane length the resonating pattern the reflection coefficient keeps on increasing.
The tension parameter also observed to play an important role in the attenuation of wave height in the
transmitted region.
Numerical results suggest that for higher values of
the tension parameter the wave reflection is higher.
The detailed analysis on various water depth, membrane length, tension parameter and wavelength are
done to understand the importance of membrane in
attenuation of wave height. The results obtained in the
present study will be helpful to designers for proper
design and analysis of floating membrane as effective
breakwater.
ACKNOWLEDGEMENTS
This work was performed within the scope of the
project FLOATERSWave Interaction with Floating Structures in Shallow Water financed by the
PTDC/ECM/111242/2009.
REFERENCES
Cho, I.H. and Kim, M.H. 1998 Interaction of a horizontal
flexible membrane with oblique incident waves. J. Fluid
Mech. 367, 139161.
Cho, I.H and Kim, M.H. 1999 Wave deformation by a
submerged circular disk. Appl. Ocean Res. 21, 263280.
Cho, I.H. and Kim, M.H. 2000 Interaction of a horizontal porous flexible membrane with waves. ASCE Journal
of Waterway, Port, Coastal and Ocean Engg. 126(5),
245253.
Dingemans, M.W. 1997 Water wave propagation over uneven
bottoms, Part-I Linear wave propagation. Advanced
series on Ocean Engineering Vol. 13, World Scientific.
Ehrenmark, U.T. 1998 Oblique Wave incidence on a plane
beach: the classical problem revisited. J. Fluid Mech. 368,
291319.
Fox, C. and Squire, V. A. 1994 On the oblique reflection and
transmission of ocean waves at shore fast sea ice. Phil.
Trans. R. Soc. Lond. A. 347, 185218.
Karmakar, D. and Sahoo, T. 2008 Gravity wave interaction
with floating membrane due to abrupt change in water
depth. Ocean Engg. 35(7), 598615.
Lee, W.K and Lo, E.Y.M. 2002 Surface-penetrating flexible
membrane wave barriers of infinite draft. Ocean Engg.
29(14), 17811804.
Magrab, E.B. 1979 Vibrations of elastic structural members,
SIJTHOFF and NOORDHOFF, Alphen aan den Rijn, The
Netherlands.
Miles, J. W. 1967 Surface-wave scattering matrix for a shelf.
J. Fluid Mech. 28, 755767.
Newman, J.N. 1965 Propagation of water waves over an
infinite step. J. Fluid Mech. 23, Part 2, 399415.
Rhee, J.P. 1997 On the transmission of water waves over a
shelf. Appl. Ocean Res. 19, 161169.
Rhee, J.P. 2001 A note on the diffraction of obliquely incident
water waves by a stepwise obstacle. Appl. Ocean Res. 23,
299304.
Rhodes-Robinson, P.F. 1971 On the forced surface waves
due to a vertical wave-maker in the presence of surface
tension. Proc. Camb. Phil. Soc. 70, 323337.
Smith, J. 1987 On surface waves crossing a step with
horizontal shear. J. Fluid Mech. 175, 395412.
Yip, T.L., Sahoo, T. and Chwang, A.T. 2001 Wave scattering
by multiple floating membranes. Proc. 11th Intl. Offshore
and Polar Engineering Conference, Stavangar, Norway,
Vol. 3, 379384.
187

Review of model test techniques of local slamming on ships

Hanbing Luo & C. Guedes Soares
ABSTRACT: The interest on slamming is often being raised from the view point of ship structural safety,
as slamming affects the strength of both local and global structures. While earlier studies of slam induced
responses have dealt with the global whipping response more recent studies have dealt with local response and
various experimental studies have been conducted. Model tests are still adopted to verify and validate results of
analytical and numerical solutions. The state of art of model test techniques and results of local slamming on
ships is reviewed here. Three main problems are considered, water entry of rigid bodies into calm water, model
tests in waves, and hydroelasticity of local slamming. The challenges for future work are discussed.
INTRODUCTION
When a ship travels in rough seas, it will impact water

because of large vertical relative motions between the
ship and wave surface. The impulsive pressure is much
larger than the wave pressure, and this hydrodynamic
impact phenomenon is defined as slamming. According to the position where it occurs, slamming on ships
can be roughly categorized as bow bottom slamming,
bow flare slamming, bow stem slamming, wet deck
slamming, and also stern slamming which is raising
attention in recent years (Luo et al. 2007).
There is a considerable amount of research conducted on slamming by experimental, analytical, and
numerical simulation methods since Von Karman
(1929) and Wagner (1932). Slamming is a so complicated physical process that it is not yet fully understood. It is related with the relative motion, body
geometry, water surface profile, air cushion, hydroelasticity of structures, compressibility of water, and
so on.
The impulsive pressure loads induced by slamming
will affect the structure of ships both locally and globally. In rough seas, this impact force is so large that
many ships have reported local structural damages due
to the slamming loads, especially in heading waves
with high forward speed. For example, the tragedy of
MV Estonia in the Baltic Sea on 28 September 1994,
one of the deadliest marine disasters of 20th century,
was mainly because of the failure of the bow door
due to the severe slamming which the Ro/Ro ferry
experienced. This impulsive water impact will induce
whipping of the ship hull, which means global transient elastic vibration. Whipping response is one main
component for nonlinear wave loads on ship hulls.
Slamming can threaten the safety of ship structures.
Structural design against loads due to slamming is a
difficult task for structural designers and researchers.
They need to consider both reliability and economic

efficiency. It is not easy to accurately predict slamming loads, correspondently for the local and global
response. Equivalent or quasi-static pressure formula
for design is always used in the relevant rules of classification societies. It is still developing and being
improved these years. And when the deadrise angle is
small, hydroelasticity will be important. It was emphasized by Faltinsen (2000) that the slamming problem
must be hydrodynamically studied from a structural
point of view. Here, the main focus is on slamming
and local structural response of ships.
Several good reviews have been published during
past years. Mizoguchi (1996) made a brief review
of the slamming phenomena, predicting methods,
and also elastic response of ship structures. Faltinsen
(2000) gave an overview on hydroelastic slamming.
Faltinsen (2004) described practical slamming problems for ships and offshore structures, including water
entry on an initially calm free surface, wet deck slamming, green water and sloshing. Xu et al. (2009)
presented a review of analysis theory and numerical simulation techniques on hydrodynamic impact of
ships.
Previous reviews dealt mainly with analytical and
numerical prediction methods. A review specially concentrating on experimental research has not yet been
published. Experiments obviously play an important
role during research procedure. Experimental methods, including model and full-scale tests, is still a
reliable approach to investigate the characteristic of
slamming loads, and obtaining design parameters,
although the cost of tests is relatively high.
Following ISSC (2009), the model tests are generally divided into two groups: one is water entry of
bodies onto a calm water surface; the other is the
model tests in waves in seakeeping basins or towing tanks. The latter tests are often carried out with
189
the seakeeping or global wave load tests in regular or

irregular waves. Full-scale tests are designed mainly
for the slamming pressures and stresses of local or
global structural response during ship trials, medium
or long-term monitoring.
Experimental methods can help to understand the
slamming phenomena, and these results are very useful, and are always adopted to verify and validate
results of analytical and numerical ones. Considering
the cost and complexity, model tests are more feasible for verification and validation, comparing with
full-scale ones.
This paper will describe the status of model test
techniques for slamming on ships, containing three
main sections: water entry of rigid body on a calm
water surface, model tests in waves, and hydroelastic
effects of local slamming.
2 WATER ENTRY INTO A CALM WATER

SURFACE
Many model tests were carried out to study the water
entry of different bodies on initially calm water. The
models can be two dimensional (2D) or three dimensional (3D), rigid or elastic, symmetric or asymmetric,
wedges or flat bodies, free-drop or velocity-controlled
drop, and so on. In this section, drop-slamming model
tests of mainly rigid bodies will be presented.
Pressure and acceleration signals will often be measured. Two methods are being used to measure impact
loads: pressure gauges (transducers) and force plates
(or pressure panels). Pressure gauges measure local
loads, and their statistical variability sometimes is
large. Their frequency responses as well as their sampling frequencies have to be very high (in the order
of kHz). Force plates measure the mean pressure over
a bigger area (typically a structural panel), and therefore they are statistically more stable. Obviously, force
plate cannot be used to measure pressure at a vey small
area. Their frequency responses have to be sufficiently
high, otherwise the effects of the resonant responses
need be corrected by means of measured accelerations.
For model tests of water entry, pressure gauges
are often adopted. Acceleration of the drop body is
always measured by acceleration transducer mounted
on the body. Correspondently velocity and displacement can be obtained by integration of the acceleration
in time, and the slamming force on the body can also be
obtained from Newtons Law. Sometimes, free surface
elevation or flow field around bodies, will be recorded
by imaging systems or Particle Image Velocimetry
(PIV), such as Ren et al. (2007).
In order to reduce errors and scatter of measuring data, guiding system sometimes will be adopted
for free drop tests, and velocity-controlled drop tests
may be carried out with guiding system. For example, Tveitnes, et al. (2008) developed a new test rig
to control the wedge to enter and also exit the water
at nearly constant velocity. In order to reduce the 3D
effect for 2D water entry problems, dummy sections
or end plates are being used, and the measuring section

is in the middle of test body.
In the early stage of model tests, the main purpose
was to get the relationship between slamming pressure
(P) and relative velocity (V) as described by Ochi and
Motter (1973),
Chuang (1967) carried out series of experiments for

wedges with various deadrise angles to study the
pressure coefficients (k) in the above equation. This
approach to predict slamming pressure on ships is
named as k-factor methods. Chuang and Milne
(1971) carried out model tests for conical shape
bodies. Up to now, the results are still always referred.
Later, the spatial and temporal distribution of pressures became of concern. Zhao and Faltinsen (1996)
presented two classic symmetric water entries model
tests, a V-shaped section with 30 degrees deadrise
angle and a typical ship bow section. Results of
velocity, pressure distributions and slamming force
in time domain were shown. The flow separation
was described. Guiding rails were used, and dummy
sections were modelled in the test body. Experimental results are compared with numerical ones using
nonlinear BEM method.
Aaesnes (1996) also carried out model tests of a bow
section with and without roll angles, which Brizzolara.
et al. (2007) compared with numerical results using
commercial code LS-DYNA, FLOW-3D and FLUENT, BEM method and SPH method. Later, Sun and
Faltinsen (2009) compared them with results by a BEM
method with different roll angles and drop velocity.
Arai and Matsunaga (1989) carried out model tests
of water entry of a bow-flare ship section with a roll
angle.
Several other model tests on 2D symmetric water
entry were published. Armand and Cointe (1986)
addressed the vertical entry of a rigid horizontal cylinder into incompressible nonviscous fluid initially at
rest focusing on the impact force. Lin and Shieh (1997)
studied the flow visualization of water impact of a
cylinder by using a modem digital imaging system.
Yettou and Champoux (2006) carried out an experimental study on water entry of symmetric wedgeshaped bodies. Yang, et al. (2007) carried out wet drop
model tests for 2D wedge sections and a ship stern section of typical modern containership. Colicchio, et al.
(2009) performed both water-entry and water-exit tests
on a circular cylinder. Simon, et al. (2010) carried out
drop test of 25 deadrise angle wedge. A high-speed
(up to 5000 frames s1 ) camera was used to visualize
the impact and subsequent formation of jet flows and
droplets. Pressure and acceleration signals were also
measured.
Not many 3D model tests of water entry are available. Peseux et al. (2005) presented a series of free
fall drop tests of rigid and deformable cone-shaped
samples with different deadrise angles and thickness.
Distribution and evolution of pressure were analyzed.
190
A numerical analysis is performed using FEM method

and predicted results are compared with experimental data. De Backer et al. (2009) performed drop tests
on a hemisphere and two conical shapes with different
angles. A guiding system was adopted to prevent small
rotations of the test body. The pressure distribution
and evolution as well as the body motion parameters were measured during impact. The water up rise
was visualized by a high speed camera. Faltinsen and
Chen (2005) presented drop tests with guide rails for
an idealized shape, which consists of cylindrical midbody and a hemispherical end, in a model composed
of four parts. Vertical force, wetted surface elevation
at key control sections, and pressure time history were
recorded. The drop tests were conducted with two different trim angles (0 and 10 ) to study the influence of
relative angle between water surface and body surface.
A generalized Wagner theory was applied to study this
3D slamming problem.
When the impact angle is small ( < 4 ), the air
trapped between the penetrating bodies and the water
surface may affect the pressure. Just as described by
Chuang (1976) empirical formula, the measured maximum pressure occurred at 23 , and it decreased
when angle becomes small, as well as the discrepancy between formula and measurements. Air cushion
effects during the impact of a finite flat horizontal and
rigid structure were examined by Verhagen (1967).
Hydroelasticity may be important, as will be discussed
in a following section.
3
MODEL TESTS IN WAVES
Hydrodynamic impact loads in extreme random waves

are an important factor to design ship structures, especially ship bows. As 3D slamming theory is still under
development, as for example, in Scolan and Korobkin
(2001), Faltinsen and Chen (2005), Korobkin and
Scolan (2006), it is not ready for practical application
to the complex 3D profiles of ship bows. Furthermore, because of the complex incident wave flow, there
are no proper 3D practical or engineering methods to
simulate this nonlinear problem for design up to now.
Impact loads on ship and offshore structures are
often predicted by model tests in severe waves. Wave
in deck impact is of concern for offshore structures
design. Baarholm and Faltinsen (2004) performed
experiments to study the global loads on a fixed horizontal deck due to the impact of a massive regular
wave. The platform deck was idealized into one flat
plate. Wetted length and vertical slamming force are
measured and compared with three numerical methods, one based on the generalization of Wagner (1932)
theory, and two different nonlinear BEM methods.
Baarholm (2009) carried out wave impact tests on a
box-shaped aluminum structure to represent an offshore platform with focus on the three-dimensional
effects. Here, local impact on ships is discussed.
Local slamming loads can be investigated by seakeeping model tests in regular and irregular waves.
ITTC (2008a, b, c) recommended procedures and

guidelines for loads and response, and seakeeping
tests, including those of high speed vehicles. It is
a practical way to obtain the design pressure from
seakeeping-slamming model tests.
Hermundstad and Moan (2005) carried out segmented model tests of a 120-m car carrier in seakeeping basin in regular oblique waves.Two pressure panels
instead of pressure transducers were mounted in the
upper part of the bow flare to study slamming pressure. Total slamming force on the bow section was also
measured. Experimental results validated numerical
simulation by direct methods, which used nonlinear strip theory and 2D BEM by Zhao (1996) with
roll angles. 3D effects of slamming pressures are discussed. Later, Hermundstad and Moan (2007) carried
out model tests to study slamming pressure on one
290-m cruise ship in irregular waves using similar
techniques described above. June and Wilson (2010)
designed Slam Patch to measure the hydro-impact
pressure and or the local structural response. The pressure panels are supposed to be rigid with high resonant
frequency. Both drop and seakeeping slamming tests
for one racing boat were carried out.
Ramos et al. (2000) conducted an experimental program assessing the slam induced loads on a segmented
ship model, which have been used recently to validate
CFD calculations by Paik, et al. (2009).
Bow slamming on a FPSO (floating production
storage and offloading ship) is different from previous bow flare or bottom slamming. The magnitude of
the wave impact at the front of the bow is dominated
by wave characteristics (namely the local wave steepness), rather than by the relative motions between ship
sections and wave surface. Sometimes it is called slapping. Maximum pressures are measured close to steep
water faces in the breaking or near breaking waves.
Voogt et al. (2004) introduced model tests for slamming pressures on FPSO in steep fronted waves. A
prediction method is developed, which can be applied
in the early design of the bow structure. An experimental program performed on a specific bow shape is
referred by Guedes Soares et al. (2007). The bow was
instrumented with pressure transducers and the test
program, also making use of video recordings, was
designed such that it was possible to determine the
correlation between undisturbed wave shape and the
impact pressure time traces. It has been found that
the wave impact at the bow is highly correlated with
the local wave steepness, which for very high waves
has at least second-order effects.
Xu et al. (2008) carried out segmented model tests
for bow slamming loads on FPSOs in towing tank.
Steep wave or breaking wave impact pressures on the
curved FPSOs bow are studied. The new-wave theory is adopted and extended to generate steep waves.
Pressure transducers were mounted in one Schiehallion model, while pressure panels in the Loch Rannoch
other model. Experimental results by these two methods are compared. Overall bow slamming forces in
two or three parts of the bow were measured.
191
HYDROELASTIC EFFECTS OF LOCAL

SLAMMING
The mechanism of hydroelastic slamming on local

structures was being studied during past years by
drop tests of flexible structure sections, or wave
impact on bodies. The main concern is of structural
responses instead of slamming pressures. Hydroelastic effects on simple structures, such as flat plate,
cylindrical shell, wedge, or cone, stiffened plating
of several stiffeners with different impact angles, are
studied.
In order to study slamming on wet-decks, theoretical and experimental studies of wave impact
on horizontal elastic plates of steel and aluminum
are presented by Kvalsvold & Faltinsen (1995) and
Faltinsen et al. (1997). One-beam model and threebeam model tests were carried out. Strain responses
in the longitudinal stiffeners and plates were measured. Significant hydroelastic effects were demonstrated. Faltinsen (2000) summarized from theoretical
and experimental research. The slamming problem
must be hydrodynamically studied from a structural
point of view. When the dead-rise angle is small,
hydroelasticity has to be considered. The higher the
value of impact velocity and the highest natural
period of local structure, the more important is the
hydroelasticity.
Some typical 2D experiments were carried out. Arai
and Miyanchi (1998) presented an experimental study
of water entry of cylindrical shell in aluminum. The
thickness was 0.003 m. Strains on the bottom of the
model were measured. Results show that hydroelasticity was important. This experiment was often cited
to validate numerical simulations.
Okada and Sumi (2000) carried out experiments
by dropping a plate and increasing the impact angle,
, from 0 to 4 with increments of 0.5 . Pressure
and strain were measured. Three patterns: the Wagner
type, the trapped-air type and the intermediate type,
were observed.
The results showed the maximum strains measured
in the plate were not so sensitive to the impact angle.
Alexander et al. (2004) described free-drop of a box
with flat stiffened panels at an impact angle of 0 .
Sun et al. (2003) carried out experiments on the fluidstructure interaction for water entry of 2D elastic
wedge made of ABS plastic.
Not much effort was devoted to the hydroelastic effect of 3D elastic bodies except asymmetric
case. Peseux et al. (2005) carried out water impact
of deformable cone shape models, but unfortunately,
only pressure results were presented, while results for
stress or deflection are not available.
Experimental study of hydroelastic impact on complex stiffened structures is not yet seen except by Wang
(2010). One complex 3D steel wedge with deadrise
angle 22 was designed. The stiffened panels on both
sides of the wedge are made up of 9 longitudinal
stiffeners and 5 transverse frames. In order to study
the effect of flexibility on the elastic responses, the
cross sections of the stiffeners and frames on each

side were designed with different sizes. Later, Luo
et al. (2011) compared stress responses on the stiffeners and frames with numerical ones using coupled
way in LS-DYNA code. Luo et al. (2012) simulated
the hydrodynamic impact and elastic responses using
an uncoupled approach with the matched asymptotic
method and FEM code. The acceleration, slamming
pressures on the plate outside, and stress responses
were compared with measured results, showing good
agreements. The high frequency oscillations on the
stress induced by hydroelastic impact effects are not
predicted due to this uncoupled way. Hydroelastic
effects were discussed.
Very few model tests of local hydroelastic effects
for composite materials were seen to be studied, which
is becoming important for high speed vehicles now.
Jensen, et al. (2000) performed drop tests of a FRP
sandwich flat panel instrumented with a network of 16
fiber optic (FO) Bragg strain sensors. The drop tests
simulated slamming loads on the wet deck of a surface
effect ship (SES).
CONCLUSIONS
The state of art of model tests techniques for local

slamming on ships is reviewed. Two main test
approaches, water entry into calm water and model
test in waves, are introduced. Model tests are the most
reliable way to predict local slamming loads on ships.
There are still many challenges due to the complex
physical phenomena.
Very few 3D hydrodynamic slamming problems
except some typical shapes, such as a hemisphere,
and a cone, were investigated. More efforts need to
be dedicated. Benchmark model tests of 3D bodies
are recommended.
Hydroelasticity due to local slamming was studied only for very simple structures, such as plate, or
panels with very few stiffeners. In this case, the lowest eigenmode had a dominant effect on the elastic
dynamic response. But model tests of complex structures section with many stiffeners and webs need to
be studied. It is not clear for these complex cases how
the structural elastic response will be influenced by so
many eigenmodes. Scaling is also one problem when
the results of model tests are transferred to those of
full scale ships.
For model tests in waves, the size of the pressure
panels needs to be carefully chosen according to the
local structures where design pressure is investigated.
The dynamic properties of panels should be studied
to eliminate resonance due to slamming. Benchmark
model tests are required to investigate the 3D effects
of structural profile and incoming flow waves. Comparisons between model tests and full scale ship
tests, especially long-term monitoring, are welcome
in order to obtain practical design pressures for local
structures.
192
ACKNOWLEDGEMENTS
The work has been performed in the scope of the
project EXTREME SEAS Design for Ship Safety
in Extreme Seas, (www.mar.ist.utl.pt/extremeseas),
which has been partially financed by the EU under
contract SCP8-GA-2009-234175.
REFERENCES
Aarsnes, J.V. 1996. Drop test with ship sections effect of
roll angle, Marintek Report 603834.00.01.
Alexander, W. J., Alex, W. V., Janssen, G. T. 2004. Response
of stiffened plate panels to hydrodynamic impact theory and experiments, Proc. 9th Int. Symp. on Practical Design of Ships and other Floating Structures;
Luebeck-Travemuende, Schiffbautechnische Gesellschaft
e.V., Hamburg, Germany.
Arai M., Matsunaga, K., 1989. A numerical and experimental
study of bow flare slamming. J of Soc Naval Archi Japan
166:343353 (in Japanese)
Arai M., Miyauchi, T. 1998. Numerical study of the impact
of water on cylindrical shells, considering fluid-structure
interactions. Practical Design of Ships and Mobile Units,
Oosterveld, M. W. C. and Tan, S. G., (Eds) Elsevier
Amsterdam, 5968.
Armand, J.L., Cointe, R. 1986. Hydrodynamic impact analysis of a cylinder. Proc. 5 th Int. Conf. on Offshore
Mechanics and Arctic Engineering (OMAE 1986); Tokyo,
Japan, 1, 237243.
Baarholm R., Faltinsen, O. M. 2004. Wave impact underneath horizontal decks. Journal of Marine Science and
Technology. 9: 113
Baarholm, R. 2009. Experimental and theoretical study of
three dimensional effect on vertical wave in deck force.
Proc 28th Int. Conf. on Offshore Mechanics and Arctic
Engineering (OMAE 2009), Honululu, Hawaii, USA.
Brizzolara S., Couty, N., Hermundstad, O., Ioan, A.,
Kukkanen, T., Viviani, M., Temarel, P. 2008. Comparison
of experimental and numerical loads on an impacting bow
section. Ships and Offshore Structures. 3, 4. 305324.
Chuang, S., Milne, D. 1971. Drop tests of cones to investigate the three-dimensional effects of slamming. Technical
report 3543, NSDRDC.
Chuang, S.L. 1967. Experiments on slamming of wedgeshaped bodies. Journal of Ship Research. 11,190198.
Colicchio, G., Greco, M., Miozzi, M., Lugni, C. 2009. Experimental and numerical investigation of the water-entry and
water-exit of a circular cylinder, Proc. 24th IWWWFB.
Zelenogorsk, Russia, April 1922.
De Backer, G., Vantorre, M., Beels, C., De Pre, J., S.
Victor, De Rouck, J., Blommaert, C., Van Paepegem,
W., 2009. Experimental investigation of water impact
on axisymmetric bodies. Applied Ocean Research. 31,
143156.
Faltinsen, O.M. 2000. Hydroelastic slamming. J. of Mar. Sci.
Tech. 5, 4965.
Faltinsen, O.M., Chen, Z. M. 2005. A generalized Wagner
method for three-dimensional slamming, Journal of Ship
Research, 49, 4, 279287
Faltinsen, O.M., Kvlsvold, J., Aarsnes, J.V., 1997. Wave
impact on a horizontal elastic plate. J Mar Sci Technol.
2, 2, 87100
Faltinsen, O.M., Landrini, M., Greco, M., 2004. Slamming in
marine application. Journal of Engineering Mathematics.
48, 187217.
Guedes Soares, C.; Pascoal, R.; Antao, E.; Voogt, A., and
Buchner, B. 2007. An Approach to Calculate the Probability of Wave Impact on a FPSO Bow. Journal of Offshore
Mechanics and Arctic Engineering. 129, 7380.
Hermundstad, O.A., Moan, T. 2007. Efficient calculation of
slamming pressures on ships in irregular seas. J. Marine
Science and Technology, 12, 60182.
Hermundstad, O.A., Moan, T. 2005. Numerical and experimental analysis of bow flare slamming on a Ro-Ro
vessel in regular oblique waves. J. Mar Sci. Technol. 10,
105122.
ISSC, 2009. Report Committee V.7 Impulsive pressure loading and response assessment. Korea. Volume 2.
ITTC recommended procedures and guidelines, 2008a. 7.502-07-02.1 Testing and extrapolation methods, load and
responses, seakeeping experiments, http://ittc.sname.org/
ITTC recommended procedures and guidelines, 2008b.
7.5-02-05-04 Testing and extrapolation methods, High
speed marine vehicles, seakeeping tests, 2008, http://ittc.
sname.org/
ITTC, recommended procedures and guidelines, 2008c. 7.502-05-06 Testing and extrapolation methods, High speed
marine vehicles, structural loads. http://ittc.sname.org/
Jensen, A. E., Havsgrd,G. B., Pranb, K., Wang, G., Vohra, S.
T., Davis, M. A., Dandridge, A., 2000. Wet deck slamming
experiments with a FRP sandwich panel using a network
of 16 fibre optic Bragg grating strain sensors. Composites:
Part B, 31,187198.
Korobkin, A.A., Scolan, Y.-M. 2006. Three-dimensional theory of water impact. Part 2. Linearized Wagner problem.
Journal of Fluid Mechanics. 549, 343373
Kvalsvold, J., Faltinsen, O.M. 1995. Hydroelastic modelling
of wetdeck slamming on multihull vessels. Journal of Ship
Research. 39, 225239.
Lee, J., Wilson, P.A. 2010. Experimental study of the hydroimpact of slamming in a modern racing sailboat. Journal
of Sailboat technology. 1, 128
Lin M.C., Shieh L.D., 1997. Flow visualization and pressure characteristics of a cylinder for water impact. Applied
Ocean Research. 19, 2, 101112
Luo, H.B., Qiu, Q., Wan, Z.Q. 2007. Experimental study
of whipping responses induced by stern slamming loads.
Proc. of the 10th Int. Symp. on Practical Design of Ships
and Other Floating Structures (PRADS 2007); Houston,
Texas, USA: ABS; 535542.
Luo, H.B., Wang. H., Guedes Soares C. 2011. Comparative
study of hydroelastic impact for one free-drop wedge with
stiffened panels by experimental and explicit finite element method. Proc. 30th Int. Conf. on Ocean, Offshore
and Arctic Engineering (OMAE 2011); Rotterdam, The
Netherlands, ASME Paper OMAE2011-49209.
Luo, H.B., Wang H., and Guedes Soares C. 2012. Numerical
and experimental study of hydrodynamic impact and elastic response of one free-drop wedge with stiffened panels.
Ocean Engineering. 2012, 40, 114
Ochi, M.K., Motter, L.E. 1973. Prediction of Slamming
Characteristics and Hull Responses for Ship Design.
Transactions SNAME, 81, 144177.
Okada, S., Sumi, Y. 2000, On the water impact and elastic
response of a flat plate at small impact angles. Journal of
Marine Science and Technology, 5, 3139.
Paik, K.-J., Carrica, P.M., Lee, D., Maki, K., 2009, Strongly
coupled fluid-structure interaction method for structural
loads on surface ships, Ocean Engineering 36, pp. 1346
1357
Peseux, B., Gornet, L. and Donguy, B. 2005. Hydrodynamic
impact: Numerical and experimental investigations.
Journal of Fluids and Structures, 21, 277303.
193
Ramos, J.; Incecik, A., and Guedes Soares, C. 2000, Experimental Study of Slam Induced Stresses in a Containership.
Marine Structures. 13(1):2551.
Ren, B., Li, X.L., Wang Y. 2007. Experimental Investigation of Instantaneous Properties of Wave Slamming on
the Plate. China Ocean Engineering, 21, 3, 533540.
Scolan, Y. M., Korobkin, A.A. 2001.Three-dimensional theory of water impact. Part1, Inverse Wagner problem.
Journal of Fluid Mechanics. 440, 293326.
Simon G Lewis, Dominic A Hudson, Stephen R Turnock,
Dominic J Taunton. 2010. Impact of a free-falling wedge
with water: synchronized visualization, pressure and
acceleration measurements. Fluid Dynamics Research,
42, 3.
Sumitoshi Mizoguchi, and Katsuji Tanizawa, 1996. Impact
wave loads due to slamming-A review. Ship Technology
Research. 43.
Sun, H., Faltinsen, O.M. 2009. Water entry of a bow flare
section with a roll angle. Journal of Marine Science and
technology. 14, 6979.
Sun, H., Lu, Z., He, Y., 2003, Experimental research on the
fluid-structure interaction in water entry of 2D elastic
wedge, Journal of Hydrodynamics, 1, 16 (in Chinese).
The Joint Accident Investigation Commission of MV ESTONIA. MV Estoniafinal report, December 1997, http://
www.edita.fi/aib_finland/estoneng/index.html, Edita Ltd.
Tveitnes, T., Fairlie-Clarke,A.C., Varyani,K., 2008. An experimental investigation into the constant velocity water
entry of wedge-shaped sections. Ocean Engineering. 35,
14631478.
Verhagen, J.H.G. 1967. The Impact of a Flat Plate on a Water
Surface. Journal of Ship Research, 11, 211223.
Von Karman, T. 1929. The impact on seaplane floats during
landing. National Advisory Committee for Aeronautics.
Technical note No. 321, 309313.
Voogt, A., Buchner, B., Joaqun Lopez-Cortijo Garcia. 2004.

Wave impacts excitation on ship type offshore structures in
steep fronted wave. OMAE-FPSO 2004, Houston, USA.
Wagner, H. 1932. Uber Stoss- und Gleitvergange an der
Oberflache von Flussigkeiten. Zeitschrift fuer angewandte
Mathematik und Mechanik 12, 193215.
Wang, H. 2010, Investigations on Ship Slamming Loads
and Structure Responses (In Chinese). PhD Thesis. China
Ship Scientific Research Center. Wuxi, China.
Wu G.X., Sun H., He Y. S. 2004. Numerical simulation and
experimental study of water entry of a wedge in free fall
motion. Journal of Fluids and Structures, 19, 3, 277289.
Xu, G.D., Duan, W.Y. 2009. Review of prediction techniques
on hydrodynamic impact of ships. Journal of Marine
Science and Applications. 8,204210
Xu, L., Barltrop, N., Okan, B., 2008. Bow impact loading on FPSOs 1Experimental investigation. Ocean
Engineering. 35, 1112, 11481157.
Yang S. H., Lee H. H., Park T. H., Lee I. H. and LeeY. W. 2007.
Experimental and Numerical Study on the Water Entry of
Symmetric Wedges and a Stern Section of Modern Containership. Proc. of the 10th Int. Symp. on Practical Design
of Ships and Other Floating Structures (PRADS 2007);
Houston, Texas, USA: ABS; 518526.
Yettou E-M, A. AD, Champoux Y. 2006. Experimental study
on the water impact of a symmetrical wedge. Fluid
Dynamics Research. 38, 1, 4766.
Zhao, R., Faltinsen, O.M. and Aarsnes, J.V. 1996. Water
entry of arbitrary two-dimensional sections with and
without flow separation. Proc. 21st Symposium on Naval
Hydrodynamics, 408423.
194

A static analytical method for the preliminary design of multiple line

mooring systems
ABSTRACT: The paper presents an analytical method for the analysis and preliminary design of multiple line
mooring systems. The method is based on the classical solution of the catenary of a single cable, namely the
curve that a uniform, flexible and inextensible cable assumes when hanging under the influence of gravity. Firstly
the formulation is presented for the case of inextensible lines and afterwards the method is extended to properly
account the material elasticity, which significantly affects the tensions along the line when the line is taut. The
resulting equations are solved by a computer code which implements algorithms able to derive the static response
in terms of end force and tension distribution in an elastic catenary by knowing the equilibrium geometry of
the system, the elasticity and the unstretched length of each line. The solution is generalized for axis-symmetric
multiple lines which resemble the typical mooring arrangement of floating offshore structures. The advantage
of this approach results from the capability to provide the existence and uniqueness of solution, which, together
with the very small computational effort, makes this method a perfect tool for a preliminary design. Finally, the
applicability of the method is demonstrated with the preliminary design of the mooring system for a generic
floating wave energy converter operating at the West coast of Portugal.
INTRODUCTION
Mooring systems are widely employed by floating

offshore structures such as ship-shape vessels and platforms for station keeping purposes. Over the last years
the development of the offshore Wave Energy Converters brought a new field of employment of these
systems. Unlike the other mentioned floating structures, WECs are classified as Near Shore Devices
and are commonly installed in water depths up to
around 80 m, but in general between 40 and 60 m.
Their location, combined with their functional requirements, makes the mooring problem of these devices to
be different and sometimes more complex from that
of the other offshore structures. As a consequence
the accumulated experience in the design of mooring systems for offshore platforms most of the times
cannot be applied directly and new codes and standards have to be developed. The aim of this paper is
to propose a semi-analytical static formulation able to
solve a three-dimensional partially grounded multiline mooring system, which results could be used
as framework for a more robust numerical formulation adapted specifically to the WECs systems design
requirements. In fact, in spite additional loads, due to
vortexes and current drifts, and the mooringsdynamic
response were to be considered, most of the times for
the purpose of preliminary design a quasi-static analysis is adequate and offers a significant reduction in
computational cost.
Nevertheless computational effort has become

cheaper bringing numerical models to play an important role in most of the research areas, but the use of
simplified analytical models has not ceased. The analytical approach provides the existence and uniqueness
of solution and if hybrid analytical-numerical schemes
are used to describe the non-linearities of the cable
configuration, fast codes can be produced. Moreover,
because of their lightness these codes can be coupled
with optimization methods, yielding to an efficient and
undoubtedly useful tool for preliminary design of the
multi-leg mooring systems.
1.1 State of the art
Huang and Vassalos (1993) adopted a semi-analytical
approach to predict the static performance and snap
loading of marine fully suspended cables. They modelled the mooring lines with lumped masses connected
by compressionless springs. Their method permits
efficient and quick parametric studies for optimal
selection of the system particulars, but it is ineffective for the analysis of partially grounded mooring
systems. An alternative semi-analytical formulation
based on the mathematical development on mechanics of catenary (O Brien, 1967) is later presented by
Chai et al. (2001). As well as the above approach Chai
adopts a quasi-static formulation but in contrast to the
Huangs model, his model permits slope continuity to
be maintained throughout the cable length.
195
Nakamura et al. (1991) used catenary formulae to

calculate the integrated quasi-static velocity, acceleration and dynamic tension along the mooring. Their
non linear time domain formulation shows how drag
forces acting on the mooring lines acts as the damping force influencing significantly the response of the
floating structures to slow drift motion.
Gobat and Grosenbaugh (2005) included in their
formulation the wave frequency effect neglected
in Nakamuras formulation and presented a simple
empirical model for determining the dynamic tension due to vertical motions at the top of a catenary
mooring.
Smith (2001) proposed an analysis of the taut-slack
dynamics of a straight, strictly vertical cable connected
to the seabed. The model, by the use of mathematical
analytical tools manages to give an interesting qualitative insight in the properties and non-linearities of the
system.
Although on one hand the mentioned semianalytical and static model has many pros in the
preliminary design step, on the other it is not able
to give a consistent understanding of the complex
behaviour of the system. In fact it is not adequate to
include the effect of inclined flat seabed or intermediate buoys neither to consider hybrid lines nor varying
cross-sectional and material properties. For these purposes and in order to have a more complete description
of the system usually numerical Finite Differences
(FD) or Finite Element (FE) are developed.
Peyrot built a computer algorithm around an efficient cable element subprogram (Peyrot and Goulios,
1979; Peyrot, 1980) able to derive displacements
or forces at any number of nodal points and the
cable tangent stiffness matrix by knowing just its end
co-ordinates, line elasticity and unstretched length.
Moreover it gives the possibility to simulate interaction of the cable with currents or wave.
The earliest investigators to solve the system of
nonlinear partial differential equations governing the
transient motion of a cable immersed in a fluid with a
finite difference technique were Walton and Polachek
(1959). They developed the so called lumped mass or
lumped-parameter model (Leonard and Nath, 1981) in
which the cable is discretized into a series of point mass
elements connected by elastic springs. By solving a
set of simultaneous equilibrium equations, which consider the external forces lumped to the point masses,
the program yields easily to the nodal displacement.
Regarding the application of the FD techniques including lumped mass models and box schemes, reference
must be done to Huang (1994), Tjavaras et al. (1998),
and Chatjigeorgiu and Mavrakos (1997).
Beside the FD another numerical method, FME,
was developed to solve the system of equations.
This technique uses a discrete number of elements
to approximate a continuum. The application of this
method to analyze the marine cables was first proposed by Leonard (1972). Later the FEM appears also
in the research by Webster (1975, 1979), who implemented a standard-finite element technique and in that
of Chen et al. (2001) who extended the CABLE3D

model developed by Ma and Webster (1994) to allow
large elongation of mooring lines.
More recent studies focus on the interaction
between the mooring line and the seabed. Chai and
Varyani (2002) are the first to formulate a lumpedmass complete three-dimensional formulation solving
the general pipeline problem with combined bending
stiffness and torsional loading effects and irregular
seabed interaction. Their study permits both static and
a dynamic analysis of most of the off-shore slim structures. Yu and Tan (2006) presented a new FEM which
described the mooring cable dynamics and interaction
with the sea bed in time domain and which included
different soil constitutive models and elastic models.
In conclusion numerical methods including
dynamic response of the moorings and interaction
with the seabed are more efficient and nowadays do
not require neither an extensive use of memory nor
an extremely powerful computer but in some cases
an static-analytical model is sufficiently accurate to
yield to the desired predesign results.
2 THEORETHICAL APROACH
2.1
Static inextensible cable in still water

(non-elastic catenary)
This paragraph overviews the classical solution of the

catenary of a single cable due to Huygens, Leibniz,
Jakob and Johann Bernoulli (1691). A catenary is the
curve that a uniform, flexible and inextensible cable
assumes when hanging under the influence of gravity.
To deduce the equation of this plane curve we consider the equilibrium of the portion AP of the chain in
figure 1. Let assume that:
T: tension in the line
H: horizontal tension in A
V: vertical force in P
w: line submerged weight per unit length
: angle to horizontal made by the line in P
s: length of the portion AP
In a mooring system the suspending part of the cable
is the so-called catenary while the lying down part of
the line on the sea floor is called grounded part. The
connecting point between these two parts of a mooring
line is called touchdown point and the top point is
called fairlead point (fig. 1).
From the equilibrium of the infinitesimal portion of
cable we can see that:
Using the Pythagoras theorem we can derive:
196
Figure 2. Mooring line element AP with or without axial

stretch.
From these equations we can obtain some simplified

formulas, which can be useful in the design of a mooring line, for instance when tension at the upper end
and the water depth is given and we want to calculate
the length necessary to prevent the uplift at the anchor
or when the tension and the line length is given and we
need to know their relative horizontal scope X.
Figure 1. Geometry of mooring line as top moves.
For practical purposes concerning mooring system

design, there is often the need to have relationships
between tension and the Cartesian coordinates x and y.
If a constant a, having the dimension of a length,
is introduced and coupled with eq. 3 we come to an
expression for the derivative ds/dx in terms of dy/dx:
a=
Horizontal projection of the line
Minimum line length required (assuming that at the

touchdown point = 0)
H dy
s
;
=
w dx
a
Horizontal force for a given tension at the upper
end of the line
Vertical force at the upper end
Considering once again the Pythagoras relation we

deduce a second expression for ds/dx and so a second order differential equation describing the catenary
shape:
Integrating eq. 6 and imposing the boundary conditions y (0) = 0 and y(0) = 0, implying that the line
length is measured from the point of tangential touchdown at a horizontal seabed we derive the expression
of the catenary geometry for an inextensible cable:
2.2 Static extensible cable in still water

(elastic catenary)
The approach presented above considers the hanging
line as inextensible, nevertheless the mooring lines
are usually made of materials with elastic proprieties, which affect significantly the tension level on
the lines. It becomes indeed important to extend the
above approach to properly account the material elasticity. To this end a linear-elastic constitutive behaviour
is considered (ds = 1 + T/AE), where E is the Youngs
Module and AE the axial stiffness.
From figure 2 we can see how the static equilibrium
equations for the lines portion will not be modified by
assuming the cable elastic but the relationships which
link the horizontal (x) and vertical (y) coordinates to
the line length (s) and angle () will be affected as
follows:
This equation can also be written in terms of x and s

remembering that dy/dx = s/a:
Substituting eqs.10 in eq. 2 and 3 and integrating it is
possible to obtain the catenary equation for an elastic
197
cable obeying Hookes law, which remains tangential

to the sea bottom under all force conditions:
New simplified formulas can be derived and compared

with those which neglect the elastic proprieties of the
material:
Figure 3. Line load excursion characteristics comparison.
Horizontal projection of the line

2.3 Touchdown point and excursion
Minimum line length required (assuming that at the
touchdown point = 0)
Horizontal force for a given tension at the upper

end of the line
Vertical force at the upper end

It is remarkable how the horizontal projection of
the line is increased by a positive factor (HL/AE),
which points out the importance of considering the
elasticity when analyzing mooring systems. Moreover
for a given horizontal force acting on the top of the
line, its maximum tension results to be overestimated
if elasticity is neglected. This is due to the fact that,
when considering the lines material as elastic the minimum length for a given horizontal force is seen to be
shorter than when considering the cable as non-elastic
(eq. 9.b compared with eq. 12.b). This affects the vertical force on the upper end, as it depends on the cable
length (eq. 9.d and 12.d), and by consequence the total
maximum tension on the line.
Finally another aspect, which emphasizes the
importance of considering the elasticity, is the forceexcursion characteristic. The horizontal stiffness of
each line as it translates on its own plane is defined
as the ratio between the increment of the horizontal
force (taking a starting force as reference) and the
respective excursion from the starting position. As it
can be observed in figure 3 when the line is considered
as elastic its non-linear stiffness is significantly lower
than when the elasticity is neglected. This huge difference between the elastic and inelastic lines behaviour
can play an important role in practical situations, when
for instance it can be required by the moored device to
have a limited excursion.
The total scope of the line will usually consist of a

grounded length (p, in fig. 1) and a suspended one
called catenary part. Either if the cable is elastic or
not when a horizontal force is applied to the top of the
cable an amount of the line, which does not depend of
the total length of the grounded cable, picks up from
the seabed.
If L0 is the starting suspended length, L1 is the
length after the application of an extra tension, X0
is the horizontal scope related to L0 and X1 the one
related to L1 , then the grounded part of each cable
configuration is described by:
Considering these formulas and observing figure 1 we

obtain the expression of the top excursion, which wont
depend on the total length of the line:
where X values are calculated from eq. 9.a or 12.a and

L values from 9.b or 12.b.
PROGRAM DEVELOPMENT
Daghigh et al. (2002) categorize mooring lines into

two types; the catenary settling on the sea bottom and
the limited one which can have a nonzero angle at its
attachment point to seabed. The difference between
these two types is that the former can maintain its zero
angle at the bottom just by decreasing its dead-length
on sea floor, while the latters bottom angle is subject
to change when the force on its other end alters. The
hereinafter computer code implements algorithms able
to derive the static response in terms of end force and
tension distribution in an elastic catenary by knowing
the equilibrium geometry of the system, the elasticity
and the unstretched length of each line.
198
2 Computation of the maximum horizontal scope,

Xmax, (which depends on the total unstretched
length of the line, eqs.8.a and 8.b), the maximum
excursion each line can undergo maintaining the
zero angle at the anchor point and finally the
maximum horizontal excursion the system can be
subjected to.
3 Calculation of the geometrical and tension condition of each line for every one-meter deflection in
the x-direction imposed on the system.
As far as the equilibrium geometry is concerned

in addition to the number of the lines composing the
system, two different inputs are accepted:
The coordinates of the anchor point of the first line
(turning anticlockwise from the horizontal axis)
which lies in the first quadrant of the Cartesian
coordinate system and the unstretched line length
The pretension on the lines, the rotation of the system from the horizontal axis, and the unstretched
line length.
From this information the program extracts either
the pretension of the system or the anchor point
Cartesian coordinates.
The assumptions of the program based on the
above-described theoretical approach are:
Small deformations only are considered (but large
displacements).
Linear-elastic constitutive behaviour only is considered ( = EA).
Only self-weight acts on the cable, and it is assumed
conservation of mass of the cable element during
the deformation process
Fixed anchor point is considered
Bending and torsion stiffness are neglected.
Plane funicular geometries, only are considered for
simplicity
Bottom friction on the dead-length on sea bottom
is neglected
Flat and horizontal seabed only is considered
All the lines have the same material (EA) and
geometrical (d and L) characteristics
The system is symmetrical
In the model the value of the submerged weight
per unit length and that of the axial stiffness per unit
length (AE) for a stud chain are calculated with the
hereinafter empirical formulas:
where d is given in mm.

3.1
Structure of the program
The main purpose of this program is to calculate the

increment of forces and tensions in each line of the
system when a horizontal displacement (x-direction)
is imposed on the fairlead point. Secondly it aims to
obtain some characteristic charts of the mooring system, useful to better understand its total behaviour in
terms of stiffness, deformations and tensions.
To this end the implementation was divided in three
main parts:
1 Calculation of the equilibrium condition, in other
words the pretension (T0 ), the respective horizontal (H0 ) and vertical forces (V0 ), the suspended length to touch down point (L0 ) and the
horizontal scope (X0 )
The last part of the program elaborates the outputs

to plot the desired charts.
3.2
Implementation choices
One of the difficulties in the implementation of the

elastic theoretical formulas is their non-linearity. Thus,
a positive excursion applied in the line path brings to
an increment in the line tension higher than the decrement we would observe for an excursion of the same
amplitude but in the negative direction. Moreover for
a catenary at to the direction of displacement, the
applied deflection and the restoring force are both
reduced by a cos (). The effect of all catenaries with
different will be summed. If the system was linear
than each catenary would contribute to the system stiffness in proportion to cos (), but the non-linearity in
the system requires the calculation to be carried out
numerically.
Secondary some of the equations, for instance
eq. 12.a, cannot be solved for H and this increase the
computational effort to be made when, for a defined
excursion, the respective variation of the tension is
required.
By considering all these observations, the program
was implemented to impose an x-direction positive
deflection to the upper point of the spread mooring
system where the lines are assumed to converge.
3.2.1 Calculation of the maximum acceptable
x-direction deflection
As already specified in this section the inputs are used
in the first part of the program to obtain the equilibrium
configuration when no external forces excite the system. It is from these geometrical and tensional results
and from the assumption that for the extreme configuration the line will be completely suspended but will
maintain the zero with the sea floor, that it is possible
to compute the maximum applicable excursion on the
line plane from the equilibrium position (eqs. 12.b and
12.c).
Once this value is known, depending on the number of lines and on the system orientation respect to the
excursion direction, the maximum x-direction excursion can be calculated from simple trigonometric and
geometric observations. The program checks which is
the critical line, in other words, the line that will reach
the maximum extension first. Because of the implementation methodology this line will always lie in the
second quadrant of the Cartesian coordinate system
199
If the studied system has n lines, then the program will generate, for every information (tension,
horizontal force, grounded line length, angle between
the line and the displacement direction,) a matrix
with n rows, each one representing one line, and
a number of columns depending on the maximum
admissible excursion. Each column will have the information related to a one-metre excursion, apart from
the last one, which represents the values referred to
the exact maximum deflection. A clearer description
of the outputs will be given in the following section
when presenting some case study results.
4
CASE STUDY
4.1 Design methodology
Figure 4. Plane view of a four-line system. Equilibrium

and extreme geometry.
and will be the one with the maximum angle . This

procedure is easily understandable from figure 4.
3.2.2 Static calculation of tension and geometry

in between the equilibrium and extreme
configurations
After the definition of the equilibrium and extreme
geometry the program computes, for each line, all the
tensional and geometrical characteristics with reference to the eqs.17.This calculation neglects the friction
between the chain or rope and the seabed during the
displacement from B to C. The change in the position
of fairlead point (from B to C) is analysed as static,
which means that the total excursion max_h is discretized in portions of one meter (BB , B B , B C)
and for each of these position the program calculates
the tensions, the touchdown point position, the inclination of the cable with respect to the horizontal axis
and the suspended length.
3.2.3 Outputs
The outputs of the program are, as already said in
the previous sections, both the tension and the configuration of the system in each 1-metre excursion
step.
This section presents the application of the described

model to design the mooring system of floating wave
energy converter. First of all one needs to know
the expected maximum loads that the system will
encounter during its operational life and in accidental conditions, namely the design storm (Hs and Tz ),
the current and the wind speed. As far as this information is concerned reference is made to Fonseca et al.
(2009). In their paper two design storms are selected
based on the climatology of the Portuguese Pilot Zone,
located at the western coast of Portugal, and statistics
of wind and current values are given. Moreover they
suggest a design methodology, which will be taken as
guideline in this section of the paper as well. The mentioned Ultimate Limit State (ULS) design procedure,
can be summarized in the following three steps:
1 Define the design storm and significant wind and
current speed, on the basis of the installation area
climatology
2 Calculate the maximum load expected on the moorings, considering a combination of the extreme
effects of waves, wind and current
3 Verify if the maximum load on the cable, weighted
by a safety factor, is smaller than the Minimum
Break Load (MBL) of the material in which the
cables are made
As can be read in Fonseca et al.s paper, to combine the loads exiting the mooring system, in order
to calculate the tension on the most tensioned line,
decomposition between static and dynamic loads is
to be done. Static loads include steady drift force for
the design storm, wind force corresponding to the
design sustained wind speed (mean wind speed), current force corresponding to the design current speed.
These forces are summed up together, resulting in the
static horizontal load exiting the system and in its relative static horizontal deflection. It is then around this
position that the dynamics component is calculated.
The load combination assumed by this design procedure results in a horizontal deflection of the mooring
system defined by the formula above:
200
Table 1.
Design conditions and loads (Fonseca et al., 2009).

Condition
Current force
Wind force
Mean drift force
Slow drift motion (max.)
Wave frequency motion
(sig. value)
Velocity = 60 cm/s
Velocity = 25 m/s
Design storm
Hs = 9.5 m,
Tz = 11.0 s
Incidence = 0
Table 2.
systems.
Load
68 [KN]
60 [KN]
353 [KN]
23.8 [m]
3.3 [m]
Geometrical characteristics of the 5 analysed
Case
N
cables
L
[m]
D
[mm]
w
[N/m]
wsubmerged
[N/m]
Case 1.a
Case 1.b
Case 2.a
Case 2.b
Case 3
4
4
6
6
4
283
294
272
283
400
147
142
147
142
105
4640.13
4336.02
4640.13
4336.02
2364.21
4036.91
3772.33
4036.91
3772.33
2056.86
where:
Xstatic : is the static deflection, determined by the
design wind and current force and by the steady
drift force,
Xpeak,sdrift : is the peak of the slow drift motion
Xwf : is the significant value of the horizontal
motion (X1/3 ) in the wave frequency.
Summarizing, to determine the maximum load on
the system, namely the design load, one needs to:
1 Define the static horizontal force, define the peak
of the slow drift horizontal force and sum them
together
2 Run the program and, from the static characteristic
of the system (load vs. displacement), calculate the
horizontal deflection relative to the first step load
3 Sum to this deflection the significant value of
the horizontal motion in the wave frequency and
extrapolate from the model output the top tension
of the maximum tensioned cable of the system.
This will be the design load.
Figure 5. Horizontal force on the global system and tension

on the most tensioned line VS deflection.
Table 3. Design loads, safety factor and minimum break
load (MBL).
Finally this load weighted for a safety factor has to

be compared with the MBL of the material composing
the lines.
Case 1.a
Case 1.b
Case 2.a
Case 2.b
Case 3
4.2 Design examples

The aim of this section is to show the potential of the
model to find the most convenient design solution,
considering a static behavior of the system.
The exiting forces and displacements, as already
introduced in the previous section, are taken from
Fonseca et al. (2009) and are summarized in table1.
Summing the steady forces (current and wind forces)
with the peak of the slow drift force yields to an initial static load of 2450 KN. Based on this value five
different design solutions are proposed and compared
assuming as inputs:
Case 1.a: 4 cables, d = 147 mm;

Case 1.b: 4 cables, d = 142 mm;
Case 2.a: 6 cables, d = 147 mm;
Case 2.b: 6 cables, d = 142 mm;
Case 3: 4 cables, d = 105 mm.
Table 2 presents the five different geometries and

cable characteristics while figure 5 represents the
graphical design procedure for a general solution. The
hypothesized installation location has a water depth
Max top
tension
[MN]
Safety
factor
Design
Load
[MN]
MBL
(Grade R3)
[MN]
3.338
3.358
3.092
3.120
3.352
2.7
2.7
2.7
2.7
2.7
9.04
9.14
8.37
8.49
9.05
15.54
14.68
15.54
14.68
10.75
about 50 m (h), and a pretension (T0 ) of 1.5wh is

applied to each of the four solutions.
Once the number of cables, the diameter of the chain
and the loads are defined the model is run for each of
these four configurations, assuming that all the load
act in the same direction with an angle of 45 and 30
for the 4 and 6 cable systems, respectively (e.g. fig. 4).
The results are the total length of each cable necessary to assure a zero angle between the cable seabed
and the top tension of the most tensioned line. These
values have to be weighted for a safety factor and must
result lower than the MBL value of the chain. The
ABS Rules for building and classing mobile offshore
drilling units (1997) specifies a minimum factor of
safety of 2.7 for quasi static mooring analysis.
201
Table 4. Comparison between the four designs in terms of

steel weight and total deflection.
Case 1.a
Case 1.b
Case 2.a
Case 2.b
Case 3
Total steel weight

[T]
T0
[kN]
Total deflection
[m]
535.44
519.79
771.94
750.52
385.60
302.77
282.93
302.77
282.93
154.26
28.88
29.29
23.81
24.15
32.62
The first four solutions are designed (case 1.a, 1.b,

2.a, 2.b) in order to maintain the ratio between the
water depth and the length of the cable around 1/5-1/6,
while the last configuration (case 3) suggest a solution
with a h/L ratio of 1/8. The aim is to get to know
which is the most convenient solution economically
speaking, in other words to understand if it is better to
have a shorter chain with a bigger diameter or a longer
chain with a smaller diameter.
To this purpose table 4 presents a comparison
between these five solutions both in terms of total
weight of steel to be used and in terms of total
deflection.
Assuming that the total weight of steel necessary
for each solution is a parameter representing the cost,
a first analysis was carried comparing the cases 1a,b
and 2a,b. What comes out is that, the configuration
that minimizes the steel weight is case 1.b, which with
4 cables of a smaller diameter and a longer cable
presents a deflection comparable with those of the
other configurations, and acceptable for the studied
device.
This result enhances the interest in considering a
fifth configuration (case 3), which, keeping the number of cables to 4, as previously found, proposes an h/L
ratio of 1/8, and a minimized diameter. The reduction
in the total steel needed is highlighted in table 4.
This last presented solution is to be considered the
best economically speaking since it requires the smallest amount of steel and in addition, because of its
lighter weight per unit length, it minimizes the installation cost. In fact, a lighter chain brings to a smaller
pretension load and by consequence to the need of
less powerful ships for the installation. Furthermore,
while the other options appear to be over dimensioned,
this configuration is not. On the other hand a limit
of this configuration is related to its inclusion in a
wave farm, since for a higher production of energy per
square meter it could be asked to each device to minimize the occupied surface. In conclusion, for a final
project many other aspects have to be considered but
in a preliminary design the 4 cable and small diameter
solution seems to be the best.
Hereinafter some outputs charts of the model are
proposed. In figure 6 the total horizontal and vertical force on the system together with the top tension
on the most stressed line are plotted. Figure 7 represents the amount of cable that lies on the sea bottom
or is lifted depending on the deflection imposed to the
Figure 6. H: total horizontal force applied to the system,

V: total vertical force, T: tension on the most stressed lines.
Figure 7. Grounded and suspended length depending on the

deflection applied to the system. p: grounded length, L: lifted
length. Blue: the taut cable, orange: the slack cable.
system. The orange lines represent the cable that with

an increment of the horizontal force on the system,
and so with an increment of the deflection, loses tension, while the blue represents the tensioned line. As
expected, an increment of the lifted length corresponds
to a reduction of the grounded length of the tensioned
cable, and vice versa for the slack cable.
CONCLUSIONS AND FUTURE WORKS
The paper presents an analytical method for the analysis of axis-symmetrical multiple line mooring systems.
The lines are assumed uniform, the material elasticity is taken into account and the dynamic effects
are neglecte. The applicability of the procedure is
demonstrated with the preliminary design of a mooring
system for a wave energy converter.
Although a dynamic analysis is in principle more
accurate, in a predesign step the presented model come
to be a useful tool, both because of its simplicity and
because of its speed in giving the output. Moreover,
202
because of its small computational effort, this code

could be coupled with optimization methods, making
the research of the best solution automatic. To this
propose it would be important to define a criterion to
define the optimal configuration, criterion that could
be based on a combination of steel weight and on minimum deflection as suggested herein. Future works
will focus on the development of these optimization
methods and on the incorporation of this first step
design model in more complex codes, considering the
dynamics of the system.
ACKNOWLEDGEMENTS
The work was performed in the scope of the project
Design of mooring systems for floating wave energy
converters, which is financed by Fundao para
a Cincia e a Tecnologia (FCT) under contract
PTDC/EME-MFE/103524/2008.
REFERENCES
Chai, Y. T., K. S. Varyani, N. D. P. Barltrop, 2002. Threedimensional Lump-Mass formulation of a catenary riser
with bending, torsion and irregular seabed interaction
effect, Ocean Engineering, 29(12):15031525,
Chai, Y.T., K.S. Varyani, N.D.P. Barltrop, 2002. Semianalytical quasi-static formulation for three-dimensional
partially grounded mooring system problems; Department
of Naval Architecture and Ocean Engineering, University
of Glasgow, Scotland, UK
Chatjigeorgiou, J.K., S.A. Mavrakos, 1997. Nonlinear Contributions in Predicting Dynamic Tensions in Mooring Lines
for High and Low Frequencies of Excitation, Proc. of
the 7th Int. Offshore and Polar Engineering Conference,
ISOPE97, 2:192199.
Chen X., X., J. Zhang, P. Johnson, M. Irani, 2001. Dynamic
analysis of mooring lines with inserted springs, Applied
Ocean Research, 23(5)277284.
Daghig, M., R.T. Paein Koulaei, M.S. Seif1, 2002. Moorign
system design and optimization for floating bridge of
Urmia Lake, Proceedings of International Conference
on Offshore Mechanics, Sharif University of Technology,
Tehran, Iran.
Fonseca, N., Pascoal, R., Morais, T., Dias, R., 2009, Design
of a mooring system with synthetic ropes for the FLOW
wave energy converter, Proceedings 27th International
Conference on Offshore Mechanics and Arctic Engineering (OMAE2009), paper OMAE2009-80223, 31 May5
June, Honolulu, Hawaii.
Gobat, J.I. and M.A. Grosenbaugh, 2006. Time domain

numerical simulation of ocean cable structures. Ocean
Engineering, 33(10):13731400.
Huang, S. and D. Vassalos, 1993. A Semi-Analytic Treatment
of Three-Dimensional Statics of Marine Cable, Ocean
Engineering, 20(4):409420.
Huang, S., 1994. Dynamics Analysis of Three-Dimensional
Marine Cables. Ocean Engineering 21(6):587605.
Huygens, Leibniz, Jakob and Johann Bernoulli, 1691. Acta
Editorum.
Leonard, J.W. and J.H. Nath, 1981. Comparison of finite element and lumped parameter methods for oceanic cables.
Engineering Structures 3(3), 153167.
Ma, W., Webster W.C., 1994. An analytical approach to
cable dynamics: theory and user manual. SEA GRANT
PROJECT R/OE-26. Department of Naval Architecture
and Offshore Engineering, University of California at
Berkeley, September 1994.
Nakamura M, W. Koterayagdgdg Y. Kyozuka, 1991. Slow
drif damping due to drag forces acting on mooring lines.
Ocean Engineering; 18:283296.
OBrien, T., 1967. General solution of suspended cable problem. Journal of Structural Division.ASCE, 93(ST1):126.
OBrien, W.T. andA.J. Francis. 1964. Cable movements under
two-dimensional loads. Journal of the Structural Division,
ASCE, 90(ST3):89123.
Peyrot,A. H., 1980, Marine Cable Structures, Journal of the
Structural Division, American Society of Civil Engineers,
Dec., 23912404.
Peyrot, A.H., A.M. Goulios, 1979. Analysis of cable structures. Computers and Structures 10 (5), 805813.
Smith, R.J., 2001. Tautslack dynamics of a vertically suspended subsea unit. Proceedings 20thInternational Conference on Offshore Mechanics and Arctic Engineering
(OMAE01), Rio de Janeiro, June 38, paper OFT-1287.
Tjavaras, A.A., Q. Zhu,Y. Liu, M.S. Triantafyllou, D.K.P.Yue,
1998. The mechanics of highly-extensible cables. Journal
of Sound and Vibration, 213:709737.
Walton, T.S., Polachek, H., 1959. Calculation of nonlinear
transient motion of cables, David Taylor Model Basin
Repost 1279.
Webster, R.L., 1975. Non-linear static and dynamic response
of underwater cable structures using finite element
approach. Proceedings of the 7th Offshore Technology
Conference, Houston, OTC 2322, 753764.
Webster, R.L. and W.R. McCreight, 1979. Analysis of deep
sea moor and cable structures. Proceedings of the 11th
Offshore Technology Conference, OTC 3623, Houston,
22992311
Yu, L., J. Tan, 2006. Numerical investigation of seabed interaction in time domain analysis of mooring cables, Journal
of Hydrodynamics, Ser. B,18(4):424430.
203

Estimation of hull bending moment and shear force from ship motions
using neural networks
L. Moreira & C. Guedes Soares
ABSTRACT: A technique is presented for estimating time-domain wave-induced ship hull bending moments
and shear forces from ship motion measurements based on Artificial Neural Networks (ANNs). The objective
of the use of this methodology is to obtain an expedite hull monitoring tool to be used on board the ships. The
ANNs are used to model the time-domain relationship between the ship motions and the wave-induced vertical
bending moments and shear forces. A mathematical model based on a strip theory formulation provides the ship
motions data in the time domain allowing the validation of the method through numerical simulations.
1
INTRODUCTION
In its most basic form, a Hull Response Monitoring System (HRMS) is a system that measures
and displays key ship motions and hull structural
responses. By monitoring real-time motions and
stresses, mariners can determine the onset and severity
of hull structural response to the sea. Hull response can
be measured either directly by strain gauges or indirectly by monitoring pressures and motions. Mariners
can then initiate ship handling changes (course and
speed) to mitigate dangerous stress levels and other
hazards (Slaughter et al., 1997).
The permanent monitoring of the main ship structures can be considered a very important issue,
because, besides representing an important tool for
checking the state of the structure during the entire
ships life, it also allows to increase the safety levels of the hull while navigating: real-time information
regarding the current status of the main parameters
of the hull response at sea are, in fact, provided to
the crew and this is a valuable help to the navigation,
particularly in severe sea states, when higher stresses
occur within the structures and the risks of a structural
failure grow up.
HRMS capabilities can be extended by measuring,
recording, and analyzing hull stresses in conjunction
with other ship motion, navigational, and performance
data. Extended benefits include fatigue assessment,
decision rules and guidance to assist the mariner in
mitigating current dangers, and quantifying design
constraints for future ships. In its most general form,
an HRMS can be integrated with remote assets such
as weather prediction to optimize routing on the basis
of hull response, ship motion, fuel consumption, and
other parameters (Slaughter et al., 1997).
Significant research has been conducted in an
attempt to estimate different parameters from ship
motion readings. It is difficult to obtain accurate predictions of wave induces loads in a ship during its
operation (Xu & Haddara, 2001). However some techniques are available to determine wave characteristics
from ship motions, as for instance the ones presented
by Pascoal & Guedes Soares (2008, 2009).
A decision support system that has been developed
to provide support to decisions about course and speed
changes in heavy weather (Perera et al., 2011) includes
both approaches for redundancy.
The wave spectra that the ship is experiencing is
estimated using the method of Pascoal & Guedes
Soares (2009) and with the estimated spectra the
response spectra of the various responses, including
wave induced loads, are calculated based on standard
response amplitude operators. The other approach is
to use directly the measured motions to assess the
corresponding wave induced loads using the method
described in this paper.
A model of an ANN is developed relating the time
domain motions that are being monitored at certain
positions of the ship, to the vertical bending moments
and shear forces that are being experienced in the
midship area of the ship. This ANN model is trained
with the results of a time domain code that predicts
ships motions and loads for various sea states, headings and speeds.
The mathematical model used here is the one proposed by Fonseca & Guedes Soares (1998), which is
a time domain code based on the strip theory formulation of Salvesen et al. (1970), which predicts the
vertical motions and sectional induced loads of ships
in large amplitude waves.
The learning approach adopted for the ANN does
not require the form of the function to be specified a
priori. In this work, the multilayer perceptron (MLP)
is used, which is one of the most widely implemented
neural network topologies. For the network training
205
Backpropagation, is employed, which is a common

method of training ANNs to perform a given task.
Simulated data from a container ship is used to
demonstrate the technique.
SHIP MOTIONS AND WAVE LOADS
The time domain method of Fonseca & Guedes Soares

(1998) is used in the present paper to predict the
motions and sectional induced loads of the ship in
waves. The solution is obtained in the time domain
using convolution integrals to account for the memory effects related to the free surface oscillations. In
this way the radiation forces are linear and represented in terms of impulse response functions, infinite
frequency added masses and radiation restoring
coefficients. The diffraction forces associated with
incident wave scattering are linear. The hydrostatic and
Froude-Krylov forces are evaluated over the instantaneous wetted surface of the hull to account for the
large amplitude motions and hull flare. The linear and
non-linear terms are used in the equations of motion
to compute the motions.
Considering a ship advancing in waves and oscillating as an unrestrained rigid body, the oscillatory
motions will consist of three translations and three
rotations. The coordinate system and the displacements are represented in Figure 1. Although the theory
which will be used can be applied for arbitrary headings relative to waves, the present work is restricted to
head waves.
Neglecting the viscous effects the fluid motion is
irrotacional and the hydrodynamic problem may be
formulated in terms of potential flow theory, thus
the fluid velocity vector may be represented by the
gradient of a velocity potential. The time domain
formulation derived by Cummins (1962) is used to
represent the radiation forces in terms of unknown
velocity potentials. The basic assumption is the linearity of the radiation forces.
The frequency dependent added masses and damping coefficients are calculated using a strip theory
(Salvesen et al., 1970). The sectional hydrodynamic
coefficients are calculated using a multi-parameter
conformal mapping method (Ramos & Guedes Soares,
1997), which ensures that the resulting added masses
and damping coefficients curves are very smooth,
even for high frequencies.
The wave exciting forces acting on the ship advancing with constant speed can be separated into the so
called Froude-Krylov part which is associated with
the wave incident potential, and the diffraction part
which represents the effects of the disturbance in the
incident potential caused by the presence of the ship.
In the method proposed by Fonseca & Guedes Soares
(1998) and used in this work the Froude-Krylov part of
the exciting forces (as well as the hydrostatic forces) is
calculated over the variable hull wetted surface under
the incident wave elevation. For the diffraction part
of the exciting forces it is assumed that the ship is
Figure 1. Conventional coordinate system for ship oscillatory motions.
advancing through the free surface and restrained at

its mean position. These forces are linear and calculated applying the Haskind-Newman relations together
with a strip theory approach (Salvesen et al., 1970).
The shear force at a cross section is the difference
between the inertia forces and the sum of the hydrodynamic forces acting on the part of the hull forward
of that section. In the same way the vertical bending moment is the difference between the moment
due to inertia and the moments due to the hydrodynamic forces. The hydrostatic contribution from each
cross section is given by the difference between the
static equilibrium hydrostatic force and the actual
hydrostatic force calculated on the exact wetted
surface. Since the ship is a stable system and the radiation forces are linear, it is possible to obtain of the
motions through Fourier transforms. With the formulation presented in Fonseca & Guedes Soares (1998)
and adopted here, the contribution for the vertical shear
force and bending moment of the radiation forces may
be calculated in the time domain for non-harmonic
motions, taking into account in a proper way the memory effects of the past history of the fluid motion. One
example of time histories of shear force and bending
moment for the ADEE containership is presented in
Figure 2, for a sea state with a significant wave height
of 2 m and a peak period of 18 s. Time histories of sway
and heave accelerations, roll, pitch and heading angles
as well as of the yaw rate are presented in Figure 3,
for a sea state with a significant wave height of 3 m
and a peak period of 6 s.
NEURAL NETWORK TRAINING
An ANN is a mathematical model or computational

model that simulates the structure and the functional aspects of biological neural networks. ANNs
are highly distributed interconnections of adaptive
nonlinear processing elements (PEs), i.e. they consist
of an interconnected group of artificial neurons and
process information using a connectionist approach
206
Figure 4. An example of a simple feedforward network.
Figure 2. Time histories of shear force and bending moment

of the ADEE containership Hs = 2 m, Tp = 18 s.
Figure 3. Time histories of the motions of the ADEE

containership Hs = 3 m, Tp = 6 s.
to computation. When implemented in digital hardware, the PE is a simple sum of products followed by
nonlinearity (McCulloch-Pitts neuron). The connection strengths, also called the network weights, can be
adapted such that the output of the network matches
a desired response.
In most cases an ANN is an adaptive system that
changes its structure based on external or internal
information that flows through the network during the
learning phase. Adaptation is the ability to change the
system parameters according to some rule (normally,
minimization of an error function). Adaptation enables
the system to search for optimal performance.AnANN
is a non-linear statistical data modeling tool. It can be
used to model complex relationships between inputs
and outputs or to find patterns in data. Distributed
computation has the advantages of reliability, fault
tolerance, high throughput (division of computation
tasks) and cooperative computing.
The MLP is one of the most widely implemented

neural network topologies. Lippmann (1987) is probably one of the best references for the computational
capabilities of MLPs. In general, for static pattern
classification, the MLP with two hidden layers is a
universal pattern classifier. In other words, the discriminant functions can take any shape, as required by
the input data clusters. Moreover, when the weights
and the output classes are properly normalized, the
MLP achieves the performance of the maximum a
posteriori receiver, which is optimal from a classification point of view (Makhoul, 1991). In terms of
mapping abilities, the MLP is believed to be capable
of approximating arbitrary functions.
MLPs are normally trained with the backpropagation algorithm. In fact the renewed interest in ANNs
was in part triggered by the existence of backpropagation. The Least Mean Squares (LMS) learning
algorithm proposed by Bernard Widrow in 1960 cannot be extended to hidden PEs, since the desired
signal it is not known there. The backpropagation rule
propagates the errors through the network and allows
adaptation of the hidden PEs.
Two important characteristics of the MLP are: its
nonlinear PEs which have a nonlinearity that must
be smooth (the logistic function and the hyperbolic
tangent are the most widely used); and their massive interconnectivity (i.e. any element of a given
layer feeds all the elements of the next layer). The
MLP is trained with error correction learning, which
means that the desired response for the system must
be known. In pattern recognition this is normally the
case, since the input data is labeled, i.e. it is known
which data belongs to which experiment.
In this work a single-hidden-layer MLP network
is used to the time-domain relationship between the
wave-induced vertical bending moment and shear
force and the ship motions. An example of a general
structure is illustrated in Figure 4. There are six input
neurons and two output neurons. A single bias neuron
is added to each of the input and hidden layers. The
number of the hidden neurons determines how well a
system can be modeled.
207
A hyperbolic tangent function is applied to each

neuron in the hidden layer, providing a network with
the ability to make soft decisions. The function is
defined by
where xilin = xi , with being an additional parameter

that controls the slope and is not adaptive; xi are the
inputs to the network and wi are the weights between
the input layer and the hidden layer.
The error correction learning works in the following
way: from the system response at PE i at iteration n,
yi (n), and the desired response di (n) for a given input
pattern an instantaneous error ei (n) is defined by
Using the theory of gradient descent learning, each

weight in the network can be adapted by correcting the present value of the weight with a term that
is proportional to the present input and error at the
weight, i.e.
The local error i (n) can be directly computed from

ei (n) at the output PE or can be computed as a weighted
sum of errors at the internal PEs. The constant is
called the step size. This procedure is called the backpropagation algorithm. Details and examples of this
algorithm are presented in Hush & Horne (1993).
Backpropagation computes the sensitivity of a cost
functional with respect to each weight in the network,
and updates each weight proportional to the sensitivity. The advantage of this procedure is that it can be
implemented with local information and requires just
few multiplications per weight, which is very efficient.
Because this is a gradient descent procedure, it only
uses the local information so it can be caught in a
local minima. Moreover, the procedure is inherently
noisy due to the use of a poor estimate of the gradient,
causing slow convergence.
In order to improve the straight gradient descent
a memory term (the past increment to the weight) is
introduced, the momentum learning, that will speed up
and stabilize the convergence. In momentum learning
the equation to update the weights becomes
where is the momentum. Normally should be set

between 0.1 and 0.9.
During the training process, the backpropagation
algorithm uses a gradient search technique to find the
connecting weights that minimize a criterion function.
In this work the criterion function is quadratic cost
function, which is the most applied cost function in

adaptive systems, given by
The error provided to the supervised learning procedure is simply the squared Euclidean distance between
the network outputs and the desired response:
The stop criterion in this work is based in a maximum number of training epochs. Once the network
is trained, its weights are saved to a file. This allows
a convenient interface with which to extract trained
network weights to be used by another application,
or to save the weights of several trials and keep the
best results. The network saves the best weights (the
weights with the lowest cross validation error).
In this work a single-hidden-layer MLP network
is implemented in NeuroSolutions 5 which is a
graphical neural network development software that
enables to easily create a neural network model for the
required data. In the training process six input variables and two output variables are used as referred
above. The number of hidden neurons chosen for this
first study was 10.
4
SIMULATED RESULTS
Simulated data is used to validate the neural network

model. Several runs were performed with the referred
seakeeping code using the M/V ADEE container
vessel. The main particulars of the ship are presented
in Table 1. Two pictures are shown in Figure 5.
As discussed in Perera et al. (2011) the onboard
system will include one accelerometer in the center of
gravity to measure the sway and heave accelerations,
obtaining also yaw rate and pitch, roll and heading
angle.
In the results presented here only simulations of
the ship in head waves are considered. The results
were obtained simulating the ship subjected to random
Table 1.
M/V ADEE ship main particulars.
Length overall (LOA) (m)

Length at waterline (LWL) (m)
Length between perpendiculars (LPP) (m)
Breadth (B) (m)
Depth (D) (m)
Draft (T) (m)
Block coefficient (CB)
Center of buoyancy aft of midship (LCB) (m)
Midship coefficient (CM)
Waterline coefficient (CWL)
Lightweight (LW) (ton)
Deadweight all told (DWAT) (ton)
Displacement () (ton)
208
128.00
119.70
117.60
20.20
10.90
8.28
0.653
0.106
0.977
0.850
4040.000
9167.000
12811.475
waves with JONSWAP spectra. The frequency range

of the wave spectra was chosen to cover frequencies
from 0.04 to 0.17 Hz. In every simulation run the
values of the midship bending moment, shear force,
sway and heave accelerations, roll, pitch and heading

angles and yaw rate are obtained.
300 training points of each input and output were
used to train the network, corresponding to 300s long
random time histories. The number of the hidden neurons was set to 10 at this first stage. During the training
process 100 points were used as cross-validation to
evaluate the convergence.
The correlation coefficient r is computed to determine how well the network output fits the desired output. By definition, the correlation coefficient between
a network output x and a desired output d is given by:
The best r values obtained for the estimation of the

bending moment are listed in Table 2.
The bending moment absolute and relative error
means as well as the corresponding standard deviations for the best r values showed in Table 2 are listed
in Table 3. In Table 3 are also presented the values
of the bias calculated through the absolute value of
Table 2. Best r results of the neural network training for the
bending moment estimation.
Figure 5. Photos of the M/V ADEE containership.

Table 3.
Peak
Frequency
(Hz)
Significant Wave
Height (m)
Shear
Force r
Bending
Moment r
0.038
0.042
0.038
0.111
0.3
2
2.5
3
0.847
0.831
0.839
0.791
0.763
0.814
0.797
0.808
Errors of the best results of the neural network training for the bending moment estimation.
Peak
Frequency (Hz)
Significant
Wave
Height (m)
Bending
Moment Absolut
Error Mean
Bending Moment
Absolut Error
Standard
Deviation
Bending Moment
Relative
Error Mean
Bending Moment
Relative Error
Standard
Deviation
Bias
0.038
0.042
0.038
0.111
0.3
2
2.5
3
0.075
0.068
0.071
0.097
0.056
0.051
0.053
0.087
0.162
0.141
0.151
0.227
0.167
0.131
0.155
0.259
0.000
0.001
0.000
0.010
Table 4. Mean, standard deviation, maximum and minimum values of r: training results for simulation runs in head waves
performed with JONSWAP spectra.
Hs (m)
Mean SF r
Mean BM r
= SF r
= BM r
Min. SF r
Min. BM r
Max. SF r
Max. BM r
0.3
2
2.5
3
0.522
0.529
0.506
0.582
0.531
0.558
0.589
0.632
0.172
0.178
0.155
0.146
0.128
0.156
0.143
0.111
0.301
0.302
0.326
0.339
0.386
0.294
0.429
0.482
0.847
0.831
0.839
0.791
0.763
0.814
0.797
0.808
209
Figure 6. Bending moment estimation Hs = 0.3 m;

Tp = 26 s.
Figure 9. Bending moment estimation Hs = 3 m; Tp = 9 s.
Figure 7. Bending moment estimation Hs = 2 m;

Tp = 24 s.
CONCLUSIONS
A methodology based on artificial neural networks has

been developed to obtain estimates of wave-induced
vertical bending moment and shear force from sway,
heave, roll, pitch, heading and yaw measurements. The
final neural network model is able to provide accurate
estimates of the instantaneous bending moment and
shear force. The main objective of this technique is to
be incorporated into a hull monitoring system.
The data used to calibrate and validate the neural network model was obtained through time-domain
simulations. In future work experimental data obtained
through tests can be also used to validate the neural
network model.
The presented results were obtained through the
use of a seakeeping code: the equations of motion are
formulated using strip theory, considering a constant
mean forward speed, for a required heading, in irregular waves described by a JONSWAP spectrum. It is
assumed that the ship is rigid and the responses are
linear.
ACKNOWLEDGEMENTS
Figure 8. Bending moment estimation Hs = 2.5 m;

Tp = 26 s.
the difference between the mean of the whole set of

original data for the bending moment and the mean of
the respective simulated data.
Data obtained from head waves simulations with
JONSWAP spectra were also used for validating the
neural network. The mean, standard deviation, maximum e minimum values of the r values are presented
in Table 4.
One can observe that reasonable results can be
obtained using a low number of neurons. However for
some frequencies the agreement is not so good but this
is expected due to resonance effects.
Figures 69 show comparisons between the neural
network estimation and the values obtained from the
simulations. Time histories of 125 s long are shown
and the values of the bending moment are normalized
between 0 and 1. The pictures illustrate the very good
agreement between the neural network estimates and
the results obtained through simulations.
This work is done within the project of Handling

Waves: Decision Support System for Ship Operation in Rough Weather, which is being partially
funded by the European Commission, under contract
TST5-CT-2006-031489.
The work of the first author has been supported by
research fellowships of the Portuguese Foundation for
Science and Technology (Fundao para a Cincia e a
Tecnologia) under contract SFRH/BPD/ 48088/2008.
REFERENCES
Cummins, W.E. 1962. The impulse response function and
ship motions. Schiffstechnik 9: 101109.
Fonseca, N. & Guedes Soares, C. 1998.Time-domain analysis
of large-amplitude vertical ship motions and wave loads.
Journal of Ship Research 42(2): 139153.
Hush, D.R. & Horne, B.G. 1993. Progress in supervised
neural networks. IEEE Signal Processing Magazine 10(1):
839.
Lippmann, R.P. 1987. An introduction to computing with
neural nets. IEEE Trans. ASSP Magazine 4(2): 422.
Makhoul, J. 1991. Pattern recognition properties of neural
networks. In Proc. 1991 IEEE Workshop on Neural
210
Networks for Signal Processing, 30 September1 October,

Princeton, NJ, USA.
Pascoal, R. and Guedes Soares, C. 2008; Non-Parametric
Wave Spectral Estimation Using Vessel Motions. Applied
Ocean Research. 30:4653.
Pascoal, R. & Guedes Soares, C. 2009. Kalman filtering
of vessel motions for ocean wave directional spectrum
estimation. Ocean Engineering 36(67): 477488.
Perera, L. P., Rodrigues, J. M., Pascoal, R., and Guedes
Soares, C. 2011, Development of an onboard decision support system for ship navigation under rough
weather conditions. Sustainable Maritime Transportation
and Exploitation of Sea Resources, E. Rizzuto, C. Guedes
Soares (Eds), Taylor and Francis Group, (in press)
Ramos, J. & Guedes Soares, C. 1997. On the assessment

of hydrodynamic coefficients of cylinders in heaving.
Ocean Engineering 24(8): 743763.
Salvesen, N., Tuck, E.0. & Faltisen, 0. 1970. Ship motions
and sea loads. Trans. SNAME 78: 250287.
Slaughter, S.B., Cheung, M.C., Sucharski, D. & Cowper, B.
1997. State of the art in hull response monitoring systems.
Ship Structure Committee, Report SSC401.
Xu, J. & Haddara, M.R. 2001. Estimation of wave-induced
ship hull bending moment from ship motion measurements. Marine Structures 14(6): pp. 593610.
211

Wave frequency dynamics of a tanker moored inside a port

A. Pao, N. Fonseca & C. Guedes Soares
ABSTRACT: The paper presents an analysis of the wave frequency induced motions of a tanker moored in
several restrained conditions that resemble the conditions inside the port. One of the objectives is to investigate
the influence of several parameters on the horizontal and vertical motions, including: the water depth, presence
and distance of a vertical wall, influence of dolphins and the effects of the mooring lines. The second objective is
to assess the ability of the existing numerical method to predict the ship dynamics in close proximity to several
boundaries and including the mooring line effects. The calculations are carried out in the frequency domain
with a three dimensional Grenns function panel method. The mooring line forces are firstly calculated exactly
as function of the ship displacements and then linearized around the equilibrium point to obtain the stiffness
matrix. The later is added to the hydrostatic stiffness matrix. Finally, numerical predictions are compared with
experimental data obtained with a scaled model on a wave tank in regular and irregular wave.
1
INTRODUCTION
the vessel motions. Some realistic sea state conditions

are also considered.
Downtime at ocean facing ports is often related

with excessive moored ship motions caused by wave
action. Reducing the amplitude of the moored ships
motions is therefore crucial to increase the on and
off-loading operations efficiency, to minimize port
operational costs, as well as to reduce environmental
risks, especially when dealing with dangerous cargoes.
In this context, the motions of an oil tanker moored
at the Berth A oil terminal of the Port of Leixes is
being studied within the scope of an R&D project
DOLPHIN. The objective is to better clarify the contribution of some of the identified critical issues on Berth
A downtime and to analyze the effectiveness of some
intervention alternatives proposed in previous studies.
This project includes physical model tests, numerical
simulations and prototype measurements at Berth A,
as described in Taveira Pinto, et al. (2008), and Rosa
Santos, et al. (2009).
This paper focuses on the results of numerical studies. The present phase is focused on the ship motions
with the ship moored at the existing terminal layout. A
large number of simulations were carried out to obtain
a wide perspective of the moored ships behavior. The
influence on the wave induced ship motions of the
following parameters is investigated:
Water depth on non-moored ship;

Proximity of a vertical wall;
Presence of the dolphins and the of the mooring
lines;
Water depth on the moored ship.
Finally, some of the results are compared with
experimental data. The study focus mainly in regular
waves with the aim of obtain the transfer functions of
2 THEORY
This section briefly presents the theory behind the
method to calculate the moored ship motions.
2.1 Transfer function calculation in regular waves
The transfer functions are calculated by solving the
6 degree of freedom motion problem. Applying the
Newtons second law to the system the motion equations might be found as:
where M is the mass matrix of the system, in this case

of the ship, A and B represents the matrices of hydrodynamic added mass and damping coefficients, both
function of the wave incident frequency. Bvisc represent
the viscous damping coefficients (in case they are necessary), C is the hydrostatic restoring matrix and K is
the stiffness matrix of the mooring lines. E represents
the vector of the exciting forces due to waves. Finally
is the resulting ships motion vector composed of
three translations (surge, sway and heave) and three
rotations (roll, pitch and yaw).
Once the coefficients of the equations of motion
are determined, the equations of motion are solved for
several wave frequencies to provide the results on a
213
large frequency range. The hydrodynamic coefficients

and exciting forces result from the solution of the linear
boundary value problem and they are given by:
where is the water density, Sb the body mean wetted

surface, j is the radiated wave velocity potential due
to the motion on mode j and ni the component in the i
direction of the unit vector normal to the body surface
pointing outwards of the body domain.
The wave exciting forces can be calculated through
the diffraction potential, simply by integrating the
pressure caused by the incident wave field and the
pressure resulting from the perturbation that the body
induces in the wave field.
The first result is known as Froude-Krilov forces
and is associated to the incident wave potential 0 .
The second is known as diffraction force and is
calculated by:
D is the diffraction velocity potential.

The inertial forces are proportional to the body
accelerations.
The hydrostatic restitution forces are a result of the
combination of the hydrostatic pressure integration
with the weight of the floating body. The integrated
pressure force is applied at buoyancy centre of the vessel. On the other hand, the weight of the body is applied
on the centre of gravity. The variation around the equilibrium condition of these forces is the cause for the
hydrostatic restoring forces. For ships with a longitudinal plane of symmetry the only nonzero restoring
coefficients are:
point in space is a stochastic variable which follows a

Gaussian distribution with zero mean.
A vessels response to an incident harmonic wave
can be seen as a linear transformation of the incident
wave.As long as the relation between the incident wave
amplitude and any given quantity remains linear, it is
possible to calculate transfer functions for that quantity
by simply calculating it for a unitary amplitude wave.
The probability theory states that the transformation
of a random, time independent and Gaussian process
will result in another process with similar characteristics. Therefore, since the vessels responses are linear,
it is possible to relate the stationary Gaussian wave
field with the stationary Gaussian vessels motion. This
means that it is possible to obtain a response spectrum
(Sj ) by combining the incident wave spectrum (S ) with
the transfer function amplitude (xja ):
The seastate is modelled as a stationary, zero mean,

Gaussian process and because the responses are linear,
the response process will be described by the same
model.
Since this is a zero mean process the standard
deviation is the root mean square value. The variance
can be obtained from the spectrum integrating:
which is applicable to both the input and the response

spectrum.
Different statistics can be derived from the assumption of the Rayleigh distribution. For example the
average of the one-third largest wave heights:
Likewise, ship responses can be calculated based on

this methodology. The following expressions are used
to determinate the mean (11), significant (11) and the
one-tenth larger amplitudes (13).
where g is the acceleration of gravity, and zb , zg represent respectively the vertical position of the center of
buoyancy and center of gravity of the ship.
2.2 Calculations in irregular seas
The seastate is assumed as a zero mean Gaussian process, meaning that the irregular wave elevation at any
2.3 Mooring stiffness coefficients

The matrix of the restoring coefficients, K, is related
to the mooring system and it is assumed that the ships
mooring lines and fenders forces on the ship are purely
elastic. Furthermore, it is assumed that the lines and
fenders have a pre-tension so that they do not become
slack with the motions of the ship.
214
Analogously, for the force applied in B:
The direction of the moment about O1 caused by the

force applied in A is:
Figure 1. Definition of the mooring line main characteristics and nomenclature.
To obtain the restoring coefficient due to a generic

spring type force with respect to a given reference system is simple. Let F be a force applied on a point with
coordinates (x, y, z):
Point A will change the location with the motions of

the ship, however B will remain constant.
Locations will be called A and B, and are given by:
where the translational and rotational motions of the

ship are given by:
The moment about the origin of the reference
system is given by:
The line stretch is given by:
where r is the position vector of the point of application

of the force with respect to the reference system. If Fk
is a generalized spring type force in the k direction
(k = 1,6), then the spring coefficient associated to
this forces due to an imposed motion in the j mode is
given by:
In the case of mooring systems of ships floating

inside ports, it is important to define correctly the force
and moment that an arbitrary line will induce on the
ship. Let O1 be the reference system which is fixed to
the ship (see Figure 1)
The point where the line is attached to ship will be
defined as A, and its coordinates in respect to O1 are:
The point where the line is connected to the berth will

be defined as B. Its coordinates are given by:
The direction of the resulting spring force applied in

A is given by the unit vector:
and so, the forces applied in A are:
where k is the stiffness constant of the line, which

depends on the material, line diameter and line length.
The former equations are applied for imposed
motions on all six directions of the coordinate system, and repeated for all mooring lines and fenders.
Combining the result with equation (16) results on the
stiffness coefficients of the mooring system restoring
matrix.
3
CASE STUDY
3.1 The berth A of Leixes oil terminal

The Port of Leixes, located in the northwest coast of
Portugal and facing the North Atlantic, has an oil terminal composed of 3 berths (Figure 2). Despite the
protection offered by the Leixes north breakwater,
Berth A, located at the harbor entrance, does not
assure the operational and safety conditions during
about 20% of the time. Ships moored at this berth may
have excessive wave induce motions and breaking of
mooring lines has occurred.
The Berth A jetty structure consists of two breasting dolphins and a loading platform. Each breasting
dolphin is equipped with a pneumatic fender and double mooring hooks. The remaining terminal mooring
hooks are located on the north breakwater superstructure. Alongside this berth the bottom is about
16 m, which allows receiving oil tankers of up to
100,000 dwt.
215
Figure 2. Leixes oil terminal, Porto, Portugal.
Table 1.
Ship main particulars.
Vessel Kornati crude oil tanker

Displacement (sea water)
Length over all
Breadth moulded
Depth moulded
Mean draught
Vertical center of gravity*
Longitudinal center of gravity **
Metacentric height
Transversal radius of gyration
Longitudinal radius of gyration
D (t)
Loa (m)
Lbp (m)
B (m)
D (m)
T (m)
VCG (m)
LCG (m)
GMt (m)
ix (m)
iy (m)
Figure 3. Physical model of the ship moored at a simplified

berth A.
122 714
245.00
236.00
43.00
20.00
14.10
12.46
128.36
5.83
15.10
61.40
*From keel; **from the stern
Figure 4. Mooring layout.
3.2 The vessel Kornati
3.4 Numerical methodology
The ship selected for the study intends to represent

the biggest class of oil tankers that regularly demand
Berth A and corresponds to an AFRAMAX crude
oil tanker of 105,000 dwt. Table 1 presents the main
particulars.
The WAMIT computer code was used to solve the

hydrodynamic problem and calculate the wave induced
ship motions. WAMIT is based on the linear potential
theory for analyzing floating or submerged bodies, in
the presence of ocean waves. The boundary integral
equation method (BIEM), also known as the panel
method, is used to solve for the velocity potential and
fluid pressure on the submerged surfaces of the bodies.
The wetted surface of the ship hull was discretized
with 898 panels and the code was used to solve the
radiation and diffraction problems of the free-floating
ship for 23 to 50 equally spaced frequencies ranging
from 0.07 rad/s to 1.00 rad/s.
The JONSWAP wave spectrum was used the irregular waves calculations.
3.3
Experimental data
The physical model study was carried out at the

Hydraulics Laboratory of the Hydraulics, Water
Resources and Environment Division of the Faculty
of Engineering of the University of Porto, on a geometric scale of 1/100. The existing wave tank is 28 m
long, 12 m wide and 1.2 m in depth.
For this study a simplified physical model of the
Berth A was considered. Bathymetry was considered
horizontal and the bottom level near the berth equal
to 16 m. The breasting and mooring dolphins were
reproduced in the physical model; however, there was
no need to construct the Leixes north breakwater as
the ship model, in this stage, is only submitted to head
waves. Those waves are the ones expected to reach
the Berth A area after diffraction around the head of
Leixes north breakwater.
The selected ship mooring layout is slightly asymmetrical and it represents the most usual mooring
layout for the largest class of oil tankers that use
Berth A. It is composed by 8 double mooring lines
(Figure 4).
This section presents the analysis of different parameters on the wave induced motion responses, namely,
the influence of the water depth on the free and isolated
floating ship, the influence of a vertical side wall and
the effects of the mooring system. The former results
are followed by comparisons with experimental data.
All results are presented in terms of transfer function amplitudes as function of the incident wave
period. The motion amplitudes are normalized by the
incident wave amplitude.
216
Figure 5. Surge transfer functions in open sea.

Figure 8. Surge transfer parallel to a vertical wall.
Figure 6. Sway transfer functions in open sea.

Figure 9. Sway transfer functions parallel to a vertical wall.
4.2 Presence and distance of a vertical wall
Figure 7. Yaw transfer functions in open sea.
4.1 Water depth influence

Regarding the water depth influence on the unrestrained ship, figures 5 to 7 present the horizontal
motions for different water depths (d between 16m and
infinite). Bow waves with a heading of 150 are considered. One observes a large increase of the motion
amplitudes with the decrease of the water depth for the
medium to large wave period range. The results show
that the ship hydrodynamics strongly modify with the
water depth.
The presence of a vertical quay in the vicinity of a

moored ship will modify the hydrodynamic responses
as compared to the isolated ship. This aspect is evaluated by computing the ship motion responses in the
presence of a vertical wall parallel to the longitudinal
plane of the ship, extending from the sea bottom to
the free surface. In the graphs of figures 8 to 10 the
horizontal ship motions are calculated for several distances between the vertical wall and the ship side (dw
from 1.5 m to 100 m). The results are for head waves.
One observes that the wall distance has no influence
on the surge motions in head waves. Furthermore the
surge motion is very small for wave periods smaller
than about 18 s. On the other hand, the influence of
the wall distance on the sway and yaw motions in head
waves is large and depends on the period range. For
example, the yaw transfer functions have a maximum
around 20 s, and then they decrease as the wave period
increases.
4.3 Effects of dolphins and mooring system
In order to assess the influence of the mooring system
on the motion responses, the dolphins, the mooring
lines and the fenders were considered in the numerical
solution. The dolphins consist of vertical structures
217
Figure 10. Yaw transfer functions.
Figure 12. Sway transfer functions.
Figure 11. Surge transfer functions.
with square cross section, fixed to the sea bottom

and intersecting the free surface. They were modeled with panels and considered on the hydrodynamic
calculations. The mooring lines and fenders effects
are included through an external stiffness matrix
calculated according to the procedure presented in
section 2.3.
The horizontal motions transfer functions are calculated for three water depths (d = 18, 24 and 32 m),
and the results of the moored ship are compared with
similar results for the free ship.
4.3.1 Head waves
Figures 11 to 13 show the amplitudes of surge, sway
and yaw in head waves as function of the incident wave
period. There is clearly a very large influence of the
mooring system on the motion responses. The surge
motion with mooring system has a natural period of
around 60 s, therefore one observes a large dynamic
amplification. Due to coupling effects, the sway and
yaw motions are amplified as well around the same
period range. The sway and yaw amplifications reduce
with as the water depth decreases due to the increase
of the related damping coefficients. The sway has a
natural period very close to the surge motion, while
the yaw motion presents a natural period around 45s.
The graphs of figures 14 to 16 present the moored
ship results with the dolphins (S. dock results)
together with the moored ship but including a vertical wall at a distance from the ships side of 50 m. One
observes that the resonant motion amplitudes increase

when side wall effects are included.
4.3.2 Beam waves
The analysis presented in the previous sub-section for
head waves is repeated for beam waves, namely the
comparison of horizontal motion responses for the free
ship and the ship moored to the dolphins (figures 17
to 19) and the comparison between the later and the
moored ship including the lateral vertical wall located
50m from the ship side (figures 20 to 22).
The conclusions are similar to the former: the mooring system introduces large dynamic amplifications
for the natural periods of the horizontal motions, and
218

the presence of the vertical wall increases the motion

responses.
4.4 Comparison of numerical predictions
with experimental data
This section presents the comparison between experimental data and numerical predictions. The first set of
results consists on the responses to regular waves of
the free model, therefore without the presence of the
dolphins and of the mooring system. Some results are
presented in the graphs of figures 23 to 26, for bow
waves (135 ) and beam waves (90 ), as function of the
incident regular wave period. The agreement between
experiments and numerical results is reasonably good.
219
Figure 25. Roll numerical and experimental amplitudes in

regular waves, free model.
Figure 23. Pitch numerical and experimental amplitudes.
Figure 26. Pitch numerical and experimental amplitudes in

Figure 24. Yaw numerical and experimental amplitudes in

The second group of comparative results is for the

tanker model moored to the dolphins, however without
the presence of the side quay (see figure 3). Figures 27
and 28 show the amplitudes of surge and heave motions
in head regular waves for three wave periods. The
agreement between experiments and numerical results
is good for the two longer periods and reasonable for
the shorter wave period.
Several experimental tests were carried out also in
irregular long crested seastates. The target wave spectra was the JONSWAP with a peakedness parameter of
3.3. The significant wave height was 2 m, while 5 peak
wave periods were tested ranging from 10 s to 18 s.
Figures 29, 30 and 31 present some statistics of
the experimental and numerical surge, heave and yaw
Figure 27. Surge numerical and experimental amplitudes in

regular waves, model moored to the dolphins.
Figure 28. Heave numerical and experimental amplitudes

in regular waves, model moored to the dolphins.
220
Figure 31. Yaw: mean, significant, and one-tenth larger

amplitudes.
Figure 29. Surge: mean, significant, and one-tenth larger
amplitudes.
Figure 30. Heave: mean, significant, and one-tenth larger

amplitudes.
motions as function of the peak wave period, namely

the mean amplitudes, the significant amplitudes and
the mean of the 1/10 largest amplitudes.
It is possible to conclude that the agreement
between experiments and predictions is good for the
surge motion, while the numerical model underestimates the heave and yaw motions. The reason for the
underestimation of the response statistics is not clear.
5
CONCLUSIONS
The paper presents an investigation of the wave

induced motions on a tanker moored in several
restrained conditions similar to the inside a port conditions. The calculations are carried out with a frequency
domain Greens function panel method. The analysis
is restricted to linear wave frequency motions.
It is concluded that the horizontal motions increase

very much as the water depth decreases from the
medium to the large wave period range. The realistic mooring system investigated introduces resonance
conditions on the horizontal motions for wave periods
around 60 s. Large dynamic amplification was identified for these conditions. The horizontal motion amplitudes increase when the vertical quay is considered in
the hydrodynamic calculations.
The numerical model was assessed by comparison
with model test results. While the agreement between
experiments and predictions is between good and reasonable for the motions in regular waves of the free
ship and ship moored to the dolphins, for irregular
waves the agreement is good only for the surge motion.
The statistics of the heave and yaw motions in irregular waves are under predicted by the numerical model
and the reasons are not clear.
ACKNOWLEDGEMENTS
The work was developed in the scope of the DOLPHIN
project Analysis of the hydrodynamics behavior of
berthing and mooring structures. The Leixes Harbour
Oil Terminal Case Study funded by the Portuguese
Science and Technology Foundation under contract
PTDC/ECM/72835/2006.
REFERENCES
Taveira Pinto, F.; Veloso Gomes, F.; Rosa Santos, P.;
Guedes Soares, C.; Fonseca, N.; Alfredo Santos, J.; Paulo
Moreira, A.; Costa, P., and Brgueira Dias, E. 2008; Analysis of the Behavior of Moored Tankers. Proceedings of
the 27thInternational Conference on Offshore Mechanics
and Arctic Engineering (OMAE 2008); Estoril, Portugal.
New York, USA: ASME; paper OMAE2008-58013.
Rosa Santos, P., Veloso Gomes, F., Taveira Pinto, F., et al.
(2009), Influence of the use of mooring line pretension
on the behaviour of a moored oil tanker, Proceedings
International Conference in Ocean Engineering, (ICOE
2009), IIT Madras, Chennai, India, 15 of March 2009.
221

Experiment and time domain method comparison for the responses of a

containership induced by the three sisters abnormal waves
S. Rajendran, N. Fonseca & C. Guedes Soares
ABSTRACT: A time domain code based on the linear model for diffraction and radiation forces combined with
a nonlinear model for Froude-Krylov and hydrostatic forces is used to predict the responses of a containership in
a predefined sea state with embedded abnormal waves. The responses include both vertical motions and bending
moments. These results are compared with experiments conducted in a wave tank with the aforementioned
containership for different Froude numbers in the predefined wave sequence known as the Three Sisters.
The main purpose behind the experiment is to check the behaviour of the ship in extreme sea states. These
experiments revealed little but the key influence of green water shipping on the ship responses especially on the
sagging moment and is used as a guideline to modify the time domain code in a heuristic manner to incorporate
the green water effect. The updated numerical model is again tested and compared with the experiment for
further validation.
1
INTRODUCTION
Extreme sea conditions that involve waves with large

amplitude up to freak waves have always been an area
of interest for naval architects. Knowledge about the
behaviour of the sea going vessels in these extreme
conditions is very limited due to the unavailability of
real data and the complexity involved with the reproduction of such large deterministic wave profile in the
wave tank. Due to the same reason validation of the
numerical codes in the extreme sea condition is not
widely done.
However several methods varying from linear strip
theory (Salvesen et al., 1970) to completely nonlinear method (Beck, 1994; Subramani et al., 1998) have
been used to predict the ship response in moderate to
extreme sea conditions. The research presented here
uses the partial nonlinear numerical method developed by Fonseca and Guedes Soares (1998) to predict
the ship response in extreme conditions. The method
makes use of two- dimensional strip theory to calculate
the diffraction and radiation coefficients combined
with nonlinear Froude-Krylov forces and hydrostatic
forces calculated over the instantaneous wetted surface
under the undisturbed wave profile.
In addition, studies had been done before on shipping of green water which is a highly nonlinear
phenomenon associated with the sea keeping problem in extreme seas. ODea and Walder (1984),
Lloyd et al. (1985), Buchner (1995), Fonseca and
Guedes Soares (2004, 2005) investigated the problem
of green water shipping by conducting experiments on
a ship advancing in waves. These studies revealed the
effect of the bow flare on shipping of green water and
behaviour of water on deck. Still data available for the

moving ship are very limited.
The responses of ships and offshore platforms to
time series containing a freak wave have been studied,
among others by Guedes Soares et al. (2006, 2008).
These studies conducted a detailed time series comparison of the experiment and numerical vertical bending
moment along with relative motion at the bow and the
green water force on the deck of a FPSO and S-175
container ship when the freak wave (New Year Wave)
was generated at different locations along the ship
asymmetric nature of the vertical bending moment
time series was clearly identified and usage of nonlinear time domain code was emphasized. Further study
was done through this research on the response of the
container ship to a group of abnormal waves.
This paper compares numerical results with experimental data for a deterministic abnormal wave group
positioned at different locations on a container ship
moving with different Froude number and tries to
clarify the physical phenomenon behind the discrepancy in the results.
2
2.1
EXPERIMENTAL SETUP AND PROGRAMME

Experimental setup
The model tests were conducted in the seakeeping

basin of the Ocean Engineering Division, Technical
University Berlin (TUB), at model scale 1:70. The
seakeeping basin with the overall length of 110 m and
an effective measuring length of about 90 m, offers a
width of 8 m at a water depth of up to 1.1 m.A computer
controlled electrical driven wave generator, which can
223
Table 1.
Main particulars of the containership.
Length between perpendiculars [m], Lpp

Breadth [m], BWL
Draught [m], Dfp
Displacement [t]
Block coefficient [], CB
Longitudinal centre of gravity [m], Xcg
Vertical centre of gravity [m], Zcg
Transversal metacentric height [m], GMt
Scale of the model
Table 2.
117.6
20.2
8.1
12499
0.65
58.8
8.12
1.1
1:70
be used both as flap type and piston type, was used for
the experiment.
Waves in the tank were created by superposing
higher order components taking into account their non
linear interaction (Clauss et al., 2010). Using this technique it was possible to create pre-defined single freak
waves, like the well known New Year Wave (Clauss
et al., 2005), so as to study the effect of these abnormal
waves on the ships.
For investigating the influence of the wave height
on the vertical bending moment with special emphasis on effects resulting from the forecastle geometry,
three ship models are investigated, a Ro/Ro vessel, a
bulk carrier and a container vessel. This paper is concerned with the analysis of response/structural loads
acting on the containership. Table 1 shows the main
dimensions and load condition of the container ship.
The container ship model consists of two segments
connected at Lpp/2 with three force transducers, two
installed close to the deck level and other underneath the bottom of the model. The force transducers register the longitudinal forces during the model
tests. Based on the measured forces and the given
geometrical arrangement of the three force transducers, the resulting vertical wave bending moment and
the longitudinal forces are obtained. On this basis,
the superimposed vertical wave bending moment
including the counteracting vertical bending moment
caused by the longitudinal forces with respect to
selected vertical levels is determined.
Undisturbed wave elevation was measured with the
wave gauge mounted on the towing arrangement moving with speeds of interest but without the ship model.
Here the waves were measured in a moving frame of
reference with the measurement carried out either at
midship or forward perpendicular location. Heave and
pitch motions were kept unrestrained by a suspension
system connected with the ship model by a thin elastic cross bar, which is mounted on the deck of the aft
segment. These motions were measured by an optical
tracking system installed on a carriage moving on roof
mounted rails with the same velocity of ship. Clauss
et al. (2009) present further details of the experimental
setup.
Table 2 gives a brief description of the list of experiments. The 3 Sisters wave group, which was said to
be occurred in Lake Superior, was recreated in the
tank embedding in a dedicated irregular seastate. This
giant wave had a maximum wave height of 18.67 m
Details of the Experiment.
No
Type of wave
Target
Fn
1
2
3
4
5
6
7
8
3 Sisters
3 Sisters
3 Sisters
3 Sisters
3 Sisters
3 Sisters
3 Sisters
3 Sisters
Midship
F.P
Midship
F.P
Midship
F.P
Midship
F.P
0.00
0.00
0.06
0.06
0.12
0.12
0.18
0.18
and a wave crest height of 11.29 m and the seastate is

characterized by a significant wave height of 10.38 m
(Hmax /Hs = 1.80).
In Table 2, Target means the position along the
ship model where the abnormal wave crest is generated, which is either the midship or the forward perpendicular. Several Froude numbers were considered
only in head waves conditions.
3 THEORY
3.1
Numerical method
Regarding the time domain seakeeping code

(Fonseca & Guedes Soares, 1998a), the method
assumes that the nonlinear contribution for the vertical bending moment is dominated by hydrostatic and
Froude-Krylov forces, thus these components depend
on the instantaneous hull wetted surface. The exciting
forces due to the incident waves are decomposed into
the diffraction part and the Froude-Krylov part. The
diffraction part is kept linear and solved in the frequency domain. The resulting transfer functions are
used to generate time histories of the diffraction heave
force and pitch moment. The Froude-Krylov part is
calculated at each time step through the integration
of the associated pressure over the wetted surface of
the hull under the undisturbed wave profile.
The radiation forces are represented in the time
domain by infinite frequency added masses, radiation restoring coefficients and convolution integrals of
memory functions. Both the radiation and diffraction
coefficients in the frequency domain are calculated by
a strip method.
The vertical forces associated with the green water
on deck, which are taken to occur when the relative
motion is larger than the free board, are calculated
using the momentum method (Buchner 1995). The
mass of water on the deck is proportional to the height
of water on the deck, which is given by the difference
between the relative motion and the free board of the
platform.
4
4.1
RESULTS
Comparison of numerical simulation with
experimental data
Figure 1 shows the comparison of the wave profile of

3 Sisters waves simulated in the tank and the measured
224
Figure 1. Comparison between numerical results and experimental data of the input wave profile and vertical ship response
motions for 0 and 0.18 Froude number.
ship vertical response motions for a Froude number

of 0 with the numerical data. It was observed that
maximum crest of the larger wave occurs approx. at
1175 s followed by two large consecutive peaks which
occurs approx. at 1187 s and 1200 s, respectively. The
response motions were measured separately when the
New Year Wave maximum peak is at two target positions, at midship and FP. Both the linear and nonlinear
numerical model was able to reproduce the exact wave
profile and calculate the vertical response motions
that match well with the experimental data. It was
seen that the agreement between the numerical model
and the experimental ship response motions even in
extreme seas is very good.
Similar comparisons were performed in the above
mentioned manner for Froude numbers 0.06, 0.12 and
0.18. The response motions other than for Froude number 0.18 (Fig. 1) were omitted from the plots since
all of them follow the same trend. The wave amplitudes were reduced for higher Froude numbers since
the wave profile was measured in the moving frame
of reference of the ship. In the case of Froude number 0.18, the maximum crest occurs approx. at 1004 s
but the following two crests were not prominent as in
the 0 Froude number case. Here too, the agreement
between the numerical model and the experimental
vertical response motions is very good.
In the next stage of the research, the vertical bending
moment was calculated using both the linear and nonlinear method and was compared with the experimental bending moment. The study was mainly focused
around the period of occurrence of the 3 Sisters peaks
and their associated induced vertical bending moment

as shown in Fig. 14. It was seen that the linear method
gives large hogging peaks and smaller sagging peaks
compared to the experimental result. Hogging peaks
calculated using the nonlinear method were in good
agreement with the experimental result while the nonlinear numerical sagging peaks were larger than the
experimental data. This trend was followed for the
bending moment when the wave acts at the midship. In
general, it was observed that the nonlinear numerical
hogging peaks calculated for the wave at midship were
higher than the ones with the wave at FP.
It was observed that for the zero Froude number
with the wave at FP, the largest sagging peak did not
occur with the largest wave crest of the 3 Sisters (occurring at 1175 s at FP) but with the following crest (at
1187 s) as shown in Fig. 2. But for the same Froude
number with the wave at midship, the largest hogging peak occurs with the largest wave (at 1175 s).
These types of observations are impossible to generalize for higher Froude number due to complicated
ship responses. For instance, for the Froude number
0.18 with the wave acting at FP between 170 s and
1180 s as shown in Fig. 4, the maximum sagging peak
didnt coincide with largest wave crest but lies between
largest wave crest and following trough.
The largest hogging and sagging peaks from the
experiment and the numerical results were plotted
as shown in Fig. 5. As discussed earlier, discrepancy between the experimental hogging peaks with
the numerical results was less for the nonlinear
method. But for the largest sagging peak, the numerical
225
Figure 2. Comparison of vertical bending moment at midship from linear and nonlinear numerical results (input wave at FP
on left hand side and input wave at midhsip on right hand side) with the experimental data for Froude numbers 0.
on left hand side and input wave at midhsip on right hand side) with the experimental data for Froude numbers 0.06.
nonlinear method highly overestimated the experimental sagging peak. On close observation it was seen that
these experimental sagging peaks were cut off making
it lesser in value than the numerically calculated ones.
Further study on the relative motion revealed that
such effects coincide with the occurrence of the green
water shipping on the deck. A detailed study was conducted to check the green water effect on the bending
moment applying a heuristic method in modifying the
relative motion at the bow.
Green water effect can be calculated using the
formula developed by Buchner et al. (1995), which calculates the force due to the green water effect using the
rate of change of momentum and water height on the

deck. However, this formula requires an exact estimate
of the height of water on deck which can be calculated
from the exact estimation of relative motion. In earlier
studies done by Fonseca and Guedes Soares (2004),
it was shown that the real relative motion calculated
using the numerical model is less than experimental
data because of additional factors like the bow wave
and dynamic swell up at the bow. It was shown in the
study that when the bow waves are separated from
the experiment, the numerical results got closer to the
experiment. Buchner et al. (1995) was successful in
predicting the green water forces on the deck when
226
on left hand side and input wave at midhsip on right hand side) with the experimental data for Froude numbers 0.12 and 0.18.
Figure 5. Comparison of vertical bending moment at midship from linear and nonlinear numerical results with the experimental data for a range of Froude numbers. Numerical on x axis stands for the Froude number and Alphabets for the target
position of the input wave (F stands for FP and M stands for Midship).
227
Figure 6. Comparison of the vertical bending moment at Midship (wave at Midship on the left and wave at FP on the right)
between the nonlinear numerical results (with and without amplification) and the experimental data for 0.06 Froude number.
exact water height on deck from experiment was used

for the calculation.
Keeping these points in mind, the following section
uses a heuristic approach to modify the relative motion
at the forward perpendicular using an amplification
factor and checks the effect of the green water height
on deck on the vertical bending moment of the ship,
thus trying to find out the reason behind the sagging
peak cutoff when the bow is immersing down into the
water.
4.2
Effect of green water on deck model
The nonlinear numerical code was modified to include

the amplification of the relative motion at the bow so
as to check the effect of the green water height on deck.
The relative motion at the bow is amplified when the
relative motion at bow is larger than the freeboard,
which means the presence of the green water on deck.
This is done in a systematic order for different sea
states and Froude numbers with 25% and 50% relative
motion amplification to investigate the effect. Figures 67 show comparisons between the experimental
and numerical values of the bending moment, without any amplification and with amplification of 25%
and 50%.
Fig. 6 shows comparison of the bending moment
between the experiment and the numerical results
without amplification and with 25% and 50% amplification. For the Froude number 0.06 with the abnormal
wave crests at FP, between 1180 and 1190 s, the relative
motion exceeds the freeboard height hence resulting
in shipping of the greenwater. The numerical code was
modified to amplify the aforementioned green water
height by amplifying the relative motion at the bow.
Such amplification resulted in reduction of the corresponding numerical bending moment bringing it closer
to the experimental value. Discrepancy between the
numerical and the experimental results were reduced
again as the amplification factor was increased from
25% to 50%. Such trends were repeated when the

wave was at midship as shown in Fig. 6.
Similar modification in the relative motion as mentioned above was done for higher Froude numbers
0.12 and 0.18 as shown in Fig. 7. A 50% relative motion amplification for Froude number 0.18,
underestimated the sagging moment while agreement
was good with the sagging peak value. It should be
noted that the adopted method for the amplification of
relative motion was highly intuitional and gave only
an outline of the effect of green water shipping on
the bending moment. The real relative motion could
be lying somewhere between the assumed amplification factors and moreover, amplification was done
only between the time periods where the numerical
result predicts shipping of the green water. In fact, the
green water shipping in the real condition might be
also happening beyond these time range.
The green water on deck reduces the sagging
moment by applying a downward vertical force on the
deck which imposes a counteracting hogging moment.
This is one of the reasons behind such a large reduction in the sagging peak value but not the only one.
There may be other reasons such as small amplitude wave and response motion assumptions of the
strip theory, the neglected surge force acting along the
longitudinal plane hence reducing the sagging peak,
resulting from the slender ship assumption, disregard
of bow wave effect resulting from the three dimensional flow, neglect of viscous effect in high seas and
three dimensional added mass and damping etc.
But as discussed earlier, the green water shipping is
found to be a major phenomenon affecting the vertical
bending moment and an exact calculation of relative
motion is very much essential for accurate prediction
of loads in high seas.
Fig. 8 compares the numerical sagging peaks after
applying 25% and 50% amplification in the relative
motion at the bow during the occurrence of the green
water with the experimental data. Similar to Fig. 5, the
228
Figure 7. Comparison of the vertical bending moment at Midship (wave at FP and Midship) between the nonlinear numerical
results (with and without amplification) and the experimental data for 0.12 and 0.18 Froude number.
Figure 8. Comparison of vertical bending moment at midship between numerical results with 25% and 50% relative motion
amplifications with the experimental data for a range of Froude numbers. Numerical on x axis stands for the Froude number
and Alphabets for the target position of the input wave (F stands for FP and M stands for Midship).
229
largest sagging peak from each result was compared.

It was clearly seen that as the amplification factor was
methodically increased from 0, as given in Fig. 5, to 25
and 50%, as given in Fig.10, the difference between the
experimental and the numerical largest sagging peak
values systematically decreased, which is an obvious
indication of the effect of green water on the vertical
bending moment.
5
CONCLUSIONS
A study has been done on the behavior of a container

ship in extreme sea conditions and the resulting vertical bending moments were analyzed. Extreme sea
conditions were simulated in the tank using a deterministic wave profile with a 3 Sisters freak wave
pattern embedded in irregular sea state. The experimental response motion and the bending moment
was recorded and compared with the numerical results
based on strip theory. A numerical code based on both
linear and nonlinear methods was used to calculate and
compare the response. Nonlinearity was introduced in
the code by including nonlinear Froude-Krylov force
and nonlinear hydrostatic force.
The linear method overestimated the hogging peak
value but underestimated the largest experimental
sagging peak value. The nonlinear peak values were in
good agreement with the experimental hogging peaks
while highly overestimated the largest sagging peak
value. The largest experimental sagging peak value
was found to be cut off which lead to further study on
the analysis of the relative motion at bow and the effect
of green water shipping.
An intuitive approach on the relative motion amplification at the bow during the instance of green water
shipping was used to study the effect of green water
on the vertical bending moment. A systematic application of amplification factor of 25% and 50% on the
relative motion gave better results reducing the difference between the largest numerical and experimental
sagging peak. The approach although did not provide a
definite solution to the problem but gave a clear insight
and suggested the direction for further studies into the
influence of green water on the sagging moment.
ACKNOWLEDGEMENTS
project HANDLING WAVES Decision Support
System for Ship Operation in Rough Weather,
(www.mar.ist.utl.pt/handling waves/), which is partially funded by the European Union through the
Growth program under contract TST5-CT-2006031489.
REFERENCES
Beck, R.F., 1994. Time-domain Computation of Floating
Bodies, Applied Ocean Research 16, 267282.
Buchner, B., 1995. The Impact of Green Water on FPSO
Design, Offshore Technology Conference 4547.
Clauss, G.F. & Schmittner, C.E., 2005. Experimental Optimization of Extreme Wave Sequences for the Deterministic Analysis of Wave/Structure Interaction. In OMAE
2005-24th International Conference on Offshore Mechanics and Arctic Engineering. Paper: OMAE 2005-67049.
Clauss, G.F., Kauffeldt, A. & Klein, M., 2009. Systematic
Investigation of Loads and Motions of a Bulk Carrier
in Extreme Seas. In: Proceeding of the ASME 2009 28th
International Conference on Ocean, Offshore and Arctic
Engineering. Paper: OMAE 2009-79389
Clauss, G.F., Klein, M. & Dudek, M., 2010. Influence of
the Bow Shape on Loads in High and Steep Waves. In:
Proceeding of the ASME 2010 29th International Conference on Ocean, Offshore and Arctic Engineering. Paper:
OMAE2010-20142
Fonseca, N. & Guedes Soares, C., 1998. Time Domain Analysis of Large Amplitude Vertical Motions and Wave Loads,
Journal of Ship Research, Vol 42, No. 2, pp. 100113
Fonseca, N., Guedes Soares, C. & Pascoal, R., 2001. Prediction of Wave Induced Loads in Ships in Heavy Weather,
Proceedings of the International Conference on Design
and Operation for Abnormal Conditions II, 67 November
2001.
Fonseca, N. & Guedes Soares, C., 2004, Green Water Effects
on the Bow of a Containership Advancing in Regular and
Irregular Waves, The 9th Int. Symposium on Practical
Design of Ships and Mobile Units (PRADS04), LuebeckTravemuende, September 2004, pp. 412419.
Fonseca, N. & Guedes Soares, C., 2005, Experimental
Investigation of the Shipping of Water on the Bow of
a Containership, Journal of Offshore Mechanics and
Arctic Engineering, Vol. 127, No. 4, 2005, pp. 322330.
Guedes Soares, C., Fonseca, N., Pascoal, R., Clauss, G.F.,
Schmittner, C.E. & Hennig, J., 2006. Analysis of Design
Wave Loads on a FPSO Accounting for Abnormal Waves.
Journal of Offshore Mechanics and Arctic Engineering.
128(3):241247.
Guedes Soares, C., Fonseca, N. & Pascoal, R., 2008. Abnormal Wave Induced Load Effects in Ship Structures.
Journal of Ship Research. 52(1):3044.
Lloyd, R.J.M., Salsich, J.O. & Zseleczky, J.J., 1985. The Effect
of Bow Shape on Deck Wetness in Head Seas, Trans.
RINA, London.
ODea, J.F. & Walden, D.A., 1984. Effect of bow shape
and Nonlinearities on the Prediction of Large Amplitude
Motions and Deck Wetness, Proc. 15th ONR Symp on
Naval Hydrodyamics, Hamburg.
Salvensen, N.,Tuck, E.O. & Faltinsen, O., 1970. Ship Motions
and Sea Loads. Trans.SNAME, 78, 250287.
Subramani, A., Beck, R.F. & Scorpio, S., 1998. Fully Nonlinear Free Surface Computations for Arbitrary and Complex
Forms. In: Proceeding of the 22nd ONR Symposium on
Naval Hydrodynamics, Washington, DC, pp. 4758.
230

Dynamic analysis of a purse seiner and a trawler fishing vessels in the

Portuguese coastal zone
M. Tello, S. Ribeiro e Silva & C. Guedes Soares
ABSTRACT: A study of the short term seakeeping performance of a set of fishing vessels is carried out.
Assuming a fully developed sea state, linear transfer functions of hullforms are assessed using a standard strip
theory method. Also from short term spectral responses, conditions at which they must operate to avoid hazards
and seasickness can be readily identified. Each fishing vessel is studied over sea state 5 to 6 and has been
evaluated at different Froude numbers and headings. The seakeeping performance is therefore determined by the
limit of the couple: speed and heading angle, at which the vessel standard operational conditions for the most
relevant performance criteria, related with the absolute motions, relative motions, accelerations, slamming, and
green water on deck, are guaranteed.
1
INTRODUCTION
Fishing vessels are subjected to different loads due

to the harsh environment where they operate. Those
loads can drive hazard on board, seasickness, and in
some cases the loss of the ship; as the tragedy of three
Spanish fishing vessels studied by Perez-Rojas et al
(2006), where loss of lives and the ships occurred.
Indeed, many fishing vessels accidents have been
reported during the last years and the most representative case has been reported by Mantari et al (2011),
where around fourteen Peruvian fishing vessels, capsized between the years 2008 to 2009. In addition,
other occurrences have been also presented in the statistical report by Wang et al (2005), where a breakdown
of the accidents recorded in 90s is: machinery damage,
with 65.97%, followed by foundering, with 15.41%
and flooding and grounding with 8.38%.
However, many of the problems which the ship
can experience can be avoided simply by relocating
the crews working places in the design stage, and
also varying the ship position with respect to waves.
This has been discussed in a previous work by Tello
et al (2011), where for a set of fishing vessels the
importance of the location of the centre of gravity, the
location for the working places and the design parameters as local freeboard, design water line, etc were
pointed out. Moreover, in Tello et al (2011), polar diagrams were also plotted, which can be useful for the
master in order to change the disposition of the ship
aiming to decrease the responses which are exceeding
their respective thresholds.
The same approach developed in Tello et al (2011)
has been used in the present work however the analysis now is focused on the study of two particular
fishing vessels: a purse seiner (FV3) and a trawler
(FV8). These particular ships are of interest due to

their smaller displacements (see Table 1). In Table 1
can be seen that the displacements of FV3 and FV8
are around half of their similar size FV2 and FV6
respectively. In addition, this analysis has been made
in the light loading condition, where the ship is operating most of the time, and in which it is subjected
to harsh responses while meeting waves. This analysis
also considers a long term statistical analysis for both
ships and focused in the short term statistics analysis
for more sea states than the ones chosen in the previous
investigations.
Finally, it is important to mention that the method
briefly described in the following section was utilized
earlier by Guedes Soares et al (1995) to investigate
the seakeeping performance of a group of fishing vessels operating in the Portuguese coast. Fonseca and
Guedes Soares et al (2002) applied the same method
to investigate the sensitivity of the expected ships
availability to different seakeeping criteria.
2
METHODOLOGY OF THE SEAKEEPING

PERFORMANCE ASSESSMENT
To analyze the seakeeping performance of fishing vessels, transfer functions (TF) of the rigid body are
obtained first. To estimate the TF, a seakeeping code
based on the strip theory formulation from Salvesen
et al (1970) has been used. The code assumes potential flow to calculate the hydrodynamic coefficients
and harmonic wave exciting forces. To apply strip
considerations as the ship is slender, small exciting
wave amplitudes and oscillatory motions and low
Froude numbers are required. Therefore the analysis of the ship will approach to the real case but it
231
is not absolutely true for ships with bulky hullforms

and appendages. Thus, the roll transfer function is corrected for viscous damping effects. An example of this
is presented in the study of Ribeiro e Silva et al (2006)
where the variations of roll motion RAO from a bare
hull configuration are much larger for hard-chinned
hullforms and ships fitted with bilge keels.
In the present study viscous effects have been estimated by the formulation proposed by Ikeda et al
(1978), resulting in a mean damping factor around
0.12 for all ships. However, it should be noted that
this formulation is not especially conceived for small
fishing vessels.
Moreover, the interactions between fishing gear
components with the vessel during haul and towing
operations, or water-on-deck effects have been all
ignored so that fishing vessels are assumed as freefloating bodies. Thus, rigid body oscillatory motions
are calculated by solving a set of coupled linear
equations, which represent the equilibrium:
These excitation forces Fk and ship motions j can

be represented on a right handed cartesian coordinate
system, X = (x, y, z), fixed with respect to the mean
position of the ship and the origin in the plane of the
undisturbed free surface, as shown in Figure 1. The
translatory displacements in x, y, and z directions are
respectively surge 1 , sway 2 , and heave 3 , while the
rotational displacements about the same axis are roll
4 , pitch 5 , and yaw 6 , where subscripts, Ckj indicate
forces in the k-direction due to motions in the j-mode,
Mkj are the components of the mass matrix for the ship,
Akj and Bkj and Bkj are the added mass and damping
coefficients, Ckj are the hydrostatic restoring coefficients and Akj and Bkj are the complex amplitudes of
the exciting forces.
The harmonic j-th response of the vessel, j will be
proportional to the amplitude of the exciting force, at
the same frequency but with phase shift, j and is then
given by:
obtained by using equation (5). Thus, the 6 DoF modes

and translational
are divided into the rotational
mode.
R
Hence, the components of the absolute displacement

vector = (XP , YP , ZP ) at any point P are:
Restricting the analysis to the ships centre plane

(yp = 0), the vertical displacement and acceleration
at the point P, can be represented by:
The lateral displacement and acceleration are similar

to the vertical acceleration, however as described by
Journe and Massie (2001), another component due
to gravity, g, should be added. Thus, the total lateral
acceleration that the vessel feels is given by:
where: YP (t, e ) = Ypa cos (e t + p )

The relative vertical displacements and velocity, is
given by: the absolute motion and wave amplitude,
P (t, e ), at the point P. This is:
Equation (10) may also be re-written as:
2.2 Irregular waves

When the ship travels at a speed U with an angle
with the direction of incoming waves (Fig. 1), then the
encounter waves and the free surface are:
2.1
To assess the response in irregular sea states, a spectral

linear formulation is adopted, following the proposals of St. Dennis and Pierson (1953). The Pierson
Moskowitz spectral form for fully developed seas
(Pierson and Moskowitz, 1964) describes the irregular sea in terms of significant wave height HS and
peak period TP , and this is given by:
Motions and derived responses
The seakeeping criteria are defined in terms of absolute motions, relative motions and accelerations, thus
these quantities are calculated for relevant points on
the ship. For small motion amplitudes, the displacement vector at any point with coordinates r can be
where N is the number of component waves, n the

circular frequency, n the random phase angle and wa n
232
Figure 1. The coordinate system and six modes of motion, and definition of the ship heading angle.
the amplitude of the n-th component wave. Because the

system is linear, the relationship of the j-th response
is given by:
well as for propeller emergence it is 1/4 of the propeller

diameter. According to Ochi (1964), slam occurs when
the relative motion is larger than the mean draught (T )
at the bow and at the same time the relative
vertical

velocity has to exceed a value of 0.093 Lpp g. This is
given by:
where Hj () is the TF from wave elevation to the j-th

mode. The variance of a record is given by:
3
Because the responses are linear, the same Gaussian
model, describes the response process. This implies
that a Rayleigh distribution describes the amplitudes
or the peaks of the processes, according to which the
probability of exceeding the level r is given by:
From the assumption of the Rayleigh distribution

(Longuet-Higgins, 1952), the average of the one-third
larger amplitudes, rS or the most probable maximum
value in N successive cycles can be obtained from (16),
they are given by:
2.3
Seakeeping performance criteria
If the criterion is defined as a probability of exceeding

a critical value pCR , the related root mean square of
the response is obtained by:
where rmax is the limiting magnitude of the response

which has the probability pCR of being exceeded. For
instance, in the occurrence of green water on deck,
rmax is the value representing the local freeboard, as
FISHING VESSELS PARTICULARS
The present analysis considers a set of Portuguese

fishing vessels, with distinct hullforms and main
dimensions, corresponding to the light loading conditions where the fishing vessels are at departure for
fishing having 100% consumables onboard.
The main particulars are presented in Table 1 and
the plan views are showed in Figure 2. The rest of
body plans of ships are omitted in here since they can
be found in Tello et al (2011).
SEAKEEPING CASE STUDY
In general, mission effectiveness will be degraded

with increasing amplitude of motions and accelerations in a seaway. From a seakeeping perspective, the
probability of success of the vessel operation can be
assessed by evaluating the effects of ship motions on
the subsystems involved.
In order to evaluate if a hull design is successful, criteria are first established against which an analysis can
then be made. Moreover, those criteria can be divided
as motion responses and derived responses criteria.
The motion response criteria are those formulated
on the basis of the ship motions. These include: the
root mean square (RMS) for roll and pitch angles, the
RMS for the vertical and lateral displacements, and
the RMS for vertical velocities. On the other hand,
derived response criteria are those formulated from
the basis of direct responses of the linear frequency
domain computations. These include: the propeller
emergence index, the slamming index, and the wetness
index (ITTC, 2005).
233
Figure 2. Body plans of fishing vessels FV2, FV3, FV6 and FV8.
Table 1.
Main particulars of the fishing vessels.
FV
Lpp
(m)
B
(m)
D
(m)
T
(m)
Cb

(tonnes)
LCG
(m)
Kx/B
Ky/Lpp
Kz/Lpp
GMt (m)
Cond.
FV1
FV2
FV3
FV4
FV5
FV6
FV7
FV8
FV9
FV10
FV11
FV12
FV13
13.4
15.0
15.9
17.8
20.0
20.1
21.4
22.0
22.1
24.5
29.0
37.4
45.7
4.7
5.4
5.6
6.0
7.4
6.1
6.7
5.8
6.9
8.0
8.0
7.3
10.1
2.2
2.6
2.7
3.0
5.5
3.3
3.4
2.4
3.4
3.5
3.7
3.5
5.3
1.4
2.1
1.8
2.0
3.0
2.5
2.5
1.6
2.5
3.1
3.3
3.0
5.0
0.55
0.62
0.35
0.57
0.64
0.43
0.45
0.48
0.46
0.63
0.62
0.67
0.55
49.00
104.88
55.49
122.90
285.71
129.50
162.60
94.70
170.30
382.70
470.90
552.80
1258.00
0.91
1.92
0.85
1.90
1.84
1.35
0.50
1.40
0.16
0.00
1.28
4.46
0.05
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.67
1.10
2.01
0.88
0.79
1.14
0.50
0.79
0.50
0.60
0.35
1.13
1.28
Light
Light
Light
Light
Light
Full
Light
Light
Light
Full
Full
In the case of fishing vessels, apart from the work

of Sarioz and Narli (2005), Fonseca et al (2002) and
Odabasi et al (1991), there are apparently not many
prescribed values available in the literature. Hence, as
used by Fonseca et al (2002), a value of 0.2g for the vertical acceleration criterion, at the bridge and working
deck locations, have been adopted in this study. Table 2
presents the chosen points with their respective criteria
for a set of thirteen fishing vessels.
The following discussions will be focused only

in the results of FV2, FV3, FV6 and FV8. As
mentioned before, similar results for all the other
ships can be found in Tello et al (2011). In this
study, mainly FV3 and FV8 are thoroughly compared
and contrasted to FV2 and FV6. The locations are
described using the reference coordinates shown in
Figure 1. While the criteria considered are presented
in Table 3.
234
Table 2.
Locations of relative points adopted for the assessment of derivate responses of each fishing vessel.
FV
WD
GW
Slam.
PE
FV1
FV2
FV3
FV4
FV5
FV6
FV7
FV8
FV9
FV10
FV11
FV12
FV13
(3.10, 0, 2.31)
(3.18, 0, 2.40)
(7.34, 0, 4.19)
(0.05, 0, 3.10)
(6.00, 0, 3.30)
(5.30, 0, 2.50)
(5.10, 0, 2.60)
(3.15, 0, 2.63)
(6.60, 0, 3.15)
(6.13, 0, 1.60)
(7.50, 0, 2.30)
(9.82, 0, 1.45)
(####, 0, 1.37)
(4.20, 0, 1.40)
(6.36, 0, 1.60)
(0.00, 0, 2.36)
(6.92, 0, 1.48)
(2.50, 0, 2.50)
(4.26, 0, 1.94)
(4.66, 0, 2.00)
(6.60, 0, 3.75)
(2.70, 0, 2.45)
(4.90, 0, 0.50)
(5.50, 0, 1.30)
(7.85, 0, 0.49)
(9.20, 0, 0.41)
(6.60, 0, 3.12)
(7.15, 0, 2.40)
(8.31, 0, 3.41)
(8.90, 0, 3.30)
(10.00, 0, 3.50)
(10.64, 0, 3.12)
(10.50, 0, 3.10)
(12.28, 0, 4.25)
(11.00, 0, 3.70)
(12.25, 0, 2.80)
(13.80, 0, 3.50)
(19.64, 0, 3.04)
(22.90, 0, 3.43)
(6.00, 0, 2.20)
(6.60, 0, 2.50)
(7.34, 0, 1.57)
(7.90, 0, 2.37)
(8.50, 0, 3.00)
(9.36, 0, 2.25)
(8.70, 0, 3.10)
(11.15, 0, 1.50)
(9.10, 0, 3.15)
(11.14, 0, 2.93)
(12.50, 0, 3.50)
(15.54, 0, 2.64)
(20.12, 0, 4.78)
(6.80, 0, 1.30)
(7.15, 0, 1.35)
(6.81, 0, 0.63)
(8.40, 0, 1.50)
(8.50, 0, 1.55)
(8.64, 0, 1.41)
(8.70, 0, 1.50)
(8.85, 0, 0.43)
(8.80, 0, 1.80)
(9.36, 0, 1.85)
(10.50, 0, 2.50)
(16.25, 0, 1.67)
(18.55, 0, 2.74)
Table 3. Prescribed criteria for the assessment of fishing

vessels seakeeping performance.
Criterion
Green Water Deck
Slamming
Propeller. Emergence
VA at Bridge
LA. at Bridge
VA at Work. Deck
LA at Work. Deck
Roll
Pitch
Prescribed value
(GW)
(Slam.)
(PE)
(VAB)
(LAB)
(VAWD)
(LAWD)
()
()
5%
3%
15%
0.2 g
0.1 g
0.2 g
0.1 g
6
3
(prob)
(prob)
(prob)
(rms)
(rms)
(rms)
(rms)
(rms)
(rms)
SEAKEEPING RESULTS AND DISCUSIONS
The analysis of the fishing vessels were made considering the Portuguese costal zone, corresponding to area
16 of Global Wave Statistics data base. These wave climate statistics used are based on visual observations
(see Hogben, da Cunha and Olliver, 1986), and the vessel is expected to operate on this specific area during
one year. From the scatter diagram, a mean significant
wave height and zero crossing wave period of Hs =
3.24 [m] and Tz = 8.42 [s] were obtained.
The results of the estimated operability indexes
per year of FV2, FV3, FV6 and FV8 are presented
in Table 4. These values mean that the ship is fully
operational (satisfies all seakeeping criteria) during a
certain percentage of the time during the year. Moreover, values have been organized for the set of chosen
criteria, and grouped in two main columns related to
the vessels Froude numbers adopted, Fn = 0.0 and
Fn = 0.3.
It can be seen is Table 4 that the operability indexes
of FV3 and FV8 are much lower than FV2 and FV6.
This is mainly due to operational restrictions imposed
by propeller emergence criterion (PE), which determines the lowest operability indexes from the set of
criteria adopted for both fishing vessels.
If a preliminary comparison is made between the
hull geometries of FV2 and FV3 relatively to FV6 and
FV8, it can be seen that the main particulars, length

and beam of the vessels do not present a significant
difference. However, the draughts of the first pair of
ships are quite distinct from the second pair of vessels, as well as the relative position of the propeller
(see Table 2). In fact, the vertical distances from the
waterline to the point in the propeller under analysis
are larger for the pair FV2 and FV6 comparatively
to the pair FV3 and FV8, and this is a reason which
contributes the most to the lowest operability index
associated with failure to comply with PE criterion for
both fishing vessels.
Other criteria which also contribute significantly
to the reduction of the operability indexes, are lateral
accelerations at the bridge and at the working deck
(LAWD), and which are closely related to the low
operability index experienced in pure rolling. However, those criteria do not seem to be as much critical
as PE.
This probabilistic analysis of the vessel responses
in waves over one year at the design point is very
important to assess the seakeeping performance in
general, but it is important to bear in mind that in
here the procedure has been simplified by neglecting
the wave directionality on the wave climate statistics.
Additionally it has been assumed that all ship headings relatively to waves are equally likely. Moreover,
it should be stressed that although the operability in
pitch and heave motion appears to stand normal operation conditions, they are no guarantee of prevention
from damaged equipments, seasickness and hazards
reported in accidents of fishing vessels. Because these
problems usually depend on specific features onboard
fishing vessels, that in practice cannot be captured by
the simplistic criteria adopted.
However, in order to better understand the dependence of the seakeeping performance of fishing vessels over specific operational scenarios, it was decided
to carry out a more detailed analysis, first, for different
combinations of significant wave heights, Hs, and zero
up crossing periods, Tz, and the results are plotted in
Figures 3 and 4. Figure 3, present the results of the analysis of fishing vessel FV3, for wave encounter heading
235
Table 4. Seakeeping criteria applied to FV2, FV3, FV6 and FV8 at Fn = 0.0 and 0.30. Coastal area 16 (from Global wave statistics).
236
Downloaded by [Florida International University] at 05:58 11 May 2014
At Froude 0.00
At Froude 0.30
Roll
180
150
120
90
60
30
0
FV2
180
150
120
90
60
30
0
1.00
0.67
1.00
1.00
1.00
0.72
1.00
0.94
0.95
0.88
1.00
0.61
0.94
1.00
0.99
0.52
0.94
0.99
1.00
0.61
0.99
0.87
1.00
0.89
1.00
0.81
1.00
1.00
Operability per year = 0.6582
1.00
0.89
1.00
1.00
1.00
0.91
1.00
0.87
0.94
0.96
1.00
0.45
0.88
0.98
1.00
0.45
0.94
0.94
1.00
0.58
1.00
0.87
1.00
0.94
1.00
0.84
1.00
1.00
1.00
0.70
1.00
1.00
0.87
0.75
1.00
0.60
0.44
0.94
0.99
0.15
0.28
1.00
0.99
0.14
0.44
0.97
0.99
0.15
0.87
0.80
1.00
0.60
1.00
0.73
1.00
1.00
1.00
0.92
1.00
1.00
0.88
0.94
1.00
0.74
0.46
0.98
1.00
0.34
0.34
0.99
1.00
0.23
0.46
0.96
1.00
0.34
0.88
0.90
1.00
0.80
1.00
0.87
1.00
1.00
FV3
180
150
120
90
60
30
0
FV6
180
150
120
90
60
30
0
FV8
Pitch
VAB
LAB
VAWD
LAWD
GW
PE
Slam.
Op
Roll
Pitch
VAB
LAB
0.95
0.95
0.96
0.98
0.98
0.97
0.97
1.00
0.96
0.72
0.72
0.72
0.94
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.88
0.94
0.99
0.99
0.96
0.76
0.72
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.67
0.72
0.61
0.52
0.61
0.76
0.72
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.95
0.69
0.53
0.60
0.94
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.27
0.27
0.27
0.27
0.27
0.27
0.27
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.27
0.27
0.27
0.27
0.27
0.27
0.27
1.00
1.00
0.99
0.99
0.99
1.00
1.00
1.00
0.67
0.23
0.14
0.23
0.61
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.98
0.99
1.00
1.00
1.00
0.97
0.95
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.70
0.60
0.15
0.14
0.15
0.60
0.73
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.78
0.34
0.21
0.23
0.69
1.00
0.98
0.99
1.00
1.00
1.00
0.98
0.98
0.26
0.26
0.27
0.26
0.27
0.27
0.27
0.87
0.91
1.00
1.00
0.98
0.87
0.87
0.26
0.26
0.27
0.21
0.21
0.27
0.27
1.00
0.44
0.83
1.00
1.00
0.49
0.87
1.00
1.00
0.75
0.94
0.95
1.00
1.00
0.99
0.89
1.00
0.98
1.00
0.97
1.00
0.94
1.00
1.00
1.00
0.91
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
0.87
0.96
1.00
1.00
0.46
0.94
1.00
1.00
0.46
0.97
1.00
1.00
0.70
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
1.00
0.58
0.60
1.00
0.99
0.60
0.61
0.88
0.86
0.87
0.80
0.46
0.84
1.00
0.99
0.45
0.94
0.98
1.00
0.70
1.00
0.88
1.00
0.99
1.00
0.84
1.00
1.00
1.00
0.97
0.95
1.00
0.97
0.97
0.96
0.90
0.79
0.99
0.98
0.56
0.70
1.00
1.00
0.43
0.85
0.99
1.00
0.61
0.99
0.98
1.00
0.94
1.00
0.97
1.00
1.00
VAWD
LAWD
GW
PE
Slam.
Op
0.52
0.61
0.76
0.98
1.00
1.00
1.00
1.00
1.00
0.98
0.98
0.99
1.00
1.00
0.91
0.95
1.00
1.00
1.00
1.00
1.00
0.67
0.72
0.87
1.00
1.00
1.00
0.98
0.91
0.96
1.00
1.00
1.00
1.00
1.00
0.44
0.49
0.75
0.89
0.97
0.94
0.91
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.94
0.69
0.58
0.74
0.98
1.00
0.92
0.92
0.92
0.92
0.92
0.85
0.85
0.27
0.27
0.27
0.27
0.27
0.27
0.27
0.85
0.85
0.86
0.87
0.89
0.89
0.89
0.27
0.27
0.27
0.27
0.27
0.27
0.27
0.75
0.80
0.88
0.99
1.00
1.00
1.00
1.00
0.94
0.61
0.46
0.70
0.99
1.00
0.87
0.94
1.00
1.00
1.00
1.00
1.00
0.92
0.94
0.99
1.00
1.00
1.00
0.99
0.75
0.87
0.99
1.00
1.00
1.00
1.00
0.58
0.60
0.46
0.45
0.70
0.88
0.84
0.95
0.96
0.98
1.00
1.00
1.00
1.00
1.00
0.90
0.56
0.43
0.60
0.93
1.00
0.97
0.98
1.00
1.00
0.99
0.94
0.92
0.21
0.21
0.21
0.21
0.21
0.21
0.21
0.67
0.75
0.97
1.00
0.94
0.94
0.94
0.21
0.21
0.21
0.21
0.21
0.21
0.21
Figure 3. Significant wave height for 90 , 150 and 180 for r all criteria at Fn = 0.0 and Fn = 0.3, for FV3.
Figure 4. Significant wave height for 90 , 150 and 180 for r all criteria at Fn = 0.0 and Fn = 0.3, for FV8.
angles of 90 , 150 and 180 , taking into account

all the criteria and for Froude numbers Fn = 0 and
Fn = 0.3. Figure 4 does present the similar results from
the evaluation made for fishing vessel FV8, where the
plots are presented as well for 90 , 150 and 180 and
two distinct Froude numbers: Fn = 0 and Fn = 0.3.
From Figures 3 and 4, it can be seen that the lowest
operability curves are those corresponding to PE criterion for both fishing vessels. Moreover, the PE curves
seem to be fairly constant for FV3 and FV8 even when
the ships speed is varied. In case of FV8, the operability is in fact limited by a larger set of criteria whose
respective thresholds are quite often exceeded, and this
feature explains why FV3 has a larger operability index
relatively to FV8. Therefore, the immersed hull geometry of FV8 should be modified in order to improve
its seakeeping performance. However, it should be also
mentioned that this course of action could be applied as
well to FV3 if a proper seagoing fishing vessel would
be seen as desirable. Notice, although FV3 presents an
enhanced performance, established prescribed values
are easily exceeded for significant wave heights above

2 meters.
The critical analysis about fishing vessels FV3 and
FV8 have been made based on the results shown in Figures 3 and 4, which correspond to the vessel responses
over a broad range of sea states, but only for two distinct ships speeds and three wave encounter heading
angles. However, these vessels are supposed to operate over a large variety of sea states, meeting waves
from any direction. Thus, in order to determine which
criteria will in practice delimit most of the vessel operability, an analysis over different sea states combined
with different heading angles and a variety of ship
speeds has been also made.
In order to find out more easily the criteria which
impair the operability of fishing vessels FV3 and FV8,
it was decided to present the failure criteria results in a
polar plot format (see Figure 5). Figure 5 presents the
analysis distributed in a grouped set of criteria which
are indicated at the bottom of the figure. This study is
carried out over a set of sea states which are specified in
237
Figure 5. Polar speed plot for fishing vessels FV3 and FV8 operating in sea states 5 and 6.
each polar plot. Those sea states correspond to those

with the highest probability of occurrence retrieved
from the Global Wave Statistics database for area 16
(Portuguese coastal zone).
In Figure 5, a polar speed plot for fishing vessels FV3 and FV8 operating in sea states 5 and 6 is
presented, the top row of polar plots belongs to FV3
while the bottom row corresponds to FV8, the columns
are aligned to a determinate seastate. The polar plots
are presented for the criteria shown at the caption
field which have exceeded their prescribed values,
therefore, omitting the ones where no occurrence is
experienced by the fishing vessels. In the polar plots
the radius is related to the ships speed in terms of
Froude numbers, while the angles are related to the
direction of the incoming waves. Moreover, the heading angle is defined in accordance to Figure 1, such
as following, beam and head waves correspond to 0 ,
90 and 180 , respectively.
In Figure 5, it can be observed that when FV3 is
operating in moderate sea states, Hs = 2.5 [m] and
Tz = 7.5 [s] and Tz 8.5 [s], only roll and pitch criteria overcome their respective thresholds. Nevertheless,
when the vessel meets the same significant wave
amplitude but with larger zero up crossing periods,
the operability is improved. This is the case of roll
in the second sea state (Hs = 2.5 [m], Tz = 8.5 [s])
where operability is improved and this is shown as
less area covered in the polar plot. Moreover, when
fishing vessel FV3 meets harsher environmental conditions, others criteria are also exceeded thus delimiting more the ship operability. This is observed for
FV3 when it operates in a sea state, Hs = 5.5 [m] and
Tz = 8.5 [s], where at such environmental conditions
all the responses in lateral accelerations at the working
deck and at the bridge, green water on deck, propeller

emergence and slamming overcome their thresholds
established in Table 3.
In the case of FV8, it experiences roll, pitch and propeller emergence for moderate sea states, Hs = 2.5 [m]
and Tz = 7.5 [s], and behaves differently from FV3.
Actually, FV8 operability is decreased when the zero
up crossing wave period changes from 7.5 [s] to 8.5 [s].
Which is due to the fact that propeller emergence
exceeds its prescribed value for more couple of heading angles and ships speed, consequently covering a
large area than in the previous sea state. On the other
hand, the operability of FV8 is better than FV3 when
operating in rough waves. This can be observed in Figure 5 for columns five to night, where FV3 presents
larger areas of the polar plots covered by the exceeding criteria than the ones experienced by FV8. As an
example, it can be observed the occurrence of propeller
emergence, which is plotted in the sixth column.
Other differences can also be observed between
FV3 and FV8, mainly in roll motion, where FV3
has better responses in waves than those experienced
by FV8. The reason for the variation in the roll
response is due to the difference in the transversal
metacentric heights, GM t , which is larger for FV3
(GMtFV3 = 2.01 > GMtFV8 = 0.79).
Hence, as suggested before a new loading condition could be disposed for FV8 in order to decrease,
for example, the emergence of the propeller and thus
avoiding engine overloading, while in the case of
FV3, the operability over moderate sea states are
not dramatic. However, it should be also noticed that
FV3 is a really small vessel, and, therefore, it may
not be considered a seagoing vessel so that in practice
FV3 will mostly operate over moderate sea states.
238
CONCLUSIONS
A methodology to predict the ship responses in waves

and related seakeeping performance is first presented
in here. The procedure requires the calculation of
transfer functions for different speeds and headings.
These transfer functions for a certain loading condition
are then combined with the wave spectral formulation
of the design point of the vessel. Finally, by defining adequate criteria the seakeeping performance can
be inferred, and, using long term statistics, the ship
operability index can be estimated.
This study presents an assessment of both the short
and long term responses aiming to identify the seakeeping criteria which limit the most the operability
of two particular fishing vessels when they operate in
different sea states and during one year at the design
point.
In the short term analysis, sea states 5 and 6 are
critical, and, therefore, were thoroughly studied. From
the analysis, it can be concluded that roll and pitch
criteria are the most often exceeded ones for the two
relatively small fishing vessels analysed. Thus in the
case of roll motion it turns very important to take into
account the transversal metacentric height, GMt. In the
analysis it was concluded that enhanced operability
would be achieved for larger values of GMt. In fact,
larger GMt values lead to smaller roll responses and
consequently derivate responses of roll motion, such
as lateral acceleration at the bridge and working deck.
In the case of pitch motion the operability is seen to
be severely decreased since the response experienced
exceeds quite easily the prescribed value. This seakeeping limitation is mainly due to the small draught
of both ships making easier to excite the responses
in pitch to large amplitudes. Moreover, the response
which seems to be also critical for those ships is propeller emergence, which is also a derived response of
pitch motion. Therefore, it can be concluded that the
draught of both fishing vessels should be increased in
order to enhance the overall seakeeping performance.
ACKNOWLEDGEMENTS
SADEP-Decision support system for the safety of
fishing vessels subjected to waves. The project has
been financed by the Foundation for Science and
Technology (Fundao para a Cincia e a Tecnologia), from the Portuguese Ministry of Science
and Technology, under contract PTDC/EME-MFE/
75233/2006.
Proceedings of the 21st International Conference on Offshore mechanics and Arctic Engineering OMAE2002.
Oslo, Norway, paper 28542. 2002.
Guedes Soares, C., Fonseca, N., Centeno, R., 1995. Seakeeping performance of fishing vessels in the Portuguese economic zone. Proceedings of the International Conference
on Seakeeping and Weather, RINA. London, UK, paper Nr.
12, pp. 110.
Hogben, N., da Cunha, L.F. and Olliver, H.N., 1986. Global
Wave Statistics, Brown Union London.
Ikeda, Y., Himeno, Y. and Tanaka, N., 1978. A prediction method for ship roll damping. University of Osaka
Prefecture, Re-port No 405.
ITTC, The Seakeeping Committee, 2005. Final report and
recommendations to the 25th ITTC Proceeding of the
24th ITTC volume I.
Journe and Massie, 2001. Offshore Hydrodynamics. Delft
University Technology, first edition.
Longuet-Higgins, M.S., 1952. On the statistical distribution
of the heights of sea waves. Journal of Marine Research,
Vol. 11, No. 3, pp. 245260, 1952.
Mantari, J.L., Ribeiro e Silva, S., Guedes Soares, C., 2011.
Loss of fishing vessels intact stability in longitudinal
waves International Journal of Small Craft Technology,
London Transactions RINA, Vol. 153, Part B1, pp. 2337.
Odabasi, A.Y., Fitzsimmons, P.A., Ankudinov, V.K. Wiley,
S.A., 1991. Seakeeping Considerations in Ship Design
and their Incorporation in HDDS, BMT. International,
Report no. HDDS.P2SPEC.
Ochi, M.K., 1964. Extreme behaviour of a ship in rough seasslamming and shipping of green water. Transactions of the
Society Naval Architects Marine Engineers (SNAME) 72,
143202.
Prez-Rojas, L., Arribas, F.P., Rodrguez, R.Z., Pacheco,
A.G., 2006. On the accidents of small fishing vessels. In: Proceedings STAB2006, Rio de Janeiro, Brazil,
pp. 669676.
Pierson, W.J., Moskowitz, L.A., 1964. A proposed spectral
form for fully developed wind seas based on the similarity theory of S.A. Kitaigorodskii. Journal of Geophysical
Research 69, 51815190.
Ribeiro e Silva, S., Fonseca, N., Guedes Soares, C., 2006.
Performance of a navy ship roll stabilisation system.
In: Proceedings STAB2006, Rio de Janeiro, Brazil,
pp. 669676.
Salvesen, N., Tuck, E.O., Faltinsen, O. 1970. Ship motion and
sea loads. Transactions of the Society Naval Architects
Marine Engineers (SNAME) 78, 250287.
Sarioz, K., Narli, E., 2005. Effect of criteria on seakeeping performance assessment. Ocean Engineering 32,
pp. 11611173.
St. Dennis, M., Pierson, W.J., 1953. On the motion ships
in confused seas. Transactions of the Society Naval
Architects Marine Engineers (SNAME) 61, 280354.
Tello, M., Ribeiro e Silva, S., Guedes Soares, C., 2011.
Seakeeping performance of fishing vessels in irregular
waves. Ocean Engineering, Vol. 38, pp. 763773.
Wang, J., Pillar, A., Kwon, Y.S., Wall, A.D., Loughran
Rodrguez, C.G, 2005. An analysis of fishing vessel accidents. Accident Analysis and Prevention 37, 10191024.
REFERENCES
Fonseca, N. and Guedes Soares, C., 2002. Sensitivity of
the expected availability to different seakeeping criteria.
239

Ship hydrodynamic interaction forces in restricted waters

Xueqian Zhou, Serge Sutulo & C. Guedes Soares
ABSTRACT: Based on the potential theory, a code using the panel method has been devised to estimate the
hydrodynamic interaction forces acting on ships in motion in confined waters, near a bank, in a canal or over a
dredged channel. The approach presumes distributing a source layer on the ships wetted surface and also on a
sufficiently large patch moving over the water boundaries in concern. The source distribution is discretized using
constant-density flat quadrilateral panels. Appropriate arrangement and panelling of the moving patch constitute
important part of the algorithm. The algorithm was tested by comparing numerical results and analytical results
for added masses of isolated ellipsoids in unbounded water, and by comparing the numerical and analytical
results for pressure distribution, surge and sway forces and the yaw moment acting on 1/8 of a uniformly moving
sphere. Systematic computations of hydrodynamic interaction forces in the ship-to-boundary interaction have
been carried out for a ship navigating in restricted waters with complex boundaries. Analysis of the results
demonstrates applicability of the developed method to the study of the ship interaction with complex flow
boundaries.
INTRODUCTION
The hydrodynamic interaction problem has been

attracting the interest of ship hydrodynamicists for
decades who were considering many various interaction situations: interaction between the ship and any
other moving or fixed objects, like a second ship, a pier,
a FPSO, a jetty; also interaction between the ship and
waterways with boundaries of any complexity, which
is typical for ships operating in a harbour, and sailing
over a dredged channel or in a canal. In recent years, the
latter situation has been increasingly seizing the attention as the main dimensions of certain ship types are
continuously getting larger and also the water traffic
becomes busier and busier in all harbours and canals
throughout the world.
Norrbin (1974) studied ship-bank interaction by
conducting a series of model tests of models moving
along vertical sidewalls of a dredged channel, with
a variety of offsets from the centre-line of the channel. A similar but more intensive experimental study
was done by Vantorre et al. (2003), who presented
the results in the form of regression formulae for the
sway force and the yaw moment which were found to
depend on the under-keel clearance, distance between
the ship and the bank, Froude number, propeller slipstream velocity and a few hull form parameters. Later
on, Lataire et al. (2007) carried out extensive model
tests to investigate bank effects induced by sloped
surface-piercing as well as submerged banks, Based
on Norrbins formula (1974), a new formula of the
relation between the height of the sidewall and the
magnitude of the sway force and yaw moment was

proposed.
Chng (1991, 1993) carried out bank-interaction
experiments with MarAd and S-175 models equipped
with propellers but without rudders. Lateral force and
yaw moment are presented as functions of parameters characterizing ship and waterway dimensions,
lateral distance, bottom clearance, Froude number,
and propeller thrust. It was demonstrated that the propeller thrust may have substantial influence on the
interaction sway force and yaw moment for waterdepth-to-draught ratio less than 1.5.
Laforce et al. (1996) presented experimental results
of systematic captive model tests on three models
of different lengths in open shallow water and in
restricted water. The influence of the ship length, water
depth and canal banks on hydrodynamic forces was
discussed.
Li et al. (2001) carried out model tests on bank
effects in extreme shallow water and near bank conditions with a tanker, a ferry, and a catamaran. Effects of
the ship speed, water depth, bank distance, propeller
action, bank slope, and bank submergence were investigated. In particular, the results showed that when the
ratio of water depth to draft goes to approximately
1.10, the sway force acting on a tanker changes from
a suction force to a repulsion force, and both the sway
force and the yaw moment rise sharply with decreasing
water depth.
Ship hydrodynamic interaction problems have also
been studied empirically and numerically. Beck (1977)
studied ship-to-bank interaction in a shallow channel.
241
The method of matched asymptotic expansions was

used to take advantage of the simplified governing
equations in the near and far fields. A similar method
was much later used by Kijima (1997) who studied
interaction between three bodies: two ships running
parallel in the proximity of a pier simulated with a
circular or an oval cylinder.
Varyani et al. (2003) studied the hydrodynamic
surge and sway forces and yaw moment on a moored
ship induced by another ship passing by the disturbances at the free surface were neglected. However,
the comparisons with experimental results showed
acceptable agreement for the hydrodynamic interaction force components. Based on this and other similar
studies, a generic set of approximations of interaction
sway forces and yaw moments was later proposed by
Varyani (2006).
Numerical methods have been becoming increasingly popular as they are able to simulate the hydrodynamic situations that experiments can not. Especially,
the recent progress in computer hardware technology
in the last decade made it feasible to employ numerical
methods taking into account viscosity and free surface.
However, these methods are still too heavy for realtime computation on an average computer although
satisfactory offline results have been obtained by some
hydrodynamicists, like Chen et al. (2003) and Sding
(2005).
Based on the method of Hess & Smith (1964, 1967),
Sutulo & Guedes Soares (2008, 2009) developed a fast
algorithm for computing interaction forces on arbitrary number of bodies in horizontal motion of any
direction. Free-surface effects were assumed small and
the double-body approximation was used. In the case
of shallow water, the problem was tackled by implementing a truncated series of mirrored images of the
double-body against the water surface.
This algorithm was later extended by Zhou et al.
(2010) to the case of two ships navigating over irregular bottoms by distributing a layer of sources on
a panelled moving patch under the ship in concern.
Added masses and hydrodynamic forces acting on
two interacting ships in shallow water with horizontal flat bottom were computed and compared with
the numerical results using a previously developed
mirrored-image method. A conclusion was drawn that
panelled moving-patch method was able to cope with
interaction problems with irregular seabed with and
produce satisfactory accuracy, however appropriate
panelling was required.
In the present paper, the algorithm was extended
to cover the problem of hydrodynamic interaction
between a ship and the complex flow boundaries of
the waterways in which the ship is navigating. The
modified algorithm was first tested by comparing
its numerical results for pressure distribution over a
sphere moving in deep water, and added masses of
ellipsoids with analytical ones; and then used to simulate interaction situations of a ship navigating in
canals and dredged channels. Computation runs have
been carried out with various configurations of the
Figure 1. Panelling scheme on the water flow boundaries.
parameters: water depth, channel width, cross-section

form of channel and canal, etc, and the results were
analyzed for the applicability of the algorithm to the
problem.
2
GENERAL THEORY
Let the global coordinate system be O, with the

axis laid on the undisturbed water surface, the axis
pointing downwards, and the axis can be determined
by right-handed convention. The advance C , transfer
C , and the heading angle are defined with the help
of the body-fixed frame Cxyz attached to the vessel
in question, with its origin located at the midship, x
pointing to the bow, y directed to the starboard, and
z directed downwards.
The geometry of any complex flow boundaries,
like the seabed, a bank, or a jetty, is expressed and
loaded in the global coordinate system at the beginning of the simulation, and at each computation
moment converted to the body-fixed frame whose
instantaneous location and orientation in the global
system can be obtained by calculating from the origins
velocity VC and the angular velocity C and time
moment t. depending on the complexity of the problem and the required accuracy of the numerical results,
a panelled moving patch of a certain size over the flow
boundaries is chosen in the vicinity of the ship in concern to distribute a layer of sources, as well as the
wetted ship surface, see Figure 1.
In the domain of perfect fluid enclosed by the body
surface, water surface and water flow boundaries (for
instance, seabed, or sidewall of a bank, etc). The total
velocity potential takes the form:
where is the total potential, Vcur is the component of the on-coming flows velocity, Vcur is the
component of the on-coming flows velocity, and
= (, , , t) is the perturbation potential.
Then the induced velocity is:
242
At any time moment, the perturbation potential should

satisfy Laplace equation
and on the ships wetted surface and chosen moving patch on the water flow boundaries, the following
non-penetration boundary condition is applied:
The total hydrodynamic inertial force Fpi and

moment Mpi can be calculated by:
and their components in the body axes are: the surge

force Xp , sway force Yp , and yaw moment Np .
The proper hydrodynamic inertial forces and
moments are
where n is the outward unity normal to the local

geometry, Vr is the relative local velocity:
where V is the absolute local velocity at a point on the

body surface and on the panelled boundaries.
Finally according to the low-Froude-number
assumption, the following equation is satisfied:
where ij are the added mass coefficients and the

pure interaction loads are:
3 VALIDATION OF THE NUMERICAL

ALGORITHM
on the water surface.
As shown in Figure 1, the union of the ships wetted
surface and the moving patch on the water flow boundaries chosen in the vicinity of the ship constitutes the
total surface S in the following equations:
where is the source density, M (x, y, z) is the field

point, P(x , y , z ) is the source point belonging to the
surface S, and G( ) is the Green function:
It is important to examine the validity and accuracy of

the algorithm before using it to compute ship hydrodynamic interaction forces. In this paper, the code is
tested firstly by comparing its results for the pressure
distribution along a sphere with analytical ones, and
through comparing the hydrodynamic forces on 1/8
of a sphere. After that the computed added masses of
ellipsoids are also compared with analytical solutions.
These comparisons partly repeat those performed in
Rodrigues et al (2006), but it was important to make
them after many modifications of the code. A number of variables, arrays and functions were added to
code for implementing a panelled moving patch on
the seabed, and most of the data structure has been
optimized for better computation performance.
where
3.1
Pressure and forces on a sphere in current
The analytical solution for the pressure along the

meridian of a sphere in motion parallel to its axis is:
Once Equation 7 is solved, the induced velocity and
induced potential at each point can be obtained by integration along the panelled surface. In turn, the pressure
can be calculated by the unsteady Bernoulli equation:
where
where VI is the induced velocity.
where p() is pressure at point when its polar angle is

(see Figure 2), p is the pressure at infinity in the
fluid, and , V are the density and velocity of the fluid
respectively.
In this comparative analysis, first, a sphere moving
in unbounded fluid is used to investigate the agreement
between numerical results and analytical ones. The
computed results for the pressure along the meridian
of a sphere show a perfect agreement with analytical
243
Figure 2. 1/8 of a moving sphere in spherical coordinate

system.
Figure 4. Relative error between numerical and analytical

results for the surge and sway force.
Figure 3. Pressure along the meridian of a sphere in the

current.
solution, as seen in Figure 3, where the dimensionless

abscissa x is defined as:
where x is the actual abscissa, and r is the radius of the

sphere.
According to the dAlembert Paradox, the resistance to a body moving with uniform velocity through
an unbounded inviscid fluid, otherwise at rest, is zero.
Therefore, integrating the pressure along the entire
sphere should result in zero force components. For this
reason, the comparison is performed by integrating the
pressure over only 1/8 of the sphere, see Figure 2.
Based on Equation 15 and the coordinate system
defined as shown in Figure 2, the analytical solutions
for the surge, sway force, and the yaw moment can be
calculated by:
Actually, the result calculated in equation (19) could

be expected from the fact that everywhere on the
sphere the vector normal to the local geometry is pointing to the centre O. Therefore, the arm of the moment
produced by the force integrated from pressure on
any infinitesimal piece of sphere surface is zero, and
consequently integrating over any parts of the sphere
would result in zero yaw moment.
The number of panels used to approximate the surface of the geometric object in concern has significant
influence on the computation time and accuracy of
results: a larger number of panels produces better
accuracy but consumes more computation time. In
order to reach a balanced compromise between accuracy and computation time, the comparison between
numerical results with analytical solutions has been
carried out for a variety of numbers of panels
distributed on the surface of a sphere moving in
unbounded deep water.
Figure 4 shows the relative errors of numerical
results for the surge and sway force, which for the
sway force is defined as [(Xn Xa )/Xa ] 100%, where
Xn , Xa are the numerical and analytical result for the
surge force respectively, and for the sway force the
relative error is defined similarly.
244
Figure 6. Axes definitions for ellipsoids.
Figure 5. Numerical results for the yaw moment at different

number of panels.
It can be seen that computation with an insufficient

number of panels only produces poor agreement. However once the number of panels exceeds approximately
40, with the increase of number of panels, the relative
error decreases and tends to zero.
The relative difference between numerical and analytical results for the yaw moment can not be processed
in the same way as done for the surge and sway force
since its precise value is equal to zero.
Figure 5 presents the tendency of the absolute error
of numerical results for the yaw moment with the
increase of number of panels, where it is shown that the
numerical result goes toward to the analytical solution,
i.e. to zero, rapidly with the number of panels increase
from 16 to about 100, after what, with further increase
of the number of panels, the convergence rate starts
to slow down gradually and at very large numbers of
panels, the numerical result becomes infinitely close to
zero. This convergence pattern suggests for this case
in concern, 100 can be a good choice of number of
panels to obtain acceptable accuracy of results without
costing too much computation time.
3.2 Added masses of ellipsoids

Ellipsoids in motion in infinite ideal fluid represent
one of the few cases for which the added masses
can be obtained analytically. This is exploited in this
paper for the purpose of revealing the behaviour of
the algorithm applied to computation of added masses
which has been carried out for an affine set of ellipsoids with various length ratios of semi-axis a/b
(Figure 6).
The analytical solution presented by Korotkin
(2009) for added masses of ellipsoids is used here
for the comparison. Figure 7 presents the relative difference between numerical results and analytical ones
for the added mass 11 of each ellipsoid in the set for
various number of approximation panels, where it is
shown that an insufficient number of panels always
Figure 7. Relative error of numerical results for 11 at

different number of panels.
results in poor accuracy for any semi-axis ratio a/b

of ellipsoid. On the other hand, one can also observe
that the algorithm requires more panels for consistent
numerical results for ellipsoids of higher ratio of a/b.
As seen in the figure, the numerical results for 11
of a sphere start to converge with the analytical one
when the number of approximating panels reaches
102.2 (about 160), in contrast to this, the algorithm
requires more than 1000 panels to produce consistent
results for the ellipsoid of a/b = 6. The third observation is that larger ratio of a/b always results in larger
relative errors.
Figure 8 and 9 respectively show 22 and 66 of
them at each ratio of a/b, except 66 for a/b = 1, when
it is zero. In all the cases, the algorithm shows robust
performance in convergence and satisfactory accuracy
as compared with analytical solutions, once a sufficient number of panels are used to approximate the
geometry.
HYDRODYNAMIC FORCES IN RESTRICTED

WATERS
The ship form that has been used for the calculation and
analysis throughout this work has main dimensions
L B T of 189.6 m 31.6 m 10.3 m, block coefficient of 0.815. As discovered in Zhou et al (2010), the
distribution of sources over a sufficiently large patch
on the seabed is required for acceptable accuracy. In
245
Figure 10. Ship moving in a dredged channel with vertical

sidewalls.

Figure 11. Ship moving in a dredged channel with sloped
sidewalls.

this study a square-shaped patch of 360 m 360 m on

the water flow boundaries around the ship under study
is used. In the case of canal sidewalls or banks where
the domain of water flow is limited in the vicinity of
the ship, the panelled moving patch is also naturally
defined by the domains boundaries.
4.1
Dredged channel with varying transverse

displacement of the ship
Two types of the dredged approach channels are studied: with the rectangular cross-section (Figure 10) and
with the trapezoidal cross-section (Figure 11). The
water depth outside the sidewall as denoted d1 and the
depth of the channel itself as d2 , and b0 stands for the
lateral distance between the ships centre plane and the
right sidewall of the channel, negative as leftward.
Figure 12. Influence of transverse displacement on the sway

force.
Both of the dredged channels are 72 m wide (corresponds to the average width in the trapezoidal
case), d1 = 14 m, d2 = 26 m. Figure 12 and Figure
13 respectively present the sway force and the yaw
moment non-dimensionalized by
as functions of the relative distance between course

and the sidewall b0 /Wd , where Wd is the width of
the channel. For example, b0 /Wd = 0.5 indicates that
the ship is moving in the middle of the channel, and
b0 /Wd = 0 means that the ship is moving right over the
sidewall.
As could be intuitively expected, the sway force and
yaw moment acting on the ship are zero when she is
246
Figure 13. Influence of transverse displacement on the yaw

moment.
Figure 14. Influence of the height of the sidewall on the

sway force.
sailing along the centre plane of the dredged channel.

However, the peaks of sway force and yaw moment
appear not exactly when the ship is riding over the
sidewall, but when its path is shifted somewhat outside of the dredge channel. Another observation is that
forces are slightly higher by absolute value in the case
of vertical walls than that in the case of sloped walls.
Figure 15. Influence of the height of the sidewall on the yaw

moment.
Figure 16. Influence of the water depth on the sway force.
As can be seen from the results shown in Figure 14

and 15, the sway force and the yaw moment increase
with the height of the sidewall, but at decreasing rates.
Especially this kind of situation happens for the yaw
moment in the case of sloped sidewall, when the height
is larger than 1.5 times the ships draught.
4.3
4.2
Dredged channels with different heights of the

sidewall
In this analysis, a ship is sailing right on top of the

sidewall of a dredged channel, the water depth d2
is 14 m, and the water depth d2 is depending on the
height of the sidewall, which varies from 2 m to 18 m,
ranging from 0.193 to 1.737 in terms of relative height
of the sidewall Hs as calculated as:
Dredged channels with various water depths
The canal can be viewed as a limiting case of a dredged

channel when d1 = 0. In this analysis, d1 varies from
0 to 14 m (1.35T ) while d2 remains equal to 18 m
(1.74T ). The computation of hydrodynamic forces has
been carried out for two different transverse shifts:
b0 /B = 0.66 and b0 /B = 0.95. The numerical results for
the sway forces and the yaw moment acting on the ship
are respectively presented in Figure 16 and Figure 17,
where d1 = 0 stands for the canal case and the rest are
the case of dredged channel. In both cases of sidewall
forms, the transition from canal to dredged channel
is very smooth for sway force, but a little abrupt for
yaw moment. This phenomenon probably arose from
247
Figure 17. Influence of the water depth on the yaw moment.

Figure 19. Influence of the width of dredged channel on the
yaw moment.
twice B, and after that it starts to decrease. However, no

explanations were found for the interesting effect that
further increase of width has no longer effect on the
hydrodynamic forces when the width reaches a certain
value. This phenomenon happens to the computation
of both cross-section forms, the only difference is this
turning point appears earlier in one of the cases, later
in the other.
5
Figure 18. Influence of the width of dredged channel on the

sway force.
the sudden change of panelling scheme on the water

boundaries. It also can be observed that also in both
cases the hydrodynamic loads on the ship drop rapidly
as the d1 increases, and become relatively small when
d1 exceeds 1.2T .
4.4
Dredged channels of various width
The only parameter that varies here is the average width

of the channel, while the depth d1 remains equal to
14 m, and d2 26 m.
The numerical results are presented in Figure 18
and Figure 19, as a function of Wd /B, where Wd is
the width of the dredged channel. As the width of the
channel increases from a small value (smaller than
the beam of the ship), both the sway force and the
yaw moment become larger. This can be explained
by the cancellation effect from two sidewalls which
reduces when one of the walls is moving away. The
yaw moment predicted by the potential code reaches its
maximum when the channels width is approximately
CONCLUSIONS
Validation of the developed algorithm shows that it

is able to produce practically acceptable results for
relatively complicated configurations. A variety of
simulated interaction cases demonstrate that the problem of the ship interaction with complex flow boundaries can be dealt with this algorithm, which can also be
extended to handle multi-ship interaction in confined
waters.
ACKNOWLEDGEMENTS
This work was performed within the project Dynamics and Hydrodynamics of Ships in Approaching
Fairways (CONFINED WATERS) financed by the
number PTDC/ECM /65806/2006.
REFERENCES
Beck , R. F. 1977. Forces and moments on a ship moving
in a shallow channel. Journal of Ship Research, Vol. 21,
No. 2, June 1977, pp. 107119.
Chen H.-Ch., Lin, W.-M., Liut, D. A., Hwang, W.-Y. 2003. An
advanced viscous flow computation method for shi-ship
dynamic interactions in shallow and restricted waterway.
248
Proceedings International Conference on Marine Simulation and Ship Maneuverability (MARSIM03), 2528 Aug
2003, Kanazawa, Japan, Vol. 3, pp. RC-35-110.
Chng P. W. 1991. An investigation into the influence of
bank effect on ship manoeuvring and its mathematics modeling for a ship-handling simulator. ME. Thesis.
School of Mechanical and Manufacturing Engineering,
The University of New South Wales.
Chng P. W., Doctors, L. J., Renilson, M.R. 1993. A method of
calculating the ship-bank interaction forces and moments
in restricted water. International. Shipbuilding. Progress.
Vol. No. 421, pp. 723.
Hess J. L., Smith, A. M. O. 1964. Calculation of nonlifting
potential flow about arbitrary three-dimensional bodies.
Journal of Ship Research. pp. 2244.
Hess J. L., Smith, A. M. O. 1967. Calculation of potential flow
about arbitrary bodies. Progress in Aeronautical Sciences,
1967, Vol. 8, pp. 1137.
Kijima K. 1997. Prediction method for ship manoeuvring
motion in the proximity of a pier. Schiffstechnik/Ship
Technology Research. Vol. 44, pp. 2231.
Korotkin A. I. 2009. Added Masses of Ship structures.
Springer.
Laforce E., Vantorre M. 1996. Experimental determination
and modelling of restricted water effects on bulkcarriers.
Marine Simulation and Ship Manoeuvarability, Chislett
(ed.), Balkema, Rotterdam. pp. 451456.
Lataire E., Vantorre, M., Laforce, E., Eloot, K., Delefortrie,
G. 2007. Navigation in confined waters: influence of bank
characteristics on ship-bank interaction. Proceedings of
the 2nd International Conference on Marine Research and
Transportation, Ischia, Naples, Italy, 2830 June 2007.
pp. 19.
Li D-Q, Leer-Andersen, M., Ottoson, P., Trgrdh, P. 2001.
Experimental investigation of bank effects under extreme
conditions. PRADS 2001 Practical Design of Ships and
other Floating Structures, Shanghai.
Norrbin N. H. 1974. Bank effects on a ship moving through a
short dredged channel. Proceedings Tenth symposium on
Naval Hydrodynamics, Cambridge.
Rodrigues J.M., Sutulo S., Guedes Soares C. 2006. Estimation of Inercial Hydrodynamic Forces on Ships in
Presence of Obstacles by Means of a 3D Panel Method
(in Portuguese), In: Inovao e Desinvolvimento
nas Actividades Martimas, C. Guedes Soares and
V. Gonalves de Brito (Eds.), Edies Salamandra, Lisboa

2006, pp. 533550.
Sding H. 1993. A method for accurate force calculations
in potential flow. Ship Technology Research, Vol. 40,
pp. 176188.
Sding H., Conrad F. 2005. Analysis of overtaking manoeuvres in a narrow waterway. Ship Technology Research,
Vol. 52, pp. 189193.
Sutulo S., Guedes Soares, C. 2008. Simulation of the
hydrodynamic interaction forces in close-proximity
manoeuvring. Proceedings of the 27th Annual International Conference on Offshore Mechanics and Arctic
Engineering (OMAE 2008), Estoril. Portugal, 1519 June
2008. Paper OMAE2008-57938, 10p.
Sutulo S., Guedes Soares, C. 2009. Simulation of closeproximity maneuvers using an online 3d potential flow
method. Proceedings of International Conference on
Marine Simulation and Ship Manoeuvrability (MARSIM 2009), Panama City, Panama, 1720 August 2009,
pp. M-9-1M-9-10.
Vantorre M., Delefortrie, G., Eloot, K., Laforce, E. 2003.
Experimental Investigation of Ship-Bank Interaction
Forces. Proceedings International Conference on Marine
Simulation and Ship Maneuverability (MARSIM 2003),
2528 August 2003, Kanazawa, Japan, Vol. 3, pp. RC-311RC-31-9.
Varyani K. S., Krishnankutty, P., Vantorre, M. 2003. Prediction of load on mooring ropes of a container ship due to
the forces induced by a passing bulk carrier. Proceedings
International Conference on Marine Simulation and Ship
Maneuverability (MARSIM 2003), 2528 August 2003,
Kanazawa, Japan. Vol. 3, pp. RC-33-1RC-33-9.
Varyani K. S. 2006. Guide to new generic equations for interaction forces/moments between moored-passing ship and
between encounter-overtaking-overtaken ships. Proceedings International Conference on Marine Simulation and
Ship Manoeuvrability (MARSIM 2006), 2530 June 2006,
Terschelling, The Netherlands. pp. M-23-1M-23-10.
Zhou, X., Sutulo S., Guedes Soares C. 2010. Computation of
ship-to-ship interaction forces by a 3d potential flow panel
method in finite water depth. Proceedings of the 29th
Annual International Conference on Offshore Mechanics
and Arctic Engineering (OMAE 2010), Shanghai, China,
611 June 2010, Paper OMAE2010-20497, 10p.
249
Ship structures

Influence of boundary constraints on the residual stresses and

deformations in a high strength steel plate after welding
M. Adak & C. Guedes Soares
ABSTRACT: The aim of this study is to investigate the influences on the residual stress field and deformations
of three different boundary constraints during butt-join welding in a high strength steel plate. The distortion of
a welded plate is a nonlinear problem in geometry and material, therefore the finite element analysis conducted
is based upon the thermo-elastic-plastic and large deflection theory. High strength shipbuilding steel AH36 with
temperature dependent material properties and nonlinear stress-strain material properties (bilinear isotropic
hardening option uses the von Mises yield criteria) are assumed for the numerical analysis. Three different
boundary conditions are considered in this study. In case 1, end edge parallel to welding line is fixed, in case 2,
edge along x-axis (top of the plate) is fixed and the edge parallel to x-axis (bottom of the plate) is fixed during
welding. The results of the thermal analysis are compared with the ones obtained by another researcher.
1
INTRODUCTION
When a material is heated to an elevated temperature

in a welded joint, the highly localized transient heat
and strongly nonlinear temperature fields in both heating and cooling processes cause nonuniform thermal
expansion and contraction and thus result in plastic
deformation in the weld and surrounding areas. During welding, one edge of the plate can be assumed to
be fixed to reduce the deformation. Therefore, stress
is generated due to fixed constraints during welding.
If the stress level is very high then it can be harmful
to the structure. So, it is necessary to investigate how
much stress is generated in a plate after welding due
to boundary constraint.
Due to the complexity of the physical processes
involved in welding, however, simple mathematical
solutions can not address the practical manufacturing process. Furthermore, it is also impossible for any
experimental technique to obtain a complete mapping
of the residual stress and deformation distribution in a
general welded structure.
The research activity in welding simulation started
decades ago. Rosenthal (1946) was among the first
researchers to develop an analytical solution of heat
flow during welding based on conduction heat transfer
for predicting the shape of the weld pool for two and
three dimensional welds. Watanabe and Satoh (1961)
used the analytical solutions resulting from the theory
of elasticity for prediction of thermal deformations due
to welding and line heating. However, since elastic
solutions are limited, application of the method is also
limited.
The first numerical application of a 2D finite element analysis was reported by Hibbit et al. (1973) to
predict residual stresses in a welded plate. Due to computational and cost limitations, FE simulation efforts
during the 70 s and 80 s were focused on simplified 2D
geometries by Karlson (1986) and Masubuchi (1980).
In reality, the thermal and stress-strain responses of all
weldments are three dimensional (3D).
With increasing computational power of present
day computers, numerical methods based on finite
element and finite difference schemes have been
extensively used. Kristina and George (1990), Argyris
et al. (1982), Rybicki et al. (1978), Papazoglou and
Masubuchi (1982), Murakawa et al. (1995) worked
out thermo-elasto-plastic problems using high speed
workstations but the resulting deformations showed
poor accuracy compared to the experimental results.
Then, 3D welding simulations were conducted
using commercial FEA software which were reported
by Tekriwal et al. (1988, 1991) and Dong et al.
(1997).
Biswas et al. (2007) modeled the angular distortion
using a plasticity-based distortion welding. Adak and
Mandal (2009, 2010) used the pseudolinear equivalent
constant stiffness system to predict the temperature
distribution and deformation shape after welding. It
is clear from the literature survey that most of them
have concentrated on 2D and 3D simulation process
for deformation shape and residual field due to welding without boundary constraints. Mulhim et al. (2002)
and Sunar et al. (2006) investigated thermal and stress
analysis with constant heat source at weld pool and
they investigated the residual stress field due to only
one side is fixed at welding region during welding in
case of 2D problem.
Not only the welding residual stress and distortion have been studied by welding researchers, but
253
the effect of welding parameters, welding sequence,

and welding joint geometry. Teng et al. (2003) and
Ji et al. (2005) investigated the distortion mechanism
and the effects of welding sequence on thin panel distortions using a finite element analysis approach. Zhu
and Chao (2002) investigated the effect of each temperature dependent material property on the transient
temperature and residual stress by using finite element
analyses approach.
Fanous et al. (2003) simulated the effects on residual stresses by choosing the appropriate boundary
conditions in modeling the welding process. Three
boundary condition sets are considered; in the first
boundary conditions set, the plates are welded along
the joining side with clamp at one end, after which the
plates are released and checked for residual stresses. In
second boundary conditions set, the plates are welded
with a large structure so that the plate may expand
freely but with no rotation allowed through out the
process. In the third set, the structural boundary conditions are the same as that of set 1 but with a decrease
in the thermal load. By the change in boundary conditions the stresses generated during the welding process
are not highly affected.
Early Sunar et al. (2006) investigated the residual
stress due to only one boundary constraint which is
applied at heating side with fixed temperature as a
heat source. The present study is an extension of that
work where three different boundary constraints are
applied at different edges of the plate to reduce the
deformation and their influence on residual stresses
are investigated after welding by using 3D finite
element analysis software ANSYS with temperature
dependent material properties and nonlinear stressstrain material properties.
NUMERICAL SIMULATION PROCEDURE
In this work, the process of welding is simulated by the

FE method. The welding process computation can be
split into two solution steps: thermal and mechanical
analysis. First, the temperature is determined as a function of time in thermal analysis. Then, the mechanical
analysis employs the previous results to get displacements at nodes and stress at integration points. Since
the thermal field has a strong influence on the stress
field, a sequentially uncoupled analysis works very
well. Therefore, in this paper, the welding process is
simulated using a sequentially uncoupled 3-D thermomechanical FE formulation based on theANSYS code.
For both the thermal and mechanical analyses, temperature dependent thermo-physical and mechanical
properties of the materials are incorporated.
2.1
Mathematical modeling
Figure 1. Variation of location for boundary constraints.
2.1.1 Thermal analysis

The mathematical model can be used threedimensional, since the plate is considered with
thickness. The conduction heating of the solid with
constant temperature source at the fixed end is considered to simulate the welding process for 2-D problem
reported by Al-Mulhim et al. (2002) and Sundar et al.
(2006). Here, the 3D transient heat conduction equation for a rectangular plate heated at one side can be
written as
where , C and k are density, specific heat and thermal

conductivity as a function respectively.
2.1.1.1 Initial condition
In case of thermal analysis, at time zero, a uniform
temperature that means room temperature is assumed
except weld pool region.
At time t = 0, T(x, y, z) = T0 and T(x, y, z) = T1
(fixed temperature for weld pool).
2.1.1.2 Boundary condition
At the free surface (except in the weld pool) of
the rectangular plate (Figure 1), convection boundary
conditions are considered. Therefore,
On the surface along the x axis and parallel to x
axis,
The mathematical models for the temperature and

stress fields are given under the appropriate subheadings.
254
Along the weld pool,

T = constant (specific temperature);
On the surface parallel to z axis,
On the surface top and bottom surface,

Figure 2. Specimen for analysis.
where h is the convection coefficient.

The heat exchange between the welded plate
and its surroundings during welding and subsequent
cooling takes place by both convection and radiation. In the thermal model the convection coefficient h = 15 W m2 , emissivity = 0.9 and the
Stefan-Boltzmann constant = 5.669 108 W m2
0 4
C are used according to Long et al. (2009). These
thermal boundary conditions are applied for all surfaces of the welding plate except for the symmetrical
plane of the plate.
Figure 3. Element type.
2.1.2 Stress field

During the welding procedure, the heating is localized and, therefore, a very large temperature variation
occurs over a small region. Consequently, considerable thermal stresses are generated due to the thermal
gradient. In the present study, the stress vector is
related to the mechanical strain vector e by
2.2 Methodology of the numerical simulation

ANSYS software (2002) is used to conduct the stress
analysis of the high strength rectangular plate with
different fixed edges. The following procedures are
adopted for the simulations.
and
2.2.1 Geometrical design

For this purpose, a rectangular 3D plate which is symmetric with respect to the welding line is assumed.
So for the analysis to reduce the calculation time,
half symmetry of the plate with length 250 mm
(z-direction), width 100 mm (x-direction) and thickness 6 mm (y-direction) is considered which is shown
in Figure 2.
The structural behaviour causes stress due to thermal gradient and the fixed edge. A rate-independent
thermo-elasto-plastic material model with temperature dependent material properties is incorporated into
the finite element modeling.
In the case of mechanical analysis, three cases for
boundary constraints are considered.
2.2.2 Element type

For thermal analysis a 3D 8-node element (solid 70)
with a single degree of freedom, temperature, at each
node is used, and for structural analysis the element
type is solid 185. This choice was arrived at a parametric study using different element types shown in
Figure 3 which led to the conclusion that choosing an element having both thermal and structural
characteristics is not optimal.
Case 1:The nodes at the extreme end from the weld line
are constrained. Symmetry boundary condition is
applied at the weld line shown in Figure 1.
Case 2: The nodes at the edge along x-axis (bottom)
are constrained. Symmetry boundary condition is
applied at the weld line shown in Figure 1.
Case 3: The nodes at the edge parallel to x-axis
(top) are constrained. Symmetry boundary condition is applied at the weld line shown in Figure 1.
These three cases are solved for residual stresses in
mechanical response.
2.2.3 Mesh generation

In order to generate the system of equations, as well
as to find the result at each point of the rectangular
plate, a mesh is generated throughout the volume. In
order to achieve this, the ANSYS program allows for
control of the size and geometry of the mesh in order to
obtain the most precise solution. However, a very large
number of grid points can compromise the computers
capacity. Figure 4 shows the mesh in the symmetry
half of the rectangular plate. Welding direction being
z-axis, sharp temperature gradients are expected along
where,
255
Table 4. Thermal expansion versus

temp.
(106 / C)
25
1450
3000
12
14
14
Table 5. Elastic Modulus and Poissons ratio

versus temp.
Figure 4. The mesh in the half of symmetry plate.

Table 1.
25
1450
3000
Density versus temperature.

Kg/m3
7800
7438
7438
GPa
25
100
200
300
400
500
600
700
800
900
1000
3000
206
203
201
200
165
100
60
40
30
20
10
10
0.296
0.311
0.330
0.349
0.367
0.385
0.405
0.423
0.442
0.461
0.480
0.480
Table 2. Thermal conductivity versus temp.
25
750
800
1450
3000
Table 3.
25
700
740
860
1450
3000
W/m C
55
29
25.4
31
31
Specific heat versus temperature.

J/kg. C
388
894
1234
610
814
814
Figure 5. Stress-strain behavior.
both x and y-axes. Therefore very fine mesh grid has

taken in weld bead region; finer mesh grid has taken
away from weld region and course grid is chosen along
welding direction (z-axis).
2.2.4 Material modeling
In this numerical simulation temperature depended
material properties are used. Steel grade AH36 (high
strength shipbuilding steel) is implemented with temperature depend on properties as specified in Tables 1
to 5 and Figure 5.
The Bilinear Isotropic Hardening option was
adopted and it uses the von Mises yield criteria coupled
with an isotropic work hardening assumption shown
Figure 6. Bilinear Isotropic hardening.
in Figure 6. This option is often preferred for large

strain analyses. The material behavior is described by
a bilinear stress-strain curve starting at the origin with
positive stress and strain values. The initial slope of the
256
Figure 7. Temperature along transverse direction (x-axis)

from Longs paper (2009).
Figure 9. Temperature distribution along centerline at

various moments.
verification model are slightly different from Longs

possibly due to mesh size. Figure 7 and Figure 8 show
the temperature distribution curves from Longs and
verification models, respectively.
4.1 Temperature history
Figure 8. Temperature along transverse direction (x-axis).
curve is taken as the elastic modulus of the material.

At the specified yield stress, the curve continues along
the second slope defined by the tangent modulus (having the same units as the elastic modulus). The tangent
modulus can not be less than zero nor greater than the
elastic modulus.
3 VERIFICATION OF BUTT-WELDING MODEL

In this study, a finite element simulation of a simple butt-joint welding for 3D model is verified along
the work of Long et al. (2009). To do this, a simple
butt joint welding whose welding parameters are consistent to those of Longs model was simulated by
using ANSYS code. The model then is verified by
comparison of its temperature distribution at some
chosen nodes with the Longs model. The temperature dependent material properties for the thermal and
mechanical analyses were equivalent to those used
in Longs analyses. Plate geometry of the butt-joint
welding is similar to the Longs model.
The results show that the temperature histories at
various locations from the weld line are in reasonable
agreement with the result obtained by Long (2009).
The peak temperatures for both models are almost
identical. The distances from the weld line of the
After having verified that the finite element adopted

produces results consistent with previous studies, this
model will be used to study the problem formulated
for this paper, i.e. the effect of the different boundary
conditions. In thermal analysis, constant temperature
is assumed for simplifying the heat source model at
butt-joint of plate during welding process. In the thermal model, the heat exchange between the welded plate
and its surroundings during welding and subsequently
cooling takes place by both convection and radiation.
The model is used for thermo-mechanical analysis
with dimensions 250 mm 100 mm 6 mm for 6 mm
thick plate. Temperature fields are obtained numerically and the corresponding stress fields with boundary constraint for three different cases are obtained
using the FEM. Temperature dependent material properties for thermo-mechanical analysis mentioned in
above tables.
The temperature distributions in the transverse
direction (perpendicular to welding line) at the centre of the welded plate for different heating periods
are shown in Figure 9. This figure shows that the peak
temperature is high at the welding line and temperature is gradually decreased along x axis due to cooling
phase. At different time steps the same pattern is coming in temperature history and it is changed in the
same place at different moments. Figure 10 shows that
during welding and heating treatment, transient temperature distribution is changed at the same time in
different places. Temperature profile along the transverse direction, repeats itself till the entire plate cools
down to room temperature.
The contour plot in Figure 11 shows that the temperature at the weld zone is very high and it is
257
Figure 10. Transient temperature distribution at various

points in transverse direction.
Figure 12. Transverse stresses (x ) along x-axis at middle

of plate.
Figure 13. Transverse stresses (x ) along welding line.
Figure 11. Temperature contour in C for 36.23 sec heating

duration.
gradually decreased away from weld zone for a particular moment through out the plate. These changes
of transient temperature will significantly affect the
forming and developing of the residual stresses.
Figure 14. Deflection along x axis at middle of the plate.
4.2 Welding residual stresses and deformations

In the mechanical analysis, the large deformation theory is used to simulate welding residual stress in the
butt-welded joint plate. Three boundary constraints
are applied for different cases in mechanical analysis
and their effects are observed. The results for different
boundary conditions show a significant influence on
transversal stresses, especially regarding a path transverse to welding direction at a centre of the plate shown
in Figure 12. From the transverse stress calculation
for three cases it is clear that stress level is very low
about 70 MPa for both transverse boundary influences
in case 2 and case 3 which is much less than the yield
strength of the plate, i.e. the elastic behavior is the
governing mechanism for the stress field. But stress
level for the boundary condition which parallel to
welding line in case 1 is high about 460 MPa which is
greater than the yield strength, i.e. the plastic behavior
is the governing mechanism for the stress field.
Those differences result from the fact that plate having no constraints at x = 100 mm can deform freely
in this direction therefore it will have smaller stress.
Transverse stresses are decreased with free ends. It
is assumed that with increasing length this positive
effect on local stress development will be negligible.
For this simulation the length of the plate is 100 mm,
which means influence of boundary changing at the
plate end is large and may be reasonable for decreased
local transverse stresses.
Along the welding line, transverse stresses (within
yield stress) show almost no differences. Transverse
stresses (Fig. 13) however present different stress
levels at the beginning as well as at the end of the
weld.
Figure 14 shows the residual deformation along x
axis at middle of the plate for three different cases.
For case 1, the deformation is much smaller than in
258
the other two cases. In case 2 and 3 deformations are

almost the same, but the highest value of deformation
is around 0.15 mm which may be negligible.Therefore,
for three different boundary constraints help the less
residual deformation, but for case 1 residual stress is
high more than yield strength of the plate.
5
CONCLUSIONS
In this study, the thermo-elastic-plastic FEM is used to

simulate the welding temperature field and the influence of three boundary constraints along different
edges on the residual stress field in a high strength
butt joint plate. Since the heating is transient, time
dependent equations of heat transfer are employed in
the analysis. It is found that the temperature distribution in the longitudinal direction (along welding line)
does not vary considerably since constant temperature is applied at the weld joint during analysis, but
the temperature varies significantly in the transverse
direction. These changes of transient temperature will
significantly affect the forming and developing of the
residual stresses.
In mechanical response, the results for three different cases show a significant influence on transversal
stresses, especially regarding a path transverse to welding line at middle of the plate. While results for both
plates fixed at transverse direction to welding line
present relatively small stress about 70 MPa, which
is much less than the yield strength of the plate, i.e.
the elastic behavior is the governing mechanism for
the stress field. The plate fixed at the side parallel
to the welded area has a stress level about 460 MPa
which is greater than the yield strength, i.e. the plastic behavior is the governing mechanism for the stress
field. Altogether results are expected that along the
welding line transverse stresses are almost same for
three different constraints. From these results it can
be concluded that no influence on residual stresses for
two boundary constraints (case 2 and case 3), but the
significant affect on residual stresses in case 1 after
welding. Therefore, case 2 and case 3 are more suitable for applying boundary constraints to reduce the
residual deformation.
ACKNOWLEDGEMENTS
This work has been performed and partially funded
by the Portuguese Foundation of Science and Technology (Fundao para Cincia e Tecnologia) through
its plurianual programme to the Centre for Marine
Technology and Engineering (CENTEC), and by the
EU funded project Network of Excellence for Marine
Structures (MARSTRUCT), under contract number
TNE3-CT-2003-506141.
REFERENCES
Adak M. and Mandal N. R., (2009), Pseudolinear equivalent constant rigidity concept for analyzing welding
deformations, Applied Mathematical Modeling, 2009,

33, 20962108.
Adak M. and Mandal N. R., (2010), Numerical and experimental study of mitigation of welding distortion, Applied
Mathematical Modeling, 34, 146158.
Al-Mulhim K., Said S.A.M., Yibas B.S., Habib M.A. and
Al-Bagawi J., (2002), Multi-dimensional transient heat
conduction in heat exchanger tube sheets, Proc. Inst.
Mech. Eng. Part B: J. Eng. Manuf., 216, 331345.
Ansys, Inc., Theory reference, Ansys Inc., 2002.
Argyris J.H., Szimmat J., and Willam K.J., (1982), Computational Aspects of Welding Stress Analysis, Computer
Methods in Applied Mechanics and Engineering, 33,
635666.
Biswas P., Mandal N.R. and Sha O. P., (2007), Three
dimensional finite element prediction of transient thermal history and residual deformation due to line heating,
Journal of Engineering for the Maritime Environment,
221, 1730.
Dong Y., Hong J.K., Tsai C.L., Dong P., (1997), Finite element modeling of residual stresses in austenitic stainless
steel pipe girth welds, Welding Journal, 76, 442s449s.
Fanous F. Z. I., Younan Y.A. and Wifi S.A., (2003), Study
of the effect of boundary conditions on residual stresses
in welding using element birth and element movement
techniques, Journal of Pressure Vessel Technology, 125,
432439.
Hibbitt H.D., Marcal P.V., (1973), A numerical thermomechanical model of the welding and subsequent loading
of a fabricated structure, Computers and Structures, 3,
11451174.
Ji S. D., Fang H. Y., Liu S. X. and Meng Q. G., (2005), Influence of a welding sequence on the welding residual stress
of a thick plate, Modelling and Simulation in Material
Science and Engineering, 13, 553565.
Karlsson L., (1986), Thermal stresses in welding, In:
Hetnarski R.B., editor, Thermal stresses I. Elsevier
Science Publishers, 1986, 299389.
Kristina, A. et al., (1990), Artificial Neural Network
Applied to Arc Welding Process Modeling and Control, IEEE Transaction on Industry Applications, 26,
824830.
Long H., Gery D., Carlier A. and Maropoulos P.G., (2009),
Prediction of welding distortion in butt joint of thin
plates, Materials and Design, 30, 41264135.
Masubuchi K., (1980), Analysis of welded structures,
Oxford, UK: Pergamon Press.
Murakawa, H., Ueda, Y., and Zhong, X. M., (1995), Buckling Behavior of Plates Under idealized Inherent Strain,
Transactions of JWRI, 24, 8791.
Papazoglou, V.J. and Masubuchi, K., (1982), Numerical
Analysis of Thermal Stresses During Welding Including
Phase Transformation Effects, Journal of Pressure Vessel
Technology, 104, 198203.
Rosenthal D., (1946), The theory of moving sources of heat
and its application to metal treatments, Transactions of
A.S.M.E., 68, 849866.
Rybicki, E. F., Schmueser, D. W., Stonesifer, R. B., Groom,
J.J., and Mishler, H. W., (1978), A Finite Element Model
for Residual Stresses and Deections in Girth-Butt Welded
Pipes, Journal of Pressure Vessel Technology, 100,
256262.
Sunar M., Yilbas B. S. and Boran K., (2006), Thermal and
stress analysis of a sheet metal in welding, J. of Materials
Proc. Tech., 172, 123129.
Tekriwal P., and Mazumder J., (1988), Finite element analysis of three dimension transient heat transfer in GMA
welding, Welding Journal, 67, 150s156s.
259
Tekriwal P., and Mazumder J., (1991), Transient and residual thermal strain-stress analysis of GMAW, Journal of
Engineering Materials and Technology, 113, 336343.
Teng T. L., Chang P. H. and Tseng W. C., (2003), Effect of
welding sequences on residual stresses, Computers and
Structures, 81, 273286.
Watanabe, M. and Satoh, K., (1961), Effect of Welding Conditions on the Shrinkage Distortion in Welded Structures,
Welding journal, 40, 377s384s.
Zhu X. K. and Chao Y.J., (2002), Effects of temperature
dependent material properties on welding simulation,
Computers and Structures, 80, 967976.
260

Innovation protecting ballast tanks of new ships against corrosion:

An IMO PSPC-compliant Portuguese-born new coating concept
J. Azevedo
Euronavy SA/The Sherwin-Williams Company, Setbal, Portugal
ABSTRACT: The new IMO Performance Standard for Protective Coatings (PSPC), mandatory since July 2008
for new ships, is described. An evaluation is made of the PSPC troublesome impact in cost and time needed
for ballast tank and double-side skin spaces coating application in new ships, if the new standard is applied
strictly, using traditional coating technologies and application methods. But the menaces of a strict application
of the new PSPC can be avoided, thanks to the PSPC openness to innovation. This paper also reviews the IMO
PSPC regulation to highlight the parts clearly encouraging innovation and offering opportunities for innovative
products & methods to be used in order to save time and costs. A leaving proof of this possibility is presented,
describing a solvent-free humidity tolerant epoxy coating system able to be applied over water jetting as surface
preparation method in the block stage and over a zinc-free organic shop primer. Current PSPC Type Approvals
and testing performance data of this innovative coating system are presented, together with hard-facts from
application in real-life PSPC projects now running. A conclusion is made suggesting that IMO new standard can
indeed be the tool for more effective corrosion protection without forgetting the need for shipyards efficiency.
1
INTRODUCTION
1.1 The PSPC

The Performance Standard for Protective Coatings
for Dedicated Seawater Ballast Tanks of all Types of
Ships and Double-Side Skin Spaces of Bulk Carriers
(International Maritime Organization, 2006), as the
new IMO PSPC is fully named, was adopted at the
International Maritime Organization (IMO) Marine
Safety Committee meeting N.82 in December 8 2006.
These rules are mandatory for tankers and bulk carriers
contracted after December 6 2006 if to be covered by
the Common Structural Rules agreed by IACS (International Association of Classification Societies). All
vessels over 500GT contracted on or after 1st July 2008
will also have to comply, independently of the CSR,
to comply with SOLAS regulation II-1/3-2 as adopted
by resolution MSC.216(82).
1.2 The challenge

The IMO PSPC poses a big challenge for owners, yards
and paint manufacturers. The new rules raised the bar
for the target useful life of corrosion protection on
ballast tanks and double skin spaces in newly built
ships. The coating system shall now be designed for
a minimum durability of 15 years, meaning that the
coating of tank and void areas shall retain the notation
GOOD during that period. It applies to both coating
materials qualification and to surface preparation and
paint application methods, which are described in IMO
PSPCs Table 1. To cope with the new rules using the
most common technologies of surface preparation and

coating materials, shipyards and owners have to face
higher costs and lengthy building schedules.
Almost one year before IMO PSPC was adopted,
the author presented an article (Azevedo, 2006) at the
Royal Institute of Naval Architects (RINA) Advanced
Marine Materials & Coatings Conference, defending the advantages of a particular approach to cope
with the Tanker Structures Cooperative Forum (TSCF)
Guidelines for Ballast Tank Coatings (Intertanko,
2002). The article also anticipated how such approach
(the use of a humidity-tolerant solvent-free coating
system over water jetting) could help to efficiently
cope to the future (now present) IMO rules for ballast tank coating. Additional data was presented at a
marine coatings conference in Hamburg (Azevedo,
2008) showing the progress done at the time in
terms of Type Approvals for the system. Four years
after the original paper and again at a RINA conference (Azevedo, 2010a), it was possible to share
how that proposed solution was implemented and to
what extent it coped with the expectations, regarding
PSPC compliance, performance target achievements
and the aimed shipyard processes efficiency. Further progress was made in 2010 and a new paper
was presented later that year (Azevedo, 2010b) in
Hamburg.
This paper is an update of the last couple of published papers. Please note that it reflects the individual
Class positions only to the extent of the TypeApprovals
already granted. Any other opinions expressed and
not yet covered by a Type Approval are of exclusive
responsibility of the author.
261
INNOVATION VS. PSPC COMPLIANCE
Table 1. PSPC testing acceptance criteria: normal vs.

alternative systems.
2.1 The innovative standard

RECOGNIZING that the Performance standard for
protective coatings () is not intended to inhibit
the development of new or novel technologies which
provide for alternative systems, () INVITES Governments to encourage the development of novel technologies aimed at providing for alternative systems and to
keep the Organization advised of any positive results.
The italic excerpt shown in the above paragraph
is extracted from the IMO PSPCs front page. It is
clear evidence that IMO PSPC does appeal for and
encourage innovative solutions. One of the intents of
this article is to defend that PSPCs interpretation must
be on technical grounds and should not conflict with
its declared spirit of openness to innovation. If this attitude is remembered by all involved (and particularly
by Classification Societies as key interpreters and
rule implementers through published (SC223) unified
interpretations (IACS, 2010), the new rules shall be
seen as an opportunity by yards and coating manufacturers and not as a menace. Instead of competing only
by lowering prices and squeezing the experience curve
of matured processes, against competitors equally proficient in doing exactly the same, innovative coating
manufacturers and shipyards should focus in developing and getting approved solutions able to cope with
the PSPC performance target in the most effective and
efficient way. This applies both to alternative systems as defined by IMO PSPC (non-epoxy or not
applied according to Table 1) or to methods that need
to be screened beyond the literal wording of PSPCs
Table 1 definitions but that can ultimately be accepted
as according to Table 1 requirements.
The PSPC text also sets when and how alternative or equivalent products and/or methods can
be accepted. Recognizing such openness is critical
to assure the coherence with the above mentioned
innovation-friendly spirit. The following sections
show some examples.
2.2
Equivalent testing methods
Extracted from PSPC, page 8: Table 1/1.3. Coating pre-qualification test: Epoxy-based systems tested
prior to the date of entry into force of this Standard
in a laboratory by a method corresponding to the test
procedure in annex 1 or equivalent ().
Interpretation: test methods different from the
ones described in the PSPC Appendix 1&2 for coating system pre-qualification can be accepted for PSPC
Type Approval purposes if the Class recognizes the
equivalence. Equivalence determination can be done
by technically recognizing that the submitted testing
methods and results are a good indication of minimum
15 years target durability. ABS, LR and GL accepted
a MIL-PRF 23236 standard (Navsea, 2003), Class7compliant testing report for the ES301 system. The
Class7 compliance means 20-years service life expectation for US Navy, so the mentioned Classification
Tested item
(criteria for both wave
test and condensation
test)
Adhesive failure
Cohesive failure
Cathodic disbondment
from artificial holiday
(wave test only)
Undercutting
from
scribe (wave test only)
Acceptance
criteria for epoxy
systems applied
according to
PSPCs Table 1
Acceptance
criteria for
Alternative
systems
>3.5 MPa
>3.0 MPa
<8 mm
>5 MPa
>5 MPa
<5 mm
<8 mm
<5 mm
NOTE: in all cases the tested panels shall present no blisters,

no rust and no pinholes.
Societies granted the IMO PSPC Type Approval based

in such equivalent testing methods.
2.3 Alternative systems
Extracted from PSPC
Page 11: 8.1All systems that are not an epoxy-based
system applied according to table 1 of this Standard
are defined as an alternative system.
Page 17 (Appendix 1) & 20 (Appendix 2):
3.4 Alternative systems not necessarily epoxy-based
and/or not necessarily applied according to table 1 of
this Standard shall satisfy the criteria for alternative
systems as indicated in the table above.
Interpretation: the above text clearly states that
PSPC can accept coating systems and methods not
necessarily following the PSPC Table 1 requirements.
If such a system is not applied according to PSPC
Table 1 it will be considered an Alternative System.
The PSPC sets the way such Alternative Systems can
be approved, by using the same prescribed testing
methods (wave tank tests and condensation chamber
for six months, see Appendix 1&2 of PSPC for details)
under different (more stringent) acceptance criteria for
adhesion, undercutting from scribe and cathodic disbondment. Both levels of acceptance are presented in
the Table 1 of this article. It is arguable why PSPC
makes the life easier for epoxy systems applied according to Table 1 of PSPC, when compared to Alternative
Systems. But at least the room for innovation is there.
2.4 Example of an alternative system
The openness to innovation is there but sometimes the
door is only slightly open. This example points out
how current PSPC text can make the life difficult to
alternative systems.
Extracted from PSPC, page 9: Table 1/2.3. Shop
Primer: Zinc containing inhibitor free zinc silicate
based or equivalent.
Interpretation: Table 1 clearly requires a zinccontaining shop primer, so any system being approved
over a zinc-free shop primer should be classified
262
colour, etc, but the photographs can still be used

to give an indication of the preparation grade.
One issue needs to be specially addressed. Detailed
as it is, PSPCs Table 1 is nevertheless unable to
cover all details involved in the application. It does
define the surface treatment grade to be reached (Sa
2.5 or Sa 2) at secondary surface treatment phase
(block stage) but does not elaborate on the specific
visual appearance of the surface immediately before
the paint application. As allowed by PSPC Table
1, the retained shop primer can be cleaned using
high-pressure water washing, for example, and this
surely impacts the build-up of flash rust on the adjacent Sa 2.5-treated areas, even if such grade was
obtained using abrasive blasting. If water jetting is
used as a generic surface preparation method, the
flash rust build up over the Sa 2.5 or Sa 2 original steel surface grade will also occurs with time.
Approved coating systems should clearly state their
tolerance grade in terms of surface visual appearance at the moment of paint application and the
testing involved in the approval should be performed
under such conditions (namely allowed flash rust
grade or dampness).
The conclusion is that a coating system to be applied
at block stage over water jetted surfaces does not
need to be scrutinized has an Alternative System but
shall nevertheless be tested over bare metal prepared
by this particular surface treatment method and
presenting the surface conditions to the maximum
extent of the paint system tolerance.
as alternative system and required to cope with the

respective acceptance criteria (more stringent). This
may seem unfair, as zinc does play a key role in scribe
and cathodic disbondment results. So a system based
on a zinc-free shop primer will face a double challenge: more stringent acceptance criteria and the lack
of the zinc-effect creating an early barrier to corrosion
spread at the scribe.
The acceptance of the shop primer is important:
lacking such approval will require removing at least
70% of intact shop primer, as prescribed by PSPC
Table 1. If the shop primer is approved it can be
retained: if the complete coating system comprising
epoxy-based main coating and shop primer has passed
a pre-qualification certified by test procedures in 1.3,
intact shop primer may be retained provided the same
epoxy coating system is used.The retained shop primer
shall be cleaned by sweep blasting, high-pressure water
washing or equivalent method (italic excerpt extracted
from PSPC).
2.5
Example of a Table 1-compliant innovation:

water jetting
Not all innovations will require the Alternative System

compliance path. In this example the compatibility of
water jetting with PSPCs Table 1 is justified, when
used as secondary surface preparation method.
The PSPC Table 1, Section 3.2 defines the required
secondary surface preparation at block stage. The
required cleanliness grades are Sa 2.5 over damaged
areas and welds or Sa 2 to remove the shop primer if
not approved. In both cases the reference standard is
the ISO 8501-1:1988/Suppl:1994. The reference standards are mentioned as foot notes in PSPC Table 1.
IACS PR34 considers these foot notes mandatory.
Earlier interpretations of this PSPC section tended
to conclude that PSPC Table 1 compliance would
imply the use of dry abrasive blasting methods for secondary surface preparation. After an enlarged debate
for more than one year now, it is becoming clear this
is not the case. Lets have a closer look:
ISO 8501-1:1988/Suppl:1994: Preparation of steel

substrate before application of paints and related
products Visual assessment of surface cleanliness.
This standard sets the degree of cleanliness to be
reached, NOT the method to get it. The text of this
standard does not mention any particular method
to reach it. It uses the expression blast cleaning,
that sometimes is wrongly associated to abrasive
blasting.
ISO 12944-4 Section 6.2.3. does provide a definition for blast cleaning. Water jetting is clearly stated
as a blast cleaning method, together with other dry
and wet blasting methods.
Can a Sa 2.5 or a Sa 2 be reached using water jetting?
The answer is yes: the mandatory ISO 8501:1 itself,
under Section 6 Photographs at line 4, clearly
states other methods, such as wet-blast cleaning,
produce surfaces that may differ in appearance,
The next section elaborates further on the water jetting

PSPC compliance issues.
2.6 Water jetting Q&A
The author has been debating the use of water jetting at ship new building with key players involved
with PSPC, from owners and shipyards to classification societies, since 2006. The following Questions
and Answers section is an attempt to present the main
conclusions of the debate, in the authors perspective.
Q: If water jetting is accepted as PSPC Table 1
compliant, does it mean that all PSPC approved coating systems are able to be applied over water jetted
surfaces at block stage?
A: No. Is the authors opinion that each individual
coating system as to be specifically tested over water
jetted surfaces (see below) in order to be type approved
together with this surface preparation method. Water
jetting compatibility check is needed, the same way
prescribed by the PSPC and PR34 for shop primers:
individual testing over required shop primer is needed
for each coating system, even if the system was already
approved over other shop primer or over bare metal.
The good news is that recognizing the PSPC Table 1
compliance of water jetting will enable the test acceptance criteria to be the normal ones, not the more
stringent criteria for alternative systems.
Q: If testing the coating system over water jetting
is needed, how should it be performed?
263
A: Simply following the PSPC testing indications

and aim for the acceptance criteria for epoxy systems applied according to Table 1 (unless the coating
system is not epoxy based). A special care has to be
taken in to consideration, though: the paint application
shall be performed over water jetted bare metal showing the flash rust and dampness conditions the coating
manufacturer claims to tolerate. The typical application over a shop primer is not possible here, as it would
not expose the paint to the conditions created by water
jetting to bare metal. PR34 does allow for approval of
coating systems applied over bare metal. The primary
surface preparation should be performed exactly as
stated in the PSPC test requirements (abrasive blasted
to Sa 2.5, providing a profile 30 to 75 microns), followed by two months weathering of the exposed bare
metal (instead of identical period weathering of the
shop primer). After weathering, the rusted bare metal
test plates shall be water jetted to the required Sa 2.5
grade of cleanliness. In the moment the paint is applied,
the surface aspect should be assessed and its condition
recorded in terms of flash rust or other specific condition, eventually using visual standard guides like SSPC
Vis 4 (SSPC, 2003).
Q: What about profile requirements?
A: Such requirement only applies mandatorily to the
primary surface preparation (that will always be performed using abrasive blasting, normally in auto-blast
chambers). PSPC Table 1 mandatory Foot Note 6 (the
only mention to profile requirements) is mentioned
in the Table 1 Section 2 (Primary Surface Preparation). No such mention or requirement is asked for
the secondary surface treatment. This is understandable, because secondary surface treatment does not
need to create any additional roughness, if the profile was already created previously. So this is not a
concern for water jetting: this blasting method is not
able to increase the profile but it will not destroy the
existing one.
Q: What will happen over the welding seams?
A: the IMO PSPC shows a surprisingly flexible attitude towards welding seams preparation. As
said before, no mandatory profile is required at secondary surface treatment level. The expression 3075
microns, otherwise as recommended by the coating
manufacturer is the only mention to profile requirements, applying to the surface condition after erection.
In the case of ES301 system the manufacturer does
not demand any profile at all. The surface treatment
of the erection joints butts, so critical for the structural integrity of the ship (and a historical source of
paint failures in the past, using traditional coating solutions), did not deserve any special care from PSPC.
These welding seams are allowed to be treated to a
St3 or better grade. No doubt that treating such areas
to a Sa 2.5 bare metal grade using water jetting will
be better than the traditional power tooling. Nevertheless recently data being collected by a Classification
Society in PSPC projects now running found surprisingly good profile roughness at water-jetted welding
seams. It appears that the proper compliance with
PSPC Table 1, section 3.1 (Steel Condition), requiring to remove sharp edges and weld spatters, will be
enough to create the needed anchorage profile. In the
ES301 system case, very high pull-off adhesion values
are being measured over water jetted welding seams
in running projects (see data presented further down).
Q: Is it difficult to get such a water jettingcompatible system passing the IMO PSPC testing
criteria?
A: Yes. The test has to be performed over bare
metal, and this is an important disadvantage to start
with. The sacrificial healing effect of zinc, present
at the scribe when using zinc based shop primers, provides benefit by reducing the undercutting from scribe
tests. This is one aspect where IMO PSPC is not perfect. Other performance-based standards, such as the
NACE SP-0108 for offshore coating systems (NACE,
2008), already took in to account this factor: rust
creepage resistance acceptance criteria is <3.5 mm for
non-zinc-primed coating systems and <1.5 mm for
zinc-primed coating systems. The zinc factor is discounted if not present, allowing for a higher tolerance
in terms of undercutting from the scribe.
A PSPC Type Approved coating system, passing
the testing criteria together with a zinc based shop
primer, can eventually be applied at a project using a
non-approved shop primer (or not submitted to a compatibility test yet). In this case, the approved coating
system will be authorized to be applied over the bare
metal surfaces, as at least 70% of the shop primer will
have to be removed. The same will occur over damaged areas that will be blasted to bare metal anyhow.
The problem is that passing the testing criteria over a
zinc shop primer does not automatically means that the
same coating system would be able to pass the same
criteria over bare metal ... Conclusion: a coating system passing water-jetting compliance testing over bare
metal will surely provide a better indication of performance under the real-life conditions, when compared
with other coating systems approved over zinc shop
primers only.
2.7 Next steps: the edge-retention issue
IMO PSPC requires the steel edges to be treated to a
rounded radius of minimum 2 mm or subjected to three
pass grinding, prior to secondary surface preparation
and painting. Also up to two stripe coats are required.
Both set of actions are mandatory because traditional
solvent-borne epoxy coatings do exhibit a poor edgeretention behaviour, leading to reduced DFT at edges
and over welding seams. Nevertheless, a specific concern with over thickness arising from such procedures
is stated in IMO PSPC Table 1, eventually allowing for
stripe coat scope reduction by the way of the welding
seams to avoid such danger.
The US Navy, in their Core Specification for
paint application on board ships (NAVSEA, 2005),
do reflect the same concern, stating that if using a
non-edge retentive coating: radius all edges, flanges,
cutouts, angles, pipe hangers and foot/hand holds to
264
a minimum radius of 3 mm. But it also states that

such radiusing of edges, flanges, cutouts, angles, pipe
hangers, and foot/hand holds is not required when
using approved edge retentive coatings. Approved
edge-retentive coatings are the ones approved by the
MIL-PRF 23236C standard, Type VII (solvent-free),
Class 7 (ballast tanks 20 years durability). The MIL
criteria asks for a minimum edge-retention ratio of
70% over a very sharp 0.1 mm edge radius.
If the US Navy, which has made huge progress
in terms of coating systems durability on board their
ships thanks to such specifications, is willing to accept
this, why not IMO PSPC? Well, this case needs additional discussion. Theoretically, a coating system can
ultimately be applied without rounding the edges or
reducing the stripe coats scope, meaning that it should
be considered an Alternative Coating system (not
applied according to Table 1). The problem arises from
the fact that current PSPC testing method does not
address this particular need: no specific sharp edge
testing nor edge retention measurement is stipulated.
Given the limitations of IMO PSPC to deal with
this particular issue, the authors suggestion is that such
case should be brought to IACS and addressed in future
UI SC223 editions (IACS, 2010). The ground for this
should be the specific concern of IMO PSPC about
excessive thickness. The edge retentive ES301 system,
for example, presents 74% edge retention ratio (DFT
@ edge/DFT @ flat surfaces) over a 0.1 mm radius
edge, and 100% over a 2 mm radius edge. It is easy to
understand that if PSPC Table 1 indications for edge
grinding and stripe coating application are strictly followed, excessive DFT (at least two-fold the required
DFT) will be observed over edges and welding seams.
To avoid this danger, the authors suggestion is for
IACS to allow Alternative Systems with proven evidence of edge-retention performance as per MIL-PRF
23236C to be applied over steel prepared to a lesser
extent (say 1 mm or two pass grinding only) and using
a reduced number of stripe coats (one stripe coat).
Figure 1. Edge retention.
Figure 2. ES301 application over water jetted surface: the

importance of humidity and flash rust tolerance.
Steel profile is not a critical factor, due to the dualbonding adhesion mechanism.
High edge retention (>70%, ES301S as 2nd coat).
3.2 Value proposition
3 THE ES301 CONCEPT APPROACH

3.1
Product features
ES301 is defined by the following key features:
Solvent free epoxy, polyamine cured

Tolerant to humidity (can be applied over damp
surfaces)
No dew point restrictions
Can be applied with relative humidity over 85%
Tolerant to adherent iron oxides (tolerates WJ2M
flash rust level SSPC Vis 4)
High compatibility with most old coatings (maintenance & repair) or shop primers.
Dual-bonding mechanism: both mechanical &
chemical interactions with steel, oxides and
moisture.
High pull-off adhesion to steel (>12 MPa over
PSPC type of profile).
The ES301 coating system value proposition is to

provide the shipyards with an IMO PSPC compliant
solvent-free coating system, able to be applied over
a zinc-free organic shop primer (PE31) and to accept
water jetting as secondary surface treatment method.
Additionally, use the edge-retentive feature as a way
to reduce the edge grinding and stripe coats scope.
This coating system is able to bring shipbuilders
important advantages in terms of process efficiency
and costs, when compared with the strictly PSPC
ship building process. These benefits are listed below,
are already enjoyed fully in non-PSPC areas and partially (to the extent of the existing type approvals or
individual Class allowances only) in water ballast tank
areas. Once getting IMO PSPC full recognition for
the entire value proposition, the below listed benefits can be enjoyed fully by the shipbuilding industry,
including in water ballast tanks areas. That moment is
much closer now than two or four years ago.
265
Table 2. ES301 key marine approvals and further IMO

PSPC testing results.
Figure 3. Water jetted plates ready to paint: PSPC testing

over flash rusted damp surfaces (results at Table 2).
Straight forward benefits of the ES301 approach:
Zinc-free organic shop primer is potentially cheaper.

Avoiding zinc means less health impact of the
welding fumes.
Avoiding zinc will eliminate a component Hazardous to the environment and Toxic to aquatic
organisms, that may cause long-term adverse effects
in the aquatic environment (R51/53 EC Risk
Phrase Classification). It can be very toxic if the
zinc content reaches 25% w/w (R50/53).
Avoiding zinc also eliminates a source of risk: in
case of zinc primer exposure to an acidic environment, the coating system will fail (zinc is dissolved
instead of generating an insoluble protective oxide).
Water jetting can be cheaper.
Water jetting will not interfere with concurrent
works to the same extent that abrasive blasting does.
Water jetting allows for time savings at the block
stage: no need for previous surface washing nor
abrasive removal after blasting. One tool only is performing washing, blasting and cleaning in one go.
Water jetting is more accurate to blast damaged
spots without damaging adjacent intact areas; no
ricochet damage expected.
The ES301 system can be applied without dew point
restrictions. No need for dehumidifiers at the paint
cell. Added flexibility to around the clock outdoors
painting, with reduced weather constraints.
Cost and time can be saved if 1 mm edge radius or
two pass grinding are accepted when using edge
retentive coating systems, instead of 2 mm/three
passes. The same applies to the reduction from two
to one single stripe coat.
Additional benefits, not PSPC-related are also
offered, as the Health & Safety and Environmental
advantages of the solvent-free nature of the products
and of the water jetting surface preparation.
3.3
System approvals and testing results
3.4
Current use
The presented coating system and associated concept

is already being used at PSPC-ruled projects currently
a. IMO PSPC Type Approvals

ES301K&S two-coat system, total DFT 320 microns,
to be applied over bare metal
ABS Certificate N. 08-HS314072-PDA.
Germanischer Lloyd Certificate N. GL-KORR 1159HH.
Lloyds Register Certificate No. MNDE/2008/2847.
approved to be applied over PE31 zinc-free shop
primer (alternative system as defined by IMO PSPC)
ABS Certificate N. 09-HS538809-PDA
approved to be applied over UHP water jetted bare
metal, abrasive blasted bare metal and PE31 zinc-free
shop primer (alternative system as defined by IMO
PSPC)
DNV Certificate N. K-4211
b. MIL-PRF23236C Qualification
ES301K&S and L&S two-coat systems, total DFT
300 microns, to be applied over bare metal
Type VII (recognized coating without solvent added)
Class 7 (approved for seawater ballast tanks,
20 years service expectation)
Class 15b (approved for application over wet surfaces)
Class 17 (approved for bilges).
c. ES301 TESTING OVER WATER JETTING
Testing performed as per IMO PSPC Appendices 1&2
ES301K&S two coat system, total DFT 320 microns,
applied over water jetted surfaces, damp and flash
rusted in the moment of application (WJ2M, SSPC Vis4).
Figure 3 shows the surface condition prior to paint
application ES301 system passed all criteria for
epoxy systems applied according to Table 1, namely:
Pull off adhesion values averaging 7.7 MPa
Cathodic disbondment 4.9 mm
Undercutting from scribe averaging 6.95 mm
d. ES301 TESTING OVER PE31 SHOP PRIMER
Testing performed as per IMO PSPC Appendices 1&2
ES301 system passed all criteria for Alternative
Systems (not applied according to Table 1), namely:
Pull off adhesion values averaging 7.1 MPa
Cathodic disbondment 0 mm
Undercutting from scribe averaging 4.96 mm
running. PE31 is the shop primer being used. Water

jetting is being used as secondary surface treatment
method in both water ballast tanks and other areas.
The applicable Flag State is accepting this method.
Given the type approvals listed in Table 3, PE31 shop
primer can be retained and UHP water jetting can be
used as secondary surface preparation method at block
stage.
Better than words, the following set of figures (figure 4 to 8) gives a fair idea how the implementation
of such system looks today. All photos have been
taken in Northern Brazil, during the constructions of
a Suez-max tanker christened in May 2010. Other
nine similar ships and other projects (namely offshore
platforms) are currently using this unique coating
approach.
266
Figure 4. Water jetting @ 2450 bar, block stage: outdoors,

around-the-clock, minimal impact in concurrent works.
Figure 5. Detail of surface preparation at block stage: full

blast to Sa 2 (ISO 8501:1) over weld seams and damaged
area, washing over intact approved shop primer.
Figure 6. Testing coating adhesion over water-jetted weld

seams: excellent performance, despite the lack of anchor
profile.
CONCLUSIONS
Ship owners and shipyards perception of the value

of top quality tank coatings for long service life is
Figure 7. Most blasting and painting work taking place

outdoors, with minimal weather restrictions.
changing. The perceived added value of such coatings,

compared with traditional non-tolerant solvent-based
solutions, is increasing. If such a coating system,
together with the performance added value, can be
applied with reduced impact on the cost of steel and
surface preparation and application delays, it may
constitute a reasonable agreement opportunity for
both owners and yards interested in answering the
current challenges and avoid the potential dangers of
conflicting interests.
Despite the focus of this article on the coating specifications for tanks at new building stage, a straightforward additional conclusion should be highlighted: the
high-performance + tolerance features joined together
in ES301 is also helping the adoption of this coating system as cost and performance effective tool for
ballast and cargo tank refurbishment. The advantages
are especially significant for dry-docking repairs using
hydroblasting or for voyage repair when a fair extension of service life of the ship (>10 years) is targeted.
A final mention should be made regarding
the cost of a solvent-free, edge-retentive and
surface + moisture tolerant system as ES301. The
pre-PSPC moment (2005) is chosen. Looking at the
specific case of a 320 microns system for 15 years
service on ballast tanks, such tolerant system cost
(material only) was about 10 US$/m2 . The cost of
a conventional good quality solvent-borne modified
epoxy system for the same application was about 6
US$/m2 (Lodhi, 2005). Taking into consideration the
high cost (as explained above) of coping with IMO
PSPC using a conventional coating system (extra edge
grinding, secondary surface preparation, extra stripe
coat and the correspondent additional delays), the gap
of 4 US$/m2 looks quite attractive if it allows the adoption of the IMO PSPC using UHP water jetting and a
zinc-free shop-primer.
The debate is ongoing and its outcome is of significant importance for the way IMO PSPC will be
remembered in the future: the catalyst for innovation it claims to encourage or a blocking piece of
paper in the way of the shipping and coating industries
competitiveness?
267
ACKNOWLEDGEMENTS
The author wish to thanks Mr. Edward Jansen from
ABS and Dr. Raouf Kattan from Safinah for reviewing
the original article presented at the Royal Institute of
Naval Architects in London (RINA, 2010] and sharing with the author a common interest in making the
IMO PSPC a practical tool to serve the marine industry
competitiveness.
REFERENCES
Azevedo, J. 2006. A New Approach For Ballast & Cargo
Tank Coating: A Solvent-Free And Humidity Tolerant
Epoxy System With Edge-Retentive Properties. The
Royal Institution of Naval Architects (RINA) Conference
on Advanced Materials and Coatings, London.
Azevedo, J., 2008. The PSPC challenges: why an innovative coating technology is needed, Marine Coatings
Conference at SMM, Hamburg.
Azevedo, J., 2010a. Innovation Achieving IMO PSPC Compliance: A Game-Changer Example. The Royal Institution of Naval Architects (RINA) Conference on Marine
Coatings, London, February 2010.
Azevedo, J., 2010b. Innovation Achieving IMO PSPC Compliance: A Game-Changer Reality. Marine Coatings
Conference at SMM, Hamburg.
IACS, 2008. UI SC223 For Application of SOLAS Regulation II-1/3-2Performance Standard for Protective Coatings
(PSPC) for Dedicated Seawater Ballast Tanks in All Types
of Ships and Double-side Skin Spaces of Bulk Carriers,
adopted by Resolution MSC.215(82), Revision 1, July
2010
International Maritime Organization, 2006. Resolution
MSC.215(82), Performance Standard For Protective Coatings For Dedicated Seawater Ballast Tanks In All Types
Of Ships And Double-Side Skin Spaces Of Bulk Carriers
(PSPC), adopted on 8 December 2006.
Intertanko, 2002. Tanker Structure Cooperative Forums
Guidelines for Ballast Tank Coatings Systems and
Surface Preparation, Witherbys Publishing.
Lodhi, A., 2005. Double Hull Tankers Ballast Tank Maintenance Experience Feed-Back, The 22nd DNV Technical
Committee Meeting, Dubai, 7th December 2005.
NACE, 2008. SP-0108 Corrosion Control of Offshore
Structures by Protective Coatings
NAVSEA US Navys Naval Sea Systems Command, 2003.
US Military standard MIL-PRF 23236C Performance
Specification Coating Systems for Ship Structures,
issued 12 August 2003.
NAVSEA US Navys Naval Sea Systems Command, 2005.
Preservation Process Instruction (PPI) CORE Specification, 63101-000 (REV 18), October 07, 2005
SSPC, 2003. Vis 4 NACE N.7, Guide and visual reference
photographs for steel cleaned by water jetting.
268

Analysis of the fluid-structure interaction of a composite motor yacht

H. Bagbanci, N. Baykut, A.S. Oktem & C. Guedes Soares
ABSTRACT: In this paper a fluid-structure interaction problem is solved by utilizing a computational fluid
dynamics (CFD) code to solve three-dimensional (3D) Reynolds Averaged Navier-Stokes Equations (RANSE) in
conjunction with a finite element program to perform a strength analysis of a composite motor yacht, performing a
fluid-structure interaction study. Hydrodynamic and hydrostatic loads are obtained in the first step. These loads
are then used to perform a strength analysis of the hull structure. A composite boat was examined for the
velocities of 0, 4, 8, 12, 16 knots to obtain the pressure loads under different working conditions. Hydrodynamic
and hydrostatic load magnitudes were found to be close in magnitude for the velocities of 0, 4 and 8 knots, but
the values for the velocities of 12 and 16 knots are different from the above counterparts. The static strength
analysis of the structure is also performed by the finite element software for the abovementioned velocities and
the results are compared.
INTRODUCTION
Fluid-structure interaction problems in all fields of

engineering as well as in naval architecture gained
importance, since ships represent a classical situation
in which these approaches are required. There is an
increasing interest in the marine industry to use composites to improve the hydrodynamic and structural
performance of naval structures. Composite materials
have high strength-to-weight and stiffness-to-weight
ratios, and the fibre orientations can be exploited to
tailor the structural deformation to reduce the load and
stress variations by automatically adjusting the shape
of the structure.
Recent advances in composite materials make it
possible to build naval ships up to 100 m, and in the
near future ships can be built with more than 150 m in
length with advanced composites (Mouritiz, 2001).
There are many practical engineering problems that
cannot lead to exact solutions. In contrast to analytical
solutions, which show the exact behaviour of a system,
numerical solutions approximate the exact solutions
only at discrete points. The first step of a numerical
procedure is discretization. This process divides the
medium of interest into a number of small sub-regions
and nodes. Among numerical methods, the finite element method is a robust numerical procedure that
can be applied to obtain solutions for a wide variety
of problems in engineering. Steady, transient, linear
or nonlinear problems in stress analysis, fluid flow,
heat transfer, and electromagnetism problems may be
analysed with the help of finite element methods.
The relation between the performance and weight
is crucial for high speed crafts. Ojeda (2003) studied a slamming problem for hogging and sagging
conditions according to the DNV HSLC (1999) rules

by employing a static linear approach using finite
element analysis. Chen et al. (2002) presented the analyses of the hull structure of a fishing vessel made of
composite material. The finite element method is utilized for the assessment of the entire ship hull based
on global and local analyses. A good example of a
complex-three dimensional finite element approach
to analyse the structural dynamic behaviour of a fast
patrol boat can be found in Santos et al. (2009).
A three dimensional computational fluid dynamics
method has been applied to solve the motions of sailing yachts (Azcueta, 2002), planning craft (Azcueta,
2003) and ships in waves (Sato, 1999), and good results
have been obtained. Ciortan and Guedes Soares (2007)
simulated the airflow around yacht sails using a CFD
code and obtained results close to experimental tests
carried out in a wind tunnel.
Fluidstructure interaction (FSI) problems are of
great importance to many fields in engineering and
applied sciences. Young (2008) studied the fluid
structure interaction analysis of flexible composite
marine propellers. The numerical solution of viscous
free surface problems was pioneered by Harlow and
Welch who developed the Marker and Cell (MAC)
method. A variant of this approach is the volume of
fluid (VOF) method developed by Hirt and Nichols.
Many models and computational approaches have
been developed in the area of fluidstructure interaction. Partitioned solution strategies have a wide variability from the loose coupling procedure of Felippa
et al. (1997) to the strong coupling algorithms of Kalro
and Tezduyar (2000), Le Tallec and Mouro (2001).
An approach to the fluidstructure interaction with
surface waves can be found in Onate and Garcia
269
Table 1. The principal parameters of the boat.

Parameter
Dimension/details
Length (Loa)
Breath (B)
Depth (D)
Draft (T)
22.00 m
5.48 m
4.00 m
1.26 m
Table 2.
Material properties.
Properties
997 kg/m3
1.184 kg/m3
8.9 104 Pa s
9.0 105 Pa s
Density of water (w)

Density of air (a)
Viscosity of water (w)
Viscosity of air (a)
Figure 1. Star CCM+ meshed structure.
Table 3. Velocity of water and air.

Current Velocity (knot)
Air Velocity (knot)
0
0
4
4
8
8
12
12
16
16
(2001a,b). Another interesting approach for fluid

object and fluidstructure interaction with free surfaces and two-fluid flows that combines attractive
features of interface tracking and interface capturing
schemes, is the mixed interface-tracking/interfacecapturing technique introduced by Tezduyar (2001).
Recently, good examples of free surface flow and
coupled fluid-structure interaction problems were presented by Ahmed and Guedes Soares (2009) and Paik
et al. (2009), respectively. Luo et al. (2011) studied the
water entry problem of one wedge, while Luo et al.
(2012) analysed the hydroelastic response of a wedge
impacting the water.
The current problem is considered a good example of FSI multi physic problem that deal with the
interaction of a ship structure with the surrounding
water. It is not a critical problem to analyse in naval
architecture, as there are no wave effects, but the aim
of the present study is to exploit the possibility of
interfacing a CFD and a finite element code. For that
purpose, a simplified fluid-structure problem is considered and a fluid dynamics software (Star CCM+) is
used in conjunction with a structural analysis program
(ANSYS).
The main emphasis of this study is to develop the
interface between the two programs aiming at producing a FSI analysis than to conduct a specific analysis
of an important problem in a specific context.
DESCRIPTION OF THE PROBLEM
A 22 meters composite motor yacht is used in the analysis. The principal parameters of the boat, properties
of water and air used in the analysis and velocities of
water and air are given inTables 1, 2 and 3, respectively.
CFD ANALYSIS
The first step considered is the CFD simulation, with

the aim to obtain hydrodynamic and hydrostatic loads
at each velocity. The simulation is run for velocities of
0, 4, 8, 12 and 16 knots. The process of each simulation
is fixed to 20 seconds, a time step length is 0.02s.
Simulation is performed for deep water. The model is
meshed by a surface remesher and trimmer and totally
28,222 faces are generated for half of the hull. The
mesh of the model is shown Figure 1.
Eulerian multiphase, volume of fluid (VOF),
k-epsilon turbulence and (VOF) waves methods are
used in the present analysis. Volume of fluid (VOF) is
a simple multi-phase model that is well suited for simulating flows of several immiscible fluids on numerical
grids capable of resolving the interface between the
mixtures phases. VOF waves can be used in conjunction with the volume of fluid (VOF) multiphase model
in STAR-CCM+. They are typically used with the sixDOF motion model for marine applications. The body
or system can rotate or translate in six degrees, it allows
pitch, roll, yaw, heave, surge and sway motions to the
vessel. These motions occur along the x, y, z axis and x,
y, z directions. Star CCM+ six-DOF solver is using the
initial orientation of the body relative to the laboratory
coordinate system. It is possible to define boundary
conditions that will simulate waves on the free surface
using the VOF wave physics model in Star CCM+.
The vessel is simulated on calm water; therefore FlatVofWave model is chosen to simulate this
condition (CD Adapco User Tutorial, 2010).
Star CCM+ uses interface capturing method by the
RANSE and the VOF methods. Reynolds Averaged
Navier-Stokes equation (RANSE) provides the solution of real flow problems. RANSE solver of the CFD
program uses the finite volume method (FVM) for
simulations, while multiphase simulations for free surface use the volume of friction method. Viscous flow
includes continuity, momentum, and volume fraction
270
Table 4.
Section locations.
Station name
Location (m)
S1
1
S2
3.4
S3
6.3
S4
9.8
S5
13.5
S6
15.4
S7
17.1
Figure 2. Control sections.
and turbulence model equations. The governing equation for mass and momentum equations are given as
follows:
Figure 3a. Total pressure distribution for 0 knot.
where, v is the velocity vector in the Cartesian coordinate system, p is the static pressure, is the stress
tensor, is the molecular viscosity, and I is the unit
tensor.
Figure 3b. Total pressure distribution for 16 knot.
3.1 VOF method

Each cell of mesh uses only one value for each dependent variable fluid state. The variables can be defined
by a function (q) (where q represents the fraction of
volume; q = 1, 2, ). The value of this variable is one
at any point in the fluid
Basically, response to a cell full of fluid, while a zero

value indicates that the cell is empty, so the cells with
values between 0 and 1. For the qth liquid, troughs
of the density are as follows:
Figure 4. Hydrodynamic and hydrostatic loads for different
velocities.
Tracking of the interfaces is done through the solution

of continuity equation for the volume fraction of the
phases. For the qth phase;
Hydrodynamic and hydrostatic loads are obtained in

the first step. In order to obtain the pressure distributions throughout the hull, imaginary control sections
are created along the hull length. These control sections which locations given in Table 4, are then used
to fit polynomial functions of pressure distributions.
3.2 Fluid dynamic analysis and discussion

At the end of each simulation, pressures on the hull are
obtained. Pressures obtained for zero and 16 knots of
the vessel are given in Figures 3a and 3b, respectively.
In Figure 3a, the vessel has zero velocity, and the
hydrostatic load is obtained as 11k Pascal on the bottom. For the velocity of 16 knots the total maximum
load (hydrostatic + hydrodynamic) is 22k Pascal.
Hydrostatic and hydrodynamic loads for the velocities considered in this work are presented in Figure 4.
The time history of drag force on the hull is also
examined for velocities 0, 12 and 16 knots of the vessel. Initially, the force is large on the hull, but after the
271
Figure 5. Time history of drag force for each speed.

Figure 7.1. Pressure distributions on control section for
0 knot
Figure 6. Time history of lift force for each speed.
first phase the drag force is stable and they are given
in Figures 5 and 6.
The reference time is important to obtain the maximum pressure distribution on the hull. The pressure
distribution is used for the finite element analysis
according to reference time. In the first simulation,
the vessel has no velocity; therefore the lift force is
stable. The reference time can be chosen for any time
to see the pressure distribution on the hull after the
first second.
For the second simulation, the vessel has a velocity of 12 knots and the lift force is sinusoidal. The
maximum pressure distribution is obtained at the 17th
second.
For the third simulation, the vessel has a velocity
of 16 knots and the lift force is also sinusoidal. The
maximum pressure distribution is obtained at the 9th
second. Lift forces reach the maximum at these time
steps. Simulations are run until the times reached the
above mentioned and the pressure distributions are
obtained at the control sections. Finally, polynomial
functions are obtained to give pressure distributions
and they are shown in Figures 7.
Polynomial functions are given in Table 6 on each
control section for the speed of 16 knots.

12 knots.

16 knots.
FINITE ELEMENT ANALYSIS
The commercially available ANSYS finite element

software is used for structural analysis of the
272
Table 4.
FE model composite layer properties.
Bottom
Side
Deck
Bulkheads
Table 5.
Table 6.
Layer Angels
Thickness (m)
Location
(0/90/0/90)s
(0/90/45)s
(0/90/0/90/0)
(0/90/0/90)
0.020 m
0.012 m
0.0075 m
0.006 m
At 1st mt
E-Glass/Vinly Ester Tow (Vf = 57%)
EX (GPa)
EY (GPa)
EZ (GPa)
GXY (GPa)
GYZ (GPa)
GXZ (GPa)
XY
YZ
XZ
4.28 1010
1.222 1010
1.222 1010
4.77 109
4.87 109
4.77 109
0.23
0.25
0.23
composite boat. The element type is chosen as SHELL

99 which is convenient for the analysis of layered composite materials. This element is an 8-node, 3-D shell,
layered element with six degrees of freedom at each
node: translation in the nodal x, y and z directions and
rotations about the nodal x, y and z axes. It is designed
to model thin to moderately thick plate and shell
structures with a side-to-thickness ratio of roughly
10 or greater. The SHELL 99 element allows a total
of 250 uniform-thickness layers.
The vessel is divided into four main parts and the
lay-up and the thicknesses of each part are given in
Table 4.
Material properties of the E-Glass composite are
given in Table 5 (Goyal, 2007).
4.1
Mesh generation
The geometry of the model, shown in Figures 3, is created by defining key points, based on the Star CCM+
model. Lines are then defined by key points and areas
are created by lines. Attributes (element type, real constants, and layer angels) are assigned to the areas based
on the information above. Following these steps, the
mesh is generated using the free meshing facility of
the pre-processor module of ANSYS 12.1.
4.2
Boundary conditions
The pressure distribution on the hull is obtained from

Star CCM+ as discussed in Section 3.2. The equations
of the pressure distribution are defined as functions
by using parameters moduled in the interface of the
structural software. The pressure functions are then
Polynominal fuctions of pressure distribution
P = 312.77z6 2122.6z5 + 5377.7z4

6216.9z3 + 3159.3z2 543.23z 0.3899
At 3.4th mt
P = 18.29z6 119.36z5 + 1241z4
3257.9z3 + 3506.8z2 1494.2z + 144.43
At 6.3th mt. P = 349.53z6 + 1263.6z5 + 189.68z4
4820.2z3 + 4793.8z2 637.23z 308.99
At 9.8th mt. P = 426.48z6 + 1891.6z5 1262.9z4
4225.3z3 + 6071.9z2 1584.9z 278.18
At 13.5th mt. P = 115.99z6 + 565.16z5 374.72z4
1897.2z3 + 3749.3z2 2773.7z + 970.21
At 17.1st mt. P = 710z6 4184.2z5 + 6659.7z4
+ 294.87z3 3357.7z2 5323.4z + 4144.4
Composite material properties.
Properties
Polynominal functions at 16 knot.
Figure 8. Weight distribution of vessel.
inserted onto the boat surface. The pressure distributions are constructed according to five different speeds
to obtain five different hydrodynamic pressures.
The boundary conditions are applied as symmetry conditions to lines located at the centre line. The
vessel is fixed in the x (length), y (width), and z
(height) direction on the aft bulkhead and on the y
and the z direction on the fore bulkhead with, symmetry conditions applied on the centre line. The rotational
condition is free for each direction. The system is then
solved to obtain the deflections and stresses due to the
weight distribution of vessel. Weight and force distribution of the vessel is shown respectively on figures 8
and 9.
4.3 Structural analysis and discussion
A linear static analysis is performed and results are
obtained by using a general postprocessor.
Due to its relative simplicity, the maximum stress
criterion is selected to check the safety, although TaiHill or Tsai-Wu failure criteria may be used for composite materials. The maximum stress failure criterion
assumes that failure occurs whenever any component
of stress attains its limiting value, independent of the
values of all other components of stress. (Herakovich,
1998).
273
Table 10.
Stress results of 16 knots.
Stresses
%50 of U. Strength (MPa)
ANSYS results
(MPa)
Sx
Sy
Sz
Sxy
Syz
Sxz
1021
54
54
60.8
60.8
60.8
6.2
7.82
5.57
2.01
3.62
2.81
Figure 9. Force distribution of vessel.

Table 7.
Composite ultimate stress properties.

E-Glass/Vinly Ester Tow (Vf = 57%)
SX
SY = SZ
SXY = SXZ
SYZ
Table 8.
(MPa)
(MPa)
(MPa)
(MPa)
2042
108.2
121.6
121.6
Stress results of 0 knot.
Stresses
ANSYS results
(MPa)
Sx
Sy
Sz
Sxy
Syz
Sxz
1021
54
54
60.8
60.8
60.8
5.63
8.1
5.64
1.64
3.68
1.5
Table 9.
Figure 10. Summation of displacement at 16 knots

condition.
Stress results of 12 knots.
Stresses
ANSYS results
(MPa)
Sx
Sy
Sz
Sxy
Syz
Sxz
1021
54
54
60.8
60.8
60.8
7.51
8.97
5.97
2.14
3.67
2.05
Figure 11. x stress at 16 knots condition.
analysis are given in Tables 8, 9 and 10, respectively.

As expected, the highest values of stresses are obtained
at 16 knots condition and for the sake of brevity,
results are provided only for the 16 knots conditions.
Finding composite properties in three directions is

quite difficult. Hence, a simple approach is assumed
by taking the yield stresses as half of the ultimate stress
properties of the material. The ultimate stress values
of the composite are given in Table 7 (Goyal, 2007).
The values obtained by taking half of the ultimate
values and the results of the static analysis for 0, 12 and
16 knots are given in Tables 8, 9 and 10, respectively.
Figures 10 to 12 present the total displacement, and
the direct and Von Misses stresses. Results of the static
CONCLUSIONS
The FluidStructure I (FSI) approach is considered

for the strength analysis of a high speed composite
boat in motion in still water.
Based on the CFD analysis, the hydrodynamic and
hydrostatic pressures are obtained. Reference time is
used to obtain the maximum pressure distribution on
the hull. Pressure distribution on control sections are
274
Figure 12. Von Misses Stress at 16 knots condition.
generated as polynomial functions and they are used

for the finite element analysis according to the reference time. On structural analysis, the SHELL99
3-D shell element is used to model composite structures. MASS21 nodal mass element and density of
material is used for modelling distribution of the ship
weight. The lay-up angles and the layer thicknesses are
optimized by utilizing the failure criterion.
The work allowed the conclusion that it is possible
and it is a relatively streamlined procedure to plan FSI
interaction problems by interfacing STAR CCM+ and
ANSYS.
ACKNOWLEDGEMENTS
The third author has been financed by the Portuguese
Foundation of Science and Technology under the
contract number SFRH/BPD/47687/2008.
REFERENCES
Ahmed, Y., Guedes Soares, C. 2009. Simulation of Free
Surface Flow around a VLCC Hull using Viscous and
Potential Flow Methods. Ocean Engineering, Vol. 36,
pp. 691696.
Azcueta, R. 2002. RANSE simulations for sailing yachts
including dynamic sinkage & trim and unsteady motions
in waves. High Performance Yacht Design Conference,
Auckland.
Azcueta, R. 2003. Steady and unsteady RANSE simulations for planing craft in the 7th Conference on Fast Sea
Transportation, FAST03, Ischia, Italy.
Ansys inc., User Manual, 2009.
Belytschko T, Geers TL, editors. 1977. Computational methods for fluidstructure interaction problems. AMD, vol.
26. NewYork:American Society of Mechanical Engineers,
pp. 95124.
CD Adapco User Tutorial, 2010.
Chen, N. Z., Garbatov, Y., Guedes Soares, C. 2002. Finite
Element Analysis of a Fishing Vessel Built of Composite Materials. Proceedings of the 6th International
Conference on Marine Science and Technology.
Ciortan, C., Guedes Soares, C. 2007 Computational study of
sail performance in upwind condition. Ocean Engineering, VOl. 34, pp. 21982206.
Felippa C. A., Park KC, DeRuntz JA. 1977. Stabilization of

staggered solution procedures for fluidstructure interaction analysis. In: Belytschko T, Geers TL, editors.
Computational methods for fluidstructure interaction
problems. AMD, vol. 26. New York: American Society of
Mechanical Engineers, pp. 95124.
Goyal, D. 2007. Analysis of Linear elasticity and nonlinearity Due to Plasticity and Material Damage in Woven
and Biaxial braided Composited. PhD. thesis, Texas A&M
University, College Station, TX.
Hirt, C.W. & Nichols, B.D., 1981. Volume of fluid (VOF)
method for the dynamics of free boundaries, Journal of
Computational physics, Vol. 39, pp. 201225.
Herakovich, C. T. 1998. Mechanics of fibrous composites,
New York, John Wiley & Sons, Inc.
Harlow F. H, Welch JE. 1965. Numerical calculation of
time-dependent viscous incompressible flow of fluid
with free surface. Int Phys Fluids 8:21822189.
Harlow F. H., Welch JE. 1966. Numerical study of largeamplitude free surface motions. Int Phys Fluids, Vol. 9,
pp. 84251.
Kwang-Jun, P., Pablo, M. C., Donghee, L., Kevin, M. 2009.
Strongly coupled fluidstructure interaction method for
structural loads on surface ships. Ocean Engineering,
Vol. 36, pp. 13461357.
Le Tallec P, Mouro J. 2001. Fluid structure interaction with
large structural displacements. Comput Methods Appl
Mech Engrg, Vol. 190, pp. 303968.
Luo, HB.; Wang, H., and Guedes Soares, C. 2012. Numerical and experimental study of hydrodynamic impact and
elastic response for one free-drop wedge with stiffened
panels. Ocean Engineering. Vol. 40, pp. 114.
Luo, H.B., Wang S., Guedes Soares C., 2011. Numerical prediction of slamming loads on rigid wedge for water entry
problem by an explicit finite element method. Advances
in Marine Structures, Guedes Soares, C, and Fricke, W.,
(Eds), Taylor & Francis, UK, pp. 4147.
Mouritz, A. P., Gellert, E., Burchill, P., Challis, K. 2001
Review of advanced composite structures for naval ships
and submarines. Composite structures, Vol. 53, pp. 2141.
Onate E, Garcia J. 2001a. Finite element analysis of
incompressible flows with free surface waves using a
finite calculus formulation. In: ECCOMAS Computational Fluid Dynamic Conference Swansea, Wales, UK.
Onate E, Garcia J. 200b. A finite element method for fluid
structure interaction with surface waves using a finite
calculus formulation. Comput Methods Appl Mech Engrg,
Vol. 191, pp. 63560.
Ojeda, R., Prusty, B.G., Salas, M. 2003. Finite element
investigation on the static response of a composite catamaran under slamming loads, Ocean Engineering, Vol.
31, pp. 901929.
Santos, F.M., Temarel, P., Guedes Soares, C. 2009a. Modal
analysis of a fast patrol boat made of composite material.
Ocean Engineering, Vol. 36, pp. 179192.
Santos, F. M., Temarel, P. A., and Guedes Soares, C. 2009b.
On the Limitations of Two and Three-dimensional Linear
Hydroelasticity Analyses Applied to a Fast Patrol Boat.
Journal of Engineering for the Maritime Environment.
Vol. 223(3), pp. 457478.
Tezduyar, T.E. 2001. Finite element methods for flow
problems with moving boundaries and interfaces. Arch
Comput Methods Engrg, Vol. 8, pp. 83130.
Young, Y.L. 2008. Fluidstructure interaction analysis of
flexible composite marine propellers Journal of Fluids
and Structures, Vol. 24, pp. 799818.
275

Structural reinforcement study of a floating dock mooring

structure using FEM software
T. Carvalho, E. Mota & A. Correia Rodrigues
Lisnave Estaleiros Navais, Setbal, Portugal
ABSTRACT: The aim of this study is to contribute for a structural solution by using finite element software
to design a steel work repair on floating dock mooring structure. An historical of damage on certain structural
elements of the dock mooring structure leads to the need of evaluation of the structure in more detail. The original
structure is modelled using finite element method software and different simulations are preformed in order to
identify structural strength weaknesses which are compared against structural damage reports. After analyzing
Von-Mises stress results and Principal Stress of original model, two different models are developed and simulated
in fem software. The first model developed is based on existing structure arrangement after several local repairs.
The second model is developed according to original structural arrangement. Both models are simulated for
the same load cases. The development of the second model is done by optimization of structural geometry of
certain elements and new structural elements are added. Afterwards both model stress results are compared and
conclusions are presented.
1
INTRODUCTION
This study intends to analyze the lower support of the

mooring of a floating dock Kumba Ndoffen Juuf
built at 1979, with a lifting capacity of 28000 t which
is operating on the west coast of Africa in Dakar,
Senegal Dakarnave Shipyard. The floating dock
is operated for ship repair since late 90s and ship
repair activity is very intensive by performing several
docking and undockings per month. This intensity of
activity and at the same time the advance age of the
floating dock is resulting in some local damages. The
study intends to understand the behaviour of the structure, considering the static forces, calculated by the
builder, mentioned in the structural mooring arrangement. Finite elements software is used to simulate
several static load conditions and others structural
solutions are developed and simulated. Hydrodynamic
effects were not considered in this study as well as
cyclic forces of other nature.
Meanwhile the structure has been subjected to small
repairs to overcome the reported damages. With this
work is intended to achieve a more definitive repair
solution with help of finite element method software.
1.1 Description of problem
Along life of the floating dock several damages
occurred due to severe weather and some cracks
appeared and propagated through structural elements.
The floating dock is located in the inside edge of CapVert Peninsula in Dakar.The dock as a length over all of
235 m, a breadth outside of 47 m, depth to upperdeck
of 17 m and a maximum draught of 15.7 m.
Figure 1. Mooring Arrangement.
The floating dock has two mooring supports and

initial crack problems were reported mainly in the
north mooring structure on the weld between top plate
guide pipe and the lower support.
Since the dock is operating for Dakarnave from
the end of 90s small damages have ocurred at the
mooring structure that have been repaired locally. In
general, the damages location is in structural areas that
are underwater, so each time a repair is done it is necessary to create dry work conditions to preform the
repair, but it is very difficult to achieve a good work
condition for a steel welding repair.
277
Figure 4. Steel repair on top plate.

Figure 2. Location of initials damages.
Figure 5. Damage report, July 2009.
Figure 3. Initial steel repair.
In 2005 Dakarnave issued a brief damage report to

Lisnave in order to study this case and to contribute
with a better steel repair design. The damage report
shows damages occur at welding connections between
the top plate and the guide pipe and on the top plate
knuckle.
Due to short time available for more accurate analysis, it was decided not only to make the welding repair
but also to replace part of the top plate with a thicker
plate (20 mm instead of 15 mm) and to add three new
brackets connecting the top plate and the guide pipe.
The renewal of the top plate it was taken in order to
restore the original mechanical proprieties of steel in
that location, which has been subjected to successive
welding repairs. The new brackets were installed to
improve the connection between the existing brackets,
top plate and the guide pipe.
At July 2009 new cracks were reported at the new
plate between the top plate and pipe guide.The knuckle
on top plate also was reported to have some cracks. The
report shows this time that cracks begun on the top of
brackets and propagated through brackets, top plate
Figure 6. Damage report, December 2009.
and reached to side girders underneath top plate, as

shown.
The repair was done by making new weldings in
crack locations and new face plates were added to the
brackets.
In December 2009 a new report was issued showing
that a few cracks appeared at the new brackets and a
few on top plate.
278
Table 1.
Description of loads considered.
Wind force
Friction force
Transversal
Longitudinal
Vertical
485 t
200 t
140 t
Table 2.
(N)
Table 3.
Forces applied in the models studied.

F1
90
F2
0
F3
Vertical
F4
45
2376500
1960000
1372000
2178227
Case tests.
F1
Case 1
Case 2
Case 3
F2
F3
F4
X
X
X
Figure 7. Case studies.
2.1
2.2 Boundary conditions
DESCRIPTION OF LOADS, BOUNDARY

CONDITIONS AND STRUCTURAL FINITE
ELEMENT
Loads
For applied forces to the model it was used the wind

force estimated for winds of 150 km/h and friction
force on the pipe that is stated in the floating dock
document DOKK-Mooring Arrangement.
For longitudinal wind force it was considered the
possibility of total force (200 t) to be applied in one
mooring point structure because the two mooring
points are aligned with longitudinal wind force direction and a small difference of distances between the
quay mooring points due to unequal erosion of wooden
battens and the floating dock mooring points can
originate this circumstance of force.
In transverse wind force it was considered the total
force (485 t) divided by two mooring points resulting
in a force of 242.5 t.
Due to friction between the guide pipe and quay
mooring structure it was considered for simulations a
vertical friction force of 140 t for each mooring point.
These forces were combined in order to create three
cases for FEM simulation. Case 1: it is only considered the longitudinal wind force and vertical friction
force. Case 2: it is considered the transverse wind
force and vertical friction force. Case 3: it is considered a wind force for an attack angle of 45 degrees
in relation with longitudinal direction of the floating
dock.
The three cases presented are intended to cover the
predominant strongest winds reported by locals in the
shipyard.
In all case studies and all models presented the

boundary conditions remained the same.
In the lower and upper mooring foundation
the degree of freedom for boundary conditions is
UX = 0, UY = 0, UZ = 0, ROTX = 0, ROTY = 0 and
ROTZ = 0.
2.3 Structural finite element
For these simulations in all models it was employed
the structural finite element SHELL 93 which is well
suitable for plan and curved elements. The element
has six degrees of freedom at each node: translations
in the nodal x, y, and z directions and rotations about
the nodal x, y, and z-axes. The deformation shapes are
quadratic in both in-plane directions. The element has
plasticity, stress stiffening, large deflection, and large
strain capabilities. The element is defined by eight
nodes, four thicknesses, and the orthotropic material
properties.
2.4 Mesh
The mesh used for finite element simulation was
free mesh, however due to the size of structure to
be modeled and to the interest of analyzing in some
detail the lower support structure it was used different
size of element of mesh in different parts of mooring
structure. All lower support structure, side brackets
and initial part of guide pipe the size of mesh elements
are 25 mm. Guide pipe area where combined forces are
applied it was used 100 mm for size of mesh element
and in transition area to lower support structure it was
used 50 mm for size of mesh elements. In upper area
279
Figure 9. 1st Principal Stresses analysis.

Figure 8. Von-Mises stress analysis.
of guide pipe it was used 200 mm for the size of mesh

elements and in the transition area to upper support
structure it was used 100 mm for size of mesh elements. All upper support structure has 50 mm for size
of mesh elements except some structural elements that
was used 25 mm for size of mesh elements.
3
ORIGINAL MOORING STRUCTURE MODEL
The original structure was modelled in a Finite Element Method software according to structural original
drawings. The entire mooring structure was modelled
including top support structure, guide pipe and lower
support, in order to better understand the structure
behaviour when it is submitted to the case studies
presented before.
3.1 Analysis and results of original mooring

structure model
3.1.1 Case 1
In this case, the simulation showed high levels of stress
at the lower support between the top plate and the pipe
guide.
Also when analysing the model for 1st Principal
Stresses the simulation showed high levels of stress.
The lower support structure makes evident a weak
zone between the top plate and the guide pipe. In this
case the simulation also shows some high levels of
stress at the knuckle on the top plate. The side girder
also shows higher levels of stress in comparison with
case 1.
3.1.2 Case 2
When analysing the structure for 1st Principal stresses
the simulation shows high levels of stress on the side
bracket over the side girder.
3.1.3 Case 3
The analysis shows two distinct zones near the same
location with high levels of stress between the top plate
and the guide pipe. Also the side bracket on the top
Figure 11. 1st Principal stress analysis.
plate has high level of stress at the end of toe bracket.

Near the knuckle of top plate there is also high levels
of stress.
The top plate according to the principal stress
analysis has high level of stress at connection with
pipe guide and also at the knuckle of top plate.
280

data of the reported damages, main damages occur on

structural elements welds between top plate and the
pipe guide.
4
MOORING STRUCTURE MODEL 1
The first approach to strength the structure was to

increase the size of central brackets over the top plate
in order to improve connection between the top plate
and the pipe guide. The thickness of side girders
above top plate was increased too. The knuckle on top
plate is eliminated to diminish the stress concentration
around this location.
3.2
4.1 Analysis and results of mooring structure

model 1
Structural condition of original mooring

structure model
The previous analysis shows that the lower support

structure has specific location of high levels of stress
when subjected to the forces considered in this work
and the locations identified are at the top plate connection with pipe guide, bracket connection with top plate,
knuckle at top plate and side girders between top plate
and bottom plate. The analysis is in accordance with
4.1.1 Case 1
In this case the bottom, top plate and side girders
have low level of stress because of the new three big
brackets at centre that contribute to dissipate stress
levels. On the other hand the face plate of brackets has
high levels of stress.
The edge of face plate has local high level of stress
and also at brackets face plate connection with guide
pipe.
281

Again at the bracket face plate there are high levels

of stress located at the edge of plate mainly, but also
in other locations.
MOORING STRUCTURE MODEL 2
From model 1 it was verified that constraining too

much the lower support of the guide pipe it would
increase the stress of certain structural elements.
So the model 2 is re-designed to strengthen the lower
support but to achieve a structure with more flexibility
than the model 1.
The geometrical structural changes comparing to
model 1 are:
connection between top plate and guide pipe is
reinforced with several small brackets;
side girders are reinforced with longer side
brackets.
4.1.2 Case 2
The stress analysis shows high levels of stress at
bracket face plate.
A small area at the bracket face plate shows some
high levels of stress.
4.1.3 Case 3
High levels of stress are experienced for this case
simulation at the bracket face plate.
5.1 Analysis and results of mooring structure

model 2
5.1.1 Case 1
From the simulation for case 1, Von Mises stress
analysis shows that stresses are dissipating well by a
large number of structural elements and there are no
concentrations of high levels of stress.
282
Figure 26. Maximum deflection of guide pipe for case 1.
5.1.3 Case 3
The structural elements present a medium level of
stress in global area of the lower support structure.
The analysis shows a small area at bracket face plate
with high levels of stress but for the rest of the structure
the stress levels are medium.
6
RESULTS COMPARISON
Critical structural elements of models are now compared in terms of deflection and stress.
6.1 Guide pipe deflection
5.1.2 Case 2
For this case the side brackets and side girders are the
most acting structural elements which is confirmed by
the higher average of stress levels.
The simulation shows low levels of stress with in a
large portion of the lower support structure.
The three models are compared in terms of deflection of guide pipe. The simulation shows that original
structure is the most flexible comparing with model
2 and model 1. On the other hand model 1 it is the
most rigid because it has the most strengthen structural
geometry at the lower support.
6.2 Stress comparison
Model 2 it is compared with the original model structure in order to check if structural modifications
283
Table 4.
Maximum deflection of guide pipe for all cases.
(mm)
Case 1
Case 2
Case 3
Original
Model 1
Model 2
27.8
19.2
23.2
23.5
18.5
18.9
25.9
19.3
21.3
6.2.3 Bottom structure Original vs

Model 2 Case 1
Comparing the overall of bottom structure the
analysis shows that the model 2 has a better distribution
of stress than original model.
improves the strengthen of lower support structure of

guide pipe.
6.2.1 Side girder Original vs Model 2 Case 1
The stress analysis shows that the geometric modification from original model to model 2 is benefit
because it clears the stress concentration around the
toe of side bracket. According to damage reports some
cracks begun propagation from bracket toe location to
top plate. The simulation for all other cases shows the
same behaviour.
6.2.2 Top plate Original vs Model 2 Case 1
Stress analysis shows that original model has high levels of stress at edge of top plate connecting to guide
pipe. Model 2 the top plate is much better relief of
stress concentration around de edge of top plate due
to the use of several brackets that help to better distribute the stresses along top plate. Simulation of all
other cases also shows a better distribution of stresses
in top plate of model 2.
CONCLUSIONS
From the analysis preformed it was verified and

checked high levels of stress between the top plate
and guide pipe.
Damage reports state that the major cracks begin in
welded connections of structural elements. In order to
improve welding quality future welding works should
be done in a dryer environment as possible.
The structural models examined indicate that several structural elements shall have their thickness
increased significantly.
With this study it was possible to improve the
original structural arrangement and test different
structural arrangements.
The study indicates that comparing model 1 and
model 2 in terms of flexibility (deformation) of the
mooring structure, model 2 it is more flexible.
From models tested it is concluded that model
2 is the better repair solution because the analysis
shows a better improvement of stress distribution from
original model to model 2. Also in terms of manufacturing structural elements in model 2 it is easier
to execute the repair because all structural elements
are straight instead model 1 that uses some curved
structural elements with high thickness.
Future work will be done to optimise the thickness
of some structural elements.
REFERENCES
ANSYS, Fundamental FEA Concepts and Applications.
Dakarnave Floating Dock damage reports, 20052009.
Dokk Mooring Arrangement DRW. N :1S-176896.
William D. Callister, Jr., Materials Science & Engineering
An Introduction Third Edition.
284

Numerical investigations to study the effect of weld parameters on the

temperature-time history in steel plates
B.Q. Chen, M. Adak & C. Guedes Soares
ABSTRACT: Numerical investigations are carried out to study the effect of weld parameters on the temperaturetime history in a butt-joint weld steel plate. A mathematical model of transient thermal process in welding is
established to simulate the transient thermal analysis with moving heat source model (Gaussian function) by using
finite element method. Results are compared to numerical and experimental results obtained from a previous
study. Parametric studies based on numerical results are carried out for different weld parameters including
welding speed, plate thickness, heat input, heat source type and finite element mesh.
1
1.1
INTRODUCTION
Introduction to welding
Welding is a complex industrial process that often

requires several trials before it can be done right. The
welding processes are carried out by skilled workers, but in the past few years automated machines
and robots are introduced in shipyards. To obtain the
expected productivity through mechanization, high
precision of parts to be assembled must be kept.
Therefore in the shipbuilding industry dimensional
predictability is important. In order to produce a highquality product, the accuracy control should be kept
through the whole assembly line. The concept of accuracy control should be incorporated in the structural
design, so that the designer can produce a better design
accounting for the geometric inaccuracy.
Generally, welding can be defined as any process in
which two or more pieces of metal are joined together
by the application of heat, pressure, or a combination
of both. Most of the processes may be grouped into two
main categories. pressure welding, in which the weld
is achieved by pressure; and heat welding, in which
the weld is achieved by heat. Heat welding is the most
common welding used today.
During the welding process, a liquid weld pool
is created through the interaction of an intense heat
source and the substrates being joined (see Figure 1).
Melting on the front side of the weld pool eliminates
the interface between the materials, while solidification on the back side of the weld pool fuses the
substrates together to create a solid joined part. Surrounding the fusion zone is a heat-affected zone, where
the substrate is heated to temperatures up to the melting point of the metal being joined. Solidification in
the fusion zone and solid-state phase transformations
in the heat-affected zone are responsible for dramatic
Figure 1. Illustration of a fusion weld.
changes in the microstructure and properties of the

welding joint.
1.2 Arc welding
Arc welding, which is heat-type welding, is one of
the most important manufacturing operations for the
joining of structural elements for a wide range of
applications, including guide way for ships, bridges,
trains, building structures, automobiles, and nuclear
reactors, to name a few. It requires a continuous supply
of either direct or alternating electric current, which
creates an electric arc to generate enough heat to melt
the metal and form a weld.
The arc welding process is a remarkably complex operation involving extremely high temperatures,
which produce severe distortions and high levels of
residual stresses. These extreme phenomena tend to
reduce the strength of a structure, which becomes
vulnerable to fracture, buckling, corrosion and other
type of failures. The most widely used arc welding
processes are shielded metal arc welding (SMAW),
285
gas tungsten arc welding (GTAW), gas metal arc

welding (GMAW), and submerged metal arc welding
(SMAW).
2 THEORY OF THE WELDING PROCESS
2.1
Heat transfer during welding process
Welding is a process with sharply local heating to

high temperature and rapid cooling afterwards, during which the temperature highly depends on the time
and location, while the mechanical properties of the
material varies with the temperature, and melting,
phase changing and latent heat exist. To sum up, the
weld induced temperature field analysis is a typical
nonlinear transient heat transfer issue.
During the welding process, the temperatures of different parts of the weldment vary immensely due to the
local heating. Consequently, heat transfer occurs both
inside the weldment and between the weldment and
surrounding.
Heat transfer mechanisms can be grouped into three
broad categories.
1) Conduction. Regions with greater molecular
kinetic energy will pass their thermal energy to
regions with less molecular energy through direct
molecular collisions, a process known as conduction. In metals, a significant portion of the
transported thermal energy is also carried by
conduction-band electrons.
2) Convection. When heat conducts into a static fluid
it leads to a local volumetric expansion. As a
result of gravity-induced pressure gradients, the
expanded fluid parcel becomes buoyant and displaces, thereby transporting heat by fluid motion
(i.e. convection) in addition to conduction. Such
heat-induced fluid motion in initially static fluids
is known as free convection.
3) Radiation. All materials radiate thermal energy in
amounts determined by their temperature, where
the energy is carried by photons of light in the
infrared and visible portions of the electromagnetic spectrum. When temperatures are uniform,
the radioactive flux between objects is in equilibrium and no net thermal energy is exchanged.
The balance is upset when temperatures are not
uniform, and thermal energy is transported from
surfaces of higher to surfaces of lower temperature.
Understanding of the theory of heat flow is essential in order to study the welding process analytically,
numerically or experimentally.
2.2
Development in prediction of welding
The problem of welding distortion during large steel

fabrications leads to dimensional inaccuracies and
misalignments of structural members, which can
result in corrective tasks or rework when tolerance limits are exceeded. This in turn, increases the cost of production and leads to delays. In fabrication and design
industries, for example, expenses for rework such as
straightening could cost millions of dollars. Therefore,

the problems of distortion and residual stresses are
always of great concern in welding industry.
In order to deal with this problem, it is necessary
to predict the amount of distortion resulting from the
welding operations. One way to predict the distortion
and shrinkage of steel welding is through numerical
analysis such as finite element analysis (FEA). Once
the techniques to predict the distortion and shrinkage are identified, then the problems can be controlled
accordingly.
Within the welding procedures, there are many factors such as welding process type, welding process
parameters, welding sequence, preheat patterns, level
of constraint and joint details that contribute to the
distortion of the welded structure. Knowing which
parameters have an effect on the quality of the weld
and which parameters give the most significant effect
on the weld quality are the main issues in welding
industry.
Welds are made at the junction of various pieces
that make up the weldment. The junctions of parts, or
joints, are defined as the location where two or more
members are to be joined. Parts being joined to produce the weldment may be in the form of rolled plate,
sheet, pipes, castings, forgings, or billets.
2.3 Mathematic model of heat source
The research activity in welding simulation started
decades ago. Rosenthal (1941) was among the first
researchers to develop an analytical solution of heat
flow during welding based on conduction heat transfer for predicting the shape of the weld pool for two
and three-dimensional welds. Using the Fourier partial differential equation (PDE) of heat conduction,
he introduced the moving coordinate system to develop
solutions for the point and line heat sources and applied
this successfully to address a wide range of welding
problems. His analytical solutions of the heat flow
made possible for the first time the analysis of the
process from a consideration of the welding parameters namely the current, voltage, welding speed, and
weld geometry.
His analytical solution (Rosenthal 1946) is based
on the concentration heat source, but it did not consider
the changes of material properties with temperature,
the phase change and latent heat. Hence, the result
deviation in the heat affect zone is relatively big.
Nevertheless, owing to its acceptable accuracy in the
low temperature zone and its simplicity, it has been
widely used in engineering applications.
Since the pioneering work of Rosenthal, Friedman
(1975) proposed to apply the Gaussian Distributed
heat source to approximately express the heat flux in
heating spot, as shown in Figure 2.
The heat flux q(r), with a distance r from the heat
source center, can be expressed as:
286
For a point within the first semi-ellipsoid located

in front of the welding arc, the heat flux equation is
described as:
where fr is the heat input proportion in the rear part.

For points (x, y, z) within the second semi-ellipsoid,
covering the rear section of the arc, the heat flux
equation is described as:
Figure 2. Gaussian distributed heat source.
In the past twenty years, considerable interest in the

thermal aspects of welding was expressed by many
researchers, such as Michaleris and DeBiccari (1997),
Wahab et al. (1998), and Gery et al. (2005).
Figure 3. Double-ellipsoidal distributed heat source.
where qmax is the maximum heat flux in the center of

the heat source; Q = IU is the heat flux of the arc;
and R is the radius of the heating spot.
For the normal welding process, such as GTAW and
SMAW, the Gaussian distributed heat source modal is
proved to provide precise enough results.
However, for the welding process in which the
momentum effect of arc is considerably large, the
disadvantage of the Gaussian distribution model in
result accuracy appears due to neglecting the effect
of the arc stiffness.
To solve this problem, Goldak et al. (1985) initially
proposed a semi-ellipsoidal heat source in which heat
flux is distributed in a Gaussian manner throughout
the heat sources volume.
Heat flux reaches the maximum value in the center of heat source, whose distribution is given by the
following equation.
where A, B, C are the heat flow distribution coefficient.

However, their experience with this heat source
showed that the predicted temperature gradients in
front of the arc were less steep than experimentally
observed ones and gradients behind the arc were
steeper than those measured. To overcome this, they
combined two semi-ellipsoids and proposed a new heat
source called double ellipsoidal heat source, as shown
in Figure 3.
Since two different semi-ellipsoids are combined
to give the new heat source, the heat flux within each
semi-ellipsoid are described by different equations.
3
3.1
FINITE ELEMENT ANALYSIS AND RESULTS

Considerations
Generally speaking, severely physical and chemical

reactions occur during the welding process, between
the base metal and the melted pool, which includes
the thermodynamics and heat transfer, the interaction
between the heat source and metal, solidification in the
welding line, the phase transformations in the welding
joint, and so on. To focus on the temperature field of
the welding structure, some factors that have a small
effect on it are supposed to deserve low weight in
consideration or sometimes even be neglected. Hence,
several simplifications have been considered during
the simulation.
First of all, the chemical reactions and agitation
and convection phenomena, and the phase change are
not taken into consideration.
Secondly, the heat transfer between the base metal
and the experiment table are neglected, assuming that
there is only convection heat transfer between the
edges of the work piece and the air, without radiation
heat transfer.
Thirdly, assume that the welding rod is of the same
material as the substrate.
Finally, assuming that the welding process is conducted with the constant heating speed V; the heat
source is complied with the Gaussian model.
3.2
Material properties
The material properties of metal, for instance, the specific heat, thermal conductivity, modulus of elasticity
287
Table 1.
Material Properties for Steel 25 (Biswas et al. 2001).
Temperature ( C)
Thermal conductivity
(W m1 C1 )
Specific heat capacity
(J kg1 C1 )
Poissons ratio ()
Thermal expansion coefficient
(106 C1 )
Modulus of elasticity (GPa)
Film coefficient (W m2 C1 )
100
51.9
450
0.28
10
200
1.0
51.1
499
0.31
11
200
6.5
300
450
46.1
566
631
0.33
12
0.34
13
200
7.5
and yield stress, vary with the change of temperature. The average values can be used in calculations
if the temperature does not vary too much. While
in the welding process, the temperature of weldment
varies shapely. In this case, to neglect the material
properties differences by temperature will result in big
deviations. Consequently, the temperature-dependent
material properties are supposed to be taken into
consideration in welding process simulation.
In this analysis, eight-node three-dimensional brick
thermal element, Solid 70, is used. The Solid 70
element has a three-dimensional thermal conduction
capability. It has eight nodes with a single degree of
freedom, temperature, at each node and is applicable to
a three-dimensional, steady-state or transient thermal
analysis. The element also can compensate for mass
transport heat flow from a constant velocity field. If
the model containing the conducting solid element is
also to be analyzed structurally, the element should be
replaced by an equivalent structural element (such as
Solid 45 or Solid 185).
The C-Mn steel whose temperature-dependent
material properties are listed in Table 1 was used.
3.3
41.1
150
7.3
550
37.5
706
0.36
14
110
7.2
600
35.6
773
720
30.6
1080
800
26.0
931
1450
29.5
438
0.37
14
0.37
14
0.42
14
0.47
15
88
7.2
20
7.1
20
7.1
2
7.0
Figure 4. Dimensions of work piece.
FE modeling
In this work, the butt welding process has been simulated in Ansys. The dimensions of the work piece
are 300*260*6 mm3 , as shown in Figure 4 in which the
arc center is moving along the x+ direction in line KI.
Furthermore, for the sake of saving computational
time and reducing computer configurations requirement, a half model has been taken in the simulation,
with symmetric boundary condition setting in the
symmetry plane, since the model is symmetric along
the X-Z plane.
Figure 5 displays the element mesh of the geometric
model to be solved. 2 mm element size is used in the
welding zone, while 8 mm element size is used in the
area far away from the heat affected zone (FAZ). Free
mesh is performed in the middle zone.
Meanwhile, a finer mesh was carried on, to study
the effect of the mesh style on the result. In this case,
four different element sizes, 1.25 mm, 2.5 mm, 5 mm,
and 10 mm are modeled. And all the elements are
hexahedrons (see Figure 6).
Figure 5. Element mesh.
Figure 6. Finer element mesh.
FE results demonstrates that the temperature distribution of the node located in the center of the welding
line are almost identical since the maximum deviation
in all time steps are only 0.21%, which means the relatively coarser model is satisfied to be utilized in the
simple butt welding process.
288
Figure 9. Temperature distribution in time = 10 s.
Figure 7. Comparison with Biswass result.
3.4 Verification
As presented by several researchers, finite element
method (FEM) has been successfully used to evaluate thermal response of the complex welding process.
In the present work, the finite element (FE) package
ANSYShas been used to simulate the arc welding
process.
Quite a few researchers have addressed this topic,
Tsai & Eagar (1983), Guedes Soares et al. (1998) and
Biswas et al. (2007), to name a few.
In Figure 7, a finite element simulation of a line
heating process was verified using the work of Biswas
et al. (2007). To do this, the same geometry and
material properties as Biswass model were using,
the heating process with total heat input 5350 Watts
and heating speed 6 mm/sec was simulated using
ANSYScodes.
From Figure 4, it shows that the temperature history
result of present work is in a good agreement with the
experimental result obtained by Biswas et al. (2007).
3.5
Results and discussion
Figure 8 to Figure 12 display the temperature field

distributions in the heating process, in the finer mesh
case. It is obvious that the temperatures increase
rapidly in the beginning, from room temperature to
over 350 C within one second. After that, the temperature field tends to be stable, which means the
temperature of a certain point is varied by time, but
the certain temperature is moving with the heat source,
meanwhile, the peak temperature is retaining of a value
of approximately 547 C.
During this period, the contour in the front of the

heat source is quite dense. On the contrary, in the rear
part, the temperature distribution gradient is comparatively small. Generally speaking, the shapes of the
contours are close to ellipsoids.
The cooling process in the finer mesh case is
described in the following four figures, from Figure
13 to Figure 15. After removing the heat source, the
peak temperatures drop to about 350 C in 50 seconds.
However, the changing rate is relatively lower comparing with the heating process. In the cooling stage, the
contours enlarge by time (see Figure 14). After 6000
seconds, the peak temperature becomes less than 70 C
(see Figure 15) and eventually the temperatures in all
nodes are tending to the room temperature.
289
Figure 18. Temperatures of nodes group 1.
Figure 19. Nodes Group 2 (parallel to Group 1).
Figure 16. Simulation results in different meshes.
Figure 17. Nodes in the welding line (Group 1).
Figure 16 demonstrates that the peak temperatures

of the node located in the center of the welding line,
in the coarser and finer mesh, are 1308 C and 1126 C
respectively.
To obtain more detailed information of the temperature field, several groups of nodes are selected.
Firstly 7 nodes in the welding line with uniform distance are chosen (see Figure 17), and the temperatures
of these 7 points are plotted in Figure 18, which indicates that, after reaching a stable state, the curve of the
temperature of each node is of the same tendency, and
for a certain node, the temperature shapely increases
when it is subject to heat input, and then fall down
with a relatively lower speed after achieving the peak
temperature. The first node and the last one are with
lower/higher temperature than the other nodes, which
is due to the less/more heat affected time.
Another 7 points are selected in Figure 19, whose
temperature distributions are similar to the previous
ones, expect the peak temperatures are much lower
since they are a little far from the heat source, and the
decrements of temperatures are slower (see Figure 20).
In the Z direction, 3 more nodes are taken as
Figure 21. Figure 22 shows their temperatures distributions. The temperature reduces with the increment
of distance from the heating surface but the effect is
not so significant.
In the Y direction, 7 nodes are marked in Figure 23,
with the distances away from the center line of 2, 4,
6, 8, 14 and 26 mm respectively. The more distances
away from the heating line, the lower temperatures
produced (see Figure 24). Figure 25 illustrates the
relationship between the peak temperatures and the
distances from the welding line.
290
Figure 23. Nodes Group 4 (in y direction).


Figure 21. Nodes Group 3 (in z direction).
Figure 25. Temperature-distance curve.

3.6
Parametric effect study
For the sake of finding out the effects of welding

process parameters (welding speed and heat input)
and geometric parameter (thickness of the metal plate),
series of calculations with different parameters have
been performed.
Figure 26 explains the temperature distributions of
the node located in the middle of the welding line,
in 5 cases of variant welding speeds, 4 mm/s, 6 mm/s,
8 mm/s, 10 mm/s and 12 mm/s. It can be concluded
that the lower the speed is, the higher temperature got
in the result.
The heat input in the previous calculation is 5350 W.
Keeping the fixed welding speed 6 mm/s, and changing the heat input to 3000 W, 6500 W and 7950 W,
then the corresponding temperature distributions are
plotted in Figure 27. There is no doubt that the higher
heat input results in higher temperature.
To study the effect of the plate thickness on the temperature field result, 5 cases are conducted in which
the welding speeds are fixed to 6 mm/s, the heat inputs
291
CONCLUSIONS
The model of transient thermal process in welding

which is a material nonlinear procedure is described
in the present work, to simulate the moving of the
heat source, and to predict the temperature field of
the entire weldment. The function of Gauss was chosen as heat source model, and was used as function
command to apply load of moving heat source.
The effects of welding speed, plate thickness and
heat input, heat source type and finite element mesh
on temperature field distribution were discussed. It
has been proved that all these three parameters play
an essential role in affecting the final temperature
distribution.
Figure 26. The effect of welding speed.
ACKNOWLEDGEMENTS
This work has been performed in a EU funded
project Network of Excellence in Marine Structures (MARSTRUCT) under contract TNE3-CT2003-506141.
REFERENCES
Figure 27. The effect of heat input.
Figure 28. The effect of plate thickness.
are set as 5350 W, but the plate thickness changes

as 4 mm, 6 mm, 8 mm, 10 mm and 12 mm. It can be
proved in Figure 28 that the thinner the plate, the higher
temperature is obtained.
ANSYS. 2007. Release 11.0 Documentation for ANSYS.

United States of America. ANSYS Inc.
Biswas, P., Mandal, N.R. & Sha, O.P. 2007.Three dimensional
finite element prediction of transient thermal history
and residual deformation due to line heating. Journal of
Engineering for the Maritime Environment, Part M 221:
1730.
Friedman, E. 1975. Thermomechanical analysis of the
welding process using the finite element method. Trans.
A.S.M.E. J. Pressure Vessel Tech 973.3: 206213.
Gery, D., Long, H. & Maropoulos, P. 2005. Effects of welding speed, energy input and heat source distribution on
temperature variations in butt joint welding. J. Materials
Processing Technology 1672-3.2-3: 393401.
Goldak, J.A., Chakravarti, A. & Bibby, M.J. 1985. A double
ellipsoid finite element model for welding and heat
sources. IIW Doc-212-603-85.
Guedes Soares, C., Gordo, J.M. & Teixeira, A.P. 1998. Elastoplastic behavior of plates subjected to heat loads. Journal
of Constructional Steel Research 452.2: 179198.
Michaleris, P. & DeBiccari, A. 1997. Prediction of welding
distortion. Welding Res. Suppl 76: 172181.
Rosenthal, D. 1941. Mathematical theory of heat distribution
during welding and cutting. Welding J 205.5: 220s234s.
Rosenthal, D. 1946. The theory of moving sources of heat
and its application to metal treatments. Trans. A.S.M.E:
849866.
Tsai, N.S. & Eagar, T.W. 1983. Temperature fields produced by traveling distributed heat sources. Welding J 62:
346s355s.
Wahab, M.A., Painter, M.J. & Davies, M.H. 1998. The prediction of the temperature distribution and weld pool
geometry in the gas metal arc welding process. J. Mater.
Process. Technol 77: 233239.
292

Mechanical properties assessment of specimens subjected to random

non-uniform general corrosion and tensile load
B.Q. Chen, Y. Garbatov & C. Guedes Soares
ABSTRACT: The objective of this work is to analyze the strength and deformability of steel specimens subjected to random non-uniform general corrosion and tensile load. The real shape of corrosion surface has
been generated by a probabilistic model of corrosion deterioration. Non-linear, large deformation and threedimensional finite element analyses have been employed to analyze the mechanical properties of deteriorated
specimens. The strength of corroded specimens is estimated as a function of the average corrosion depth and
the minimum cross sectional area. The deformability and energy absorption of the corroded plates are estimated
as a function of the surface roughness represented by the difference of averaged plate thickness and that at the
section of minimum cross sectional area.
1
INTRODUCTION
Steel structures such as ships, offshore platforms and

land-based structures operate in a very complex sea
water environment. The corrosion in steel structures is
often very severe, not only under sustained immersed
condition in ballast tanks, but also under general exposure to atmospheric conditions. To assess the structural
performance of steel structures such as aged ships,
it is important to analyze the residual strength and
energy absorption of corroded plates.
The corrosion attacking on metals could be divided
into general corrosion, galvanic cells, under-deposit
corrosion, CO2 corrosion, top-of-line corrosion, weld
attack, erosion corrosion, corrosion fatigue, pitting
corrosion, microbiological corrosion and stress corrosion cracking.Among these various types of corrosion,
general wastage that results in a generalized decrease
of plate thickness and pitting which consists of much
localized corrosion with very deep holes appearing in
the plate are two main corrosion mechanisms generally
presented in the steel plates. Most of the work referred
in this paper relate to the former, general corrosion.
Nowadays, numerical simulation is being used to
replacing time-consuming and expensive experimental work. A great number of researchers have studied
structural performance of steel structures subjected to
corrosions.
Paik et al. (2003) investigated the ultimate strength
characteristics of steel plate elements with pit corrosion wastage and under axial compressive load. While
the similar case but under in-plane shear loads had
been investigated as well.
Garbatov and Guedes Soares (2008) analyzed the
non-linear corrosion wastage of main deck plate of
tanker ships and bulk carriers based on the model
proposed by Guedes Soares and Garbatov (1999).

The corrosion wastage of deck plates was examined
with respect to how the corrosion wastage varies in
time as a result of generalized corrosion.
Nakai et al. (2004a, 2004b, 2006) analyzed circular
cone-shaped pits in the hold frames of bulk carriers,
and ellipsoidal-shaped pits in the bottom shell plates
of a tanker. They investigated actual pitting corrosion
observed on hold frames of bulk carriers in different
studies.
A total of 256 nonlinear FEA which was the full
combination of two cases of transverse pitting locations, five cases of plate thicknesses, four cases of
pitting breadths, four cases of pitting lengths and two
cases of pitting depths had been carried out by Ok
et al. (2007). They had proved that the length, breadth
and depth of pit corrosion have weakening effects on
the ultimate strength of the plates while plate slenderness has only marginal effect on strength reduction.
Empirical formulae to predict strength reduction due
to pitting corrosion had been derived by applying the
multi-variable regression method.
Jiang and Guedes Soares (2010) studied the influence of scattered pitted plates on the collapse strength
by using the mathematical model proposed by Daidola
et al. (1997), who developed a method to estimate the
residual thickness of pitted plates.
Silva et al. (2011) studied the effect of non-linear
randomly distributed non-uniform corrosion on the
ultimate strength of unstiffened rectangular plate subjected to axial compressive loading has been studied.
A series of 570 plate surface geometries are generated by Monte Carlo simulation for different degree
of corrosion, location and ages and nonlinear finite
element analyses were carried out, using a commercial finite element code and based on a regression
293
analysis, empirical formulae to predict strength reduction because of corrosion have been derived demonstrating a good accuracy.
The objective of the present work is to apply a
complete true stressstrain relationship to an elastoplastic material model of LS-DYNA (Livermore Software Technology, Livermore, CA, USA) to assess
the strength and deformability of steel specimen
subjected to random non-uniform general corrosion
and tensile load.
2
2.1
SIMULATION OF CORRODED SURFACES

Non-linear corrosion wastage model
Guedes Soares and Garbatov (1999) proposed a model

that describes the growth of corrosion wastage by a
non-linear function of time in three phases. In the first
phase, it is assumed that there is no corrosion because
a corrosion protection system is effective. Failure of
the protection system will occur at a random point of
time and the corrosion wastage will start a non-linear
process of growth with time, which levels off asymptotically at a long-term value of corrosion wastage.
Since then, several authors have proposed some variants of this model or have compared it with other
modifications.
The time dependent model of corrosion degradation
is illustrated in Figure 1, separated into four phases,
where in the first one there is no corrosion (O O).
The second phase corresponds to the initiation of failure of the corrosion protection system, which leads to
corrosion with the decrease of thickness of the plate
(OB, fast growing corrosion). The third phase BC corresponds slowly growing corrosion and the last one
(t > C) corresponds to a stop in the corrosion process
when the corrosion rate becomes zero.
The model is based on the solution of a differential
equation of the corrosion wastage:
Figure 1. Thickness of corrosion wastage as a function of

time.
where d is the long-term corrosion wastage, d(t) is

the corrosion wastage at time t, c is the time without
corrosion which corresponds to the start of failure of
the corrosion protection coating (when there is one),
and t is the transition time duration, which may be
calculated as d /tg, where is the angle defined by
OA and OB.
2.2
Non-uniform general corroded surface model
Based on 1168 measurements of deck plates from

ballast tanks with original nominal thicknesses varying from 13.5 to 35 mm on ships with lengths
between perpendiculars in the range of 163.5 to 401 m,
Garbatov et al. (2007) determined the parameters of
the mean value of the corrosion depth as a function
of time under the assumption that it is approximated by the exponential function given in Eqn. (1),
where the long-term corrosion wastage is equal to
1.85 mm, the time without corrosion is equal to 10.54
years and the transition period is equal to 11.14 years.
Another important statistical descriptor of the data
set was the standard deviation for each yearly subset
of data. The standard deviation as a function of time
was fitted to a logarithmic function:
where the coefficients a1 and b1 are defined as 0.384

and 0.710 respectively, for the deck plates of ballast
tanks.
Several distributions were evaluated by Garbatov
et al. (2007) and it was concluded that corrosion
wastage depth is the best fit by the Log-normal distribution, which was as well adopted in the present
work. Based on their work, a reformative non-uniform
general corroded surface model is proposed here.
Figure 2c displays the non-uniform general corroded
surfaces models of 8 mm thickness tensile specimens
with three different levels of corrosion defined by
controlling the minimum plate thickness.
The corroded plate surface is modelled as random
plate thickness that results in the random vertical position of the coordinates of corroded surface for equally
spaced reference points positioned along the x and
y direction of the plate. These reference points are
defined in a Monte Carlo simulation as being the nodes
of the finite element mesh on the plate.
The plate thickness, Zijcorroded , at any reference
point with coordinates x, y for the corroded plate
surface, is defined by the random thickness of the
intact plate surface, Zijintact affected by the random
vertical reduction resulting from the corrosion depth,
corrosion depth
Zij
as defined by Silva et al. (2011):
where Z are the matrixes of the corroded and intact

surface and corrosion depth.
This convention is used to derive the formulation
that describes the vertical position of the surface of the
294
Figure 3. Geometry of tensile specimen.

Figure 2a. Corroded surfaces, Case 1.
Figure 2b. Corroded surfaces, Case 2.

Figure 4a. True stress true strain relation (Ahmmad and
Sumi, 2010), full range curve.
Figure 2c. Corroded surfaces, Case 3.
non-linear corroded plate in the Monte Carlo simulation resulting in randomly distributed plate thicknesses
for randomly defined reference nodes at a specific year
based on Eqn. (3) applying the corrosion degradation
levels as defined by Eqn. (1) and (2).
The vertical random coordinates (corrosion depth)
of the corroded and intact plate surfaces and corrosion depths are modelled by a log-normal distribution.
The intact plate surface coordinates and corresponding
corrosion depths are considered as not correlated.
The average thicknesses of plates subjected to these
levels of corrosion titled as case 1, case 2 and case 3,
are 7.380 mm, 6.554 mm and 5.852 mm, respectively.
GEOMETRIC AND MATERIAL PROPERTIES

OF STEEL SPECIMEN
3.1 Tensile specimen

A typical geometry of the specimen of tensile test,
which has enlarged ends or shoulders for gripping,
is shown in Figure 3. The cross-sectional area of the
gage section that is the important part of the specimen is reduced relative to that of the remainder of the
specimen so that deformation and failure will be localized in this region. The gauge length is the region over
Figure 4b. True stress true strain relation (Ahmmad and

Sumi, 2010), detailed curve of 0 0.1 true strain.
which measurements are made and is centered within

the reduced section.
Ahmmad and Sumi (2010) estimated the strength
and deformability of steel plates with various pit sizes,
degrees of pitting intensity, and general corrosion both
experimentally and numerically. The vision-sensor
technology was used in their study to obtain the relationship between the true stress and the true strain
beyond the onset of localized necking. The true stress
strain relationship, including the material response in
both pre- and post-plastic localization phases, is necessary as input for numerical analyses (see Figure 4).
The mechanical properties of the material used in their
work are listed in Table 1.
3.2 Stress-strain relationship
In the conventional engineering tension test, the engineering measures of stress and strain, denoted in this
module as e and e respectively, are determined from
295
Table 1.
Material properties.
Setting
Value
Unit
Material
Yield strength
Mass density
Youngs modulus
Tensile strength
SM490A
365.0
7853.6
206.9
509.4
MPa
kg/m3
GPa
MPa
Figure 5. Finite element model.
the measured the load and deflection using the original

specimen cross-sectional area A0 and length L0 as:
where is the controlled displacement applied in one

end of the specimen in a tensile test, while P is the load
corresponding to the displacement. When the stress
(e ) is plotted against the strain (e ), an engineering
stress-strain curve is obtained.
However, the engineering stress-strain curve must
be interpreted with caution beyond the elastic limit,
since the specimen dimensions experience substantial
change from their original values during the test. If
the true stress, based on the actual cross-sectional area
of the specimen (A), is used, the stress-strain curve
increases continuously to fracture. If the strain measurement is also based on instantaneous measurement,
the curve that is obtained such as that shown in Figure 4
is known as true stress-strain curve.
Any point on the true stress-strain curve can be considered the yield stress for a metal strained in tension
by the amount shown on the curve. Thus, if the load is
removed at this point and then reapplied, the material
will behave elastically throughout the entire range of
reloading. The true stress t is expressed in terms of
engineering stress e and engineering strain e by:
While the true strain t , is determined from the

engineering strain e by:
4
4.1
STRESS-STARAIN ANALYSIS
Finite element model
Numerical analyses were carried out by using a nonlinear implicit finite element code, LS-DYNA, applying
a quasi-static type of load. The constitutive material
model is an elastoplastic material where an arbitrary
stress versus strain curve can be defined. This material model is based on the J2 flow theory with isotropic
hardening.
The geometric model of the tensile specimen has

been shown in Figure 3, in which the total and gauge
length are 308 and 140 mm respectively, the width is
50 mm and the reduced width in the gauge section is
40 mm. The thickness of the intact specimen is 8 mm,
while three different levels of corrosion are used using
the reformative non-uniform general corroded surface
model.
Figure 5 demonstrates the finite element model
of the tensile specimen. Four different element sizes,
10 mm, 5 mm, 2.5 mm and 1.25 mm were used in
different regions.
In the left side of specimen, the degrees of freedom
of all nodes are constrained in all directions, while
a constant displacement of 3 mm/min was applied in
all nodes of the right side, in the loading direction.
Fracture is introduced by allowing elimination of
elements when strain to failure is achieved.
The material selected from the library of LS-DYNA
is Mat.024 Piecewise linear plasticity, which allows
the definition of a true stressstrain curve as an offset
table. The true stress-strain curve used in the study of
Ahmmad and Sumi (2010) (see Figure) are used as an
input for the calculations.
4.2 Effect of failure criterion
The failure criterion used for the numerical simulation of the tensile test is an important issue. In the
present analyses, the failure criterion is defined as the
plastic strain tends to fail. The element where necking occurs is deleted from the model when the plastic
strain reaches the setting value predefined as a failure
criterion.
Normally, the failure criterion in terms of the plastic
strain behavior in an element can be determined in such
a way that the total elongation in FE analysis reaches
the experimental elongation. For the sake of studying
the effect of the strain to fail on the tensile properties,
three different failure criteria for strains, 0.25, 0.55
and 0.85, are applied for the Case 1 in the numerical
simulations. Three resultant curves shown in Figure 6
indicate that the initial linear portions of the curves are
exactly identical, which means that the strain to fail
has no effect on the modulus of elasticity which can
be calculated as the slope of the initial linear portion of
the curve. However, it is found from this figure that the
failure strain has a significant effect on the elongation.
With the increase of strain to failure from 0.25 to 0.55,
the elongation increases up to twice, and it increases
to more than 15% when the strain to fail reaches the
value 0.85. There is a small increase in elongation from
296
Table 2. Tensile strength as a function of corrosion level.

Case
Intact
Corrosion rate (%)

Tensile strength (MPa)
Strength loss (%)
0.0
548.3
0.0
7.8
504.1
8.1
18.1
444.3
19.0
26.9
438.6
20.0
Figure 6. Stress elongation curve for different failure

criteria.
Figure 8. Liner part of the nominal stress nominal strain

curve as a function corrosion levels.
Figure 7. Nominal stress versus percent elongations curve.
strain to fail of 0.85 to that of 0.932. However, for the

present study, the strain failure criterion assumed is
0.932.
4.3
Effect of corrosion level
Figure 7 shows the nominal stress versus percent

elongations curves, of the non-corroded specimen
and three different cases of different corrosion levels defined in Section 2.2, in which the strain to
fail value is assumed as 0.932. The Case 1, Case 2
and Case 3 refer to the specimens subjected to three
different levels of general non-uniformly distributed
corrosion with the average plate thickness of 7.380 mm
(7.8%), 6.554 (18.1%) mm and 5.852 (26.9%) mm,
respectively.
Tensile strength is one of the most important
parameters that describes the mechanical properties
of a steel plate under tensile load. As can be seen from
Figure 7, the intact specimen has the highest tensile
strength, while the strength decreases with the increase
of the corrosion rate. Table 2 lists the tensile strength
of each case, in which 8.1% of the strength is lost
when the intact specimen suffers to 7.8% general corrosion, after that the strength loss still increases to 19%
as the corrosion rate reaches 18.1%. From Case 2 to
Case 3, the changing rate of the strength loss becomes
lower, from 19.0% in the case of the 18.1% general
corrosion to 20.0% in that of 26.9% general corrosion. And it implies that the severer corrosion results
in less elongation.
Figure 9. The average thicknesses of all cross sections

along specimen length.
Another important mechanical property of a steel

plate is the modulus of elasticity. Figure 8 shows the
initial liner portion of the nominal stress nominal
strain curve with various corrosion levels. According to Hookes Law, the slope of the stress strain
curve stands for the modulus of elasticity which is
also known as the Youngs modulus. It is evident that
the plate modulus of elasticity is decreased as the
corrosion deterioration increases.
4.4 Effect of minimum cross section
Since the corroded surfaces are defined from random
variants, it is also interesting to pay special attention
on the cross section, where the minimum thickness is
located. Figure 9 shows the average thickness of each
cross section along the specimen length in all the three
cases. The locations, where the failures occurred are
indicated.
The marked failures occurring locations show a
small correlation with the locations of minimum
thicknesses of the cross sections, given that they are all
297
identical to one of the minimum section average thickness, but not the least value. It may be concluded that
the determination of the failures occurring locations is
complicated and depends on various factors including
the minimum section average thickness. From the FE
analysis it was observed that the failure criteria affect
the locations where the failures occur as well.
4.5
Stress-strain analysis
The true stress-strain curve of many metals in

the region of uniform plastic deformation can be
expressed by the simple power law (Ludwik, 1909):
Figure 10. Loglog plot of the stressstrain curve.
where n is the strain hardening exponent, K is the

strength coefficient, and p is the plastic strain.
A log-log plot of the true stress-strain curve from
yield point up to the maximum load will result in
a straight line if Eqn. (7) is fitted to the observed/
experimental data.
Another common variation on the simple power
law is the Ramberg-Osgood equation (Ramberg and
Osgood, 1943):
Table 3.
Case
n value
K value
Intact
Case 1
Case 2
Case 3
0.251
0.249
0.250
0.251
1054 MPa
971 MPa
859 MPa
843 MPa
Table 4.
levels.
where and m are dimensionless constants, E is the

Youngs modulus and R is a reference stress. If m is
very large, then p remains small until approaches
R , and increases rapidly when exceeds R , so that R
may be regarded as an approximate yield stress. In the
limit as m becomes infinite, the plastic strain is zero
when < R , and is indeterminate when = R , while
> R would produce an infinite plastic strain and is
therefore impossible. This limiting case accordingly
describes a perfectly plastic solid with yield stress R .
If the deformation is sufficiently large for the
elastic strain to be neglected, then Ramberg-Osgood
equation can be solved for in terms of :
which is equivalent to the power law, Eqn. (8) if

m1
= K.
consider m1 = n and R ER
Figure 10 shows the log-log plot of the stress-strain
relationships in the present study. All curves of the
four cases with different corrosion deteriorations are
expressed by a linear regression y = ax + b in which
a is equal to the strain hardening exponent (n), and b
stands for the logarithmic form of the strength coefficient K. All the values of n and K are listed in Table 3.
It can be observed that the strains hardening exponent are around 0.250 in all corrosion rates. While the
K value decreases with the increase of the corrosion
deterioration.
n and K values, with various corrosion levels.
Resilience and toughness, with various corrosion
Case
Unit
Ur
Ut
Intact
Case 1
Case 2
Case 3
MPa
MPa
MPa
MPa
0.459
0.380
0.344
0.367
95.9
74.8
62.3
56.5
4.6 Energy absorption analysis

Moreover, two parameters with respect to the energy
absorption are also analyzed here. One is the modulus of resilience (Ur ) which means the capacity to
absorb energy when deformed elastically and recover
all energy when unloaded. It can be calculated as the
area under the elastic portion of the stress-strain curve.
The other is modulus of toughness Ut that stands for
the energy to break a unit volume of material, or absorb
energy to fracture. It can be calculated as the area
under the entire stress-strain curve.
From Figure 7 and Figure 8, the resilience and
toughness can be easily calculated by integration. The
results are shown in Table 4, where one can observe
that the modulus of resilience is decreasing with the
increase of the corrosion deterioration, except for
Case 3. In Case 3, the resilience is even bigger than
the one of Case 2, though the Youngs modulus that is
the slope of the initial linear portion of the curve in
Case 3 is less than that of Case 2. It is because in Case
2 the elongation of the linear portion of the stress-strain
curve is bigger.
Figure 11 shows the capacity of energy absorption
with respect to the general non-uniformly distributed
298
In different cases with various level of corrosion

deterioration, the strains hardening exponent n are
almost the same. While the strength coefficient K
decreases with the increase of the corrosion rate.
The capacity of energy absorption decreases from
95.9 MPa to 74.8 MPa as the intact specimen suffers
7.8% general corrosion. After that the energy absorption still decreases with the increase of the corrosion
rate, but with a lower rate.
REFERENCES
Figure 11. Energy absorption with respect to corrosion
deterioration.
corrosion deterioration. It is concluded that the capacity of the energy absorption decreases from 95.9 MPa
to 74.8 MPa as the intact specimen suffers to 7.8%
general corrosion. After that the energy absorption
still decreases with the increase of the corrosion rate,
but with a lower rate.
5
CONCLUSIONS
This work presented an analysis of the strength and

deformability of steel specimen subjected to random
non-uniform general corrosion and tensile load. Based
on the results obtained in the present work, several
conclusions can be drawn.
The strain to failure has no effect on the modulus of
elasticity which can be calculated as the slope of the
initial linear portion of the stress-strain curve. However, it has a significant effect on the elongation. With
the increase of strain to failure from 0.25 to 0.55, the
elongation increases to twice as the previous value.
And it increases to more than 15% when the strain to
fail reaches the value 0.85. There is a small increase in
elongation from strain to fail of 0.85 to that of 0.932.
The tensile strength decreases with the increase of
the corrosion deterioration. 8.1% of the strength is
lost when the specimen is subjected to 7.8% general
non-uniform corrosion, after that the strength loss still
decreases with the increase of the corrosion deterioration, but with a lower changing rate, that it varies
from 19.0% to 20.0% as the corrosion changes from
18.1% to 26.9%. The severer corrosion results in less
elongation.
The modulus of elasticity and the energy absorption capacity of a steel plate is decreased as corrosion
deterioration increases.
The failures occurring locations show little correlation with the locations of minimum thicknesses of
the cross sections. It may be concluded that the failure occurring locations is complicated and depends on
various factors including the minimum section average
thickness as well as the failure criteria.
Ahmmad, M. M. and Sumi, Y., 2010, Strength and deformability of corroded steel plates under quasi-static tensile
load, J Mar Sci Technol, 15, pp. 115.
Daidola, J., Parente, J. and Orisamolu, I., 1997, Residual
strength assessment of pitted plate panels, Ship Structure
Committee.
Garbatov, Y. and Guedes Soares, C., 2008, Corrosion
Wastage Modeling of Deteriorated Ship Structures,
International Shipbuilding Progress, 55, pp. 109125.
Garbatov, Y., Guedes Soares, C. and Wang, G., 2007,
Non-linear Time Dependent Corrosion Wastage of Deck
Plates of Ballast and Cargo Tanks of Tankers, Journal
of Offshore Mechanics and Arctic Engineering, 129,
pp. 4855.
Guedes Soares, C. and Garbatov, Y., 1999, Reliability of
Maintained, Corrosion Protected Plates Subjected to
Non-Linear Corrosion and Compressive Loads, Marine
Structures, pp. 425445.
Jiang, X. and Guedes Soares, C., 2010, Ultimate Compressive
Capacity of Rectangular Plates with Partial Depth Pits,
Proceedings of the OMAE, Paper n. OMAE201021050.
Ludwik, P, 1909, Elemente der Technologischen Mechanik,
Applied Mechanics (Springer Verlag, Berlin).
Nakai, T., Matsushita, H. and Yamamoto, N., 2004a, Effect of
pitting corrosion on local strength of hold frames of bulk
carriers (2nd report) Lateral-distortional buckling and
local face buckling, Marine Structures, 17, pp. 612641.
Nakai, T., Matsushita, H. and Yamamoto, N., 2006, Effect of
pitting corrosion on the ultimate strength of steel plates
subjected to in-plane compression and bending, J Mar Sci
Technol, 11 (1), pp. 5264.
Nakai, T., Matsushita, H., Yamamoto, N. and Arai, H., 2004b,
Effect of pitting corrosion on local strength of hold
frames of bulk carriers (1st report), Marine Structures, 17,
pp. 403432.
Ok, D., Pu, Y. and Incecik, A., 2007, Computation of
ultimate strength of locally corroded unstiffened plates
under uniaxial compression, Marine Structures, 20 (12),
pp. 100114.
Paik, J., Lee , J., Hwang, J. and Park, Y., 2003, A TimeDependent Corrosion Wastage Model for the Structures
of Single and Double Hull Tankers and FSOs, Marine
Technology, 40, pp. 201217.
Ramberg, W and Osgood, W. R, 1943, Description of stressstrain curves by three parameters. Tec. Note 902, National
Advisory Comm. Aeronaut., Wangshington, D.C.
Silva, J. E., Garbatov, Y. and Guedes Soares, C., 2011,
Ultimate strength assessment of ageing steel plates
subjected to random non-uniform corrosion wastage.
In: Advances in Marine Structures. C. Guedes Soares and
W. Fricke, (Eds.), Taylor & Francis, pp. 213220.
299

Fatigue damage assessment of stiffener-frame structure

Wei Huang, Y. Garbatov & C. Guedes Soares
ABSTRACT: This paper presents a fatigue damage analysis of very complex stiffener frame ship structures.
A three-dimensional finite element model is generated and linear elastic finite element analysis is performed.
Hot-spot stresses of a stiffener-frame structure are analyzed applying 20-node solid elements using a commercial
finite element code. Based on a FEM analysis, hot-spot stresses are determined and fatigue damage is quantified
employing the S-N approach. Comparative analysis of fatigue damage for all hot spots is presented and the
difference between numerical predictions is discussed.
INTRODUCTION
Fatigue assessment has been gaining much attention

during the last decades and much effort has been
spent to develop methodologies to predict the fatigue
strength or life of welded joint structures. The fatigue
life of a welded detail may be estimated based on the
S-N approach and in the presence of a crack by fracture
mechanics methods.
The main steps in fatigue analysis based on direct
calculations involve the description of the wave
induced loading, the stress distribution in the structure,
the model of fatigue damage and the probabilistic evaluation of the different steps to arrive at a safety index
or time dependent reliability. The analysis of stresses is
a difficult task due to the complexity of a typical ship
structure. Fatigue analysis has to be based on combined load response that includes local stress response
and global load response. The global load considers a
reliable description of the overall stiffness and global
stress distribution in the hull. It takes into account vertical hull girder bending including shear lag effects,
vertical shear distribution between ship side and bulkheads, horizontal hull girder bending including shear
lag effects, torsion of the hull girder and transverse
bending and shear. The global analysis may be carried
out with a relatively coarse mesh.
Simple stress calculation methods cannot accurately predict local stress response. Sophisticated
finite element models can describe local stress
response but yield typicality to hierarchical multilevel
modeling where a lower level global model is used
for calculating load for the higher-level local model
(Guedes Soares et al., 2003, Garbatov et al., 2010). Not
only mesh density of the local finite element model
but also the accuracy of the load applied to the local
model governs the accuracy of calculated local stress.
The aim of the local finite element analysis is normally not to calculate directly the stress at a detail,
but to calculate the stress distribution in the region of

the hot spot and to use these stresses as a basis for
derivation of the geometric stress concentration factors, which can even be considered as time dependent
(Garbatov et al., 2002a). In this aspect in the open literature has a large spectrum of applied finite element
modeling techniques (Saad-Eldeen et al., 2010, Gaspar
et al., 2011).
The analysis of the local effects at critical
points such as hot spot stresses becomes a routine
practice nowadays (Almar-Naess, 1985, Fricke and
Petershagen, 1992, Matoba et al., 1994). Different
detailed stress analyses have been performed on largescale specimen (Yuasa and Watanabe, 1999, Rudan
et al., 2000, Garbatov et al., 2002b, 2004).
The structural hot spot stress is calculated based on
an idealized, aligned weld joint and it is related to the
plate in front of the weld toe. The IIW (Niemi et al.,
2004) recommendations include two fillet weld S-N
curves, FAT 90 and FAT 100, depending on whether
the weld is load carrying or non-load-carrying. The hot
spot stress method is not concerned with the size effect
of the base plate, nor the size of the throat thickness
of the welded joint. Xiao and Yamada (2004) proposed
a method for evaluating the structural stress approach
based on the stress at a location 1 [mm] below the
weld toe surface in the direction corresponding to the
expected crack path. Finite element analysis is necessary with a mesh having element size of 1 [mm] or
less.
Dong (2001) proposed a method that has been
claimed to be mesh-size insensitive and can be used in
combination with both solid and shell element models.
Normally both fatigue and corrosion will be present
and their combined effect needs to be considered in
that the decreased net section due to corrosion will
increase the stress levels, which in turn increase the
rate of crack growth. This effect has been considered
by Garbatov et al. (2002a), which adopted linear and
301
Figure 1. Structural hot-spot stress.

Figure 2. Frame structure.
nonlinear models of the effects of corrosion wastage

on the plate thickness reduction and stress concentration factors and fatigue damage are calculated as a
function of time. It is concluded that the stress concentration factors have a nonlinear dependency with the
time and this leads to a significant difference of the
fatigue damage of structural components subjected to
corrosion as compared with the traditional predictions.
A formulation for the assessment of a local ship hull
structure regarding to fatigue failure and reliability
has been presented by Garbatov and Guedes Soares
(1998, 2002). The potential cracks are considered to
occur in the side shell, in the connections between
longitudinal stiffeners and transverse web frame. The
model accounts for the crack growth process applying
linear elasto fracture mechanic approach.
The objective of this work is to analyze fatigue
damage of a local ship hull structure of a very complex stiffener-frame in the side shell applying the S-N
approach. The work covers several steps of calculations including generation of a three-dimensional
finite element model of stiffener-frame structure with
welded details by the use of ANSYS commercial finite
element code, applying 20-node solid elements. A
linear elastic finite element analysis is performed to
determine the stress distribution around the welded
joint details studied here and to define structural
stresses of all hot-spots by the means of the hot-spot
stress assessment procedure. Finally, this study quantifies and compares the fatigue damage for all hot spots
employing an S-N approach based on hot-spot stress.
HOT-SPOT STRESS
Hot spot is defined as a weld toe position in a structure,

where fatigue crack may initiate due to the combined
effect of structural stress fluctuation and the weld
geometry. The structural stress at a hot spot is defined
as hot-spot stress, HS . The structural stress includes
all stress raising effects excluding the stress due to the
local weld profile itself. The stresses are determined
on the structural surface at the point of the vicinity of
the welded joints as may be seen from Figure 1.
For fatigue assessment, the hot-spot stress HS is

determined at hot spot of a welded joint, where fatigue
crack initiation is expected. In general, the maximum
principle stresses are considered for the hot spot calculation. The hot-spot stress HS is the maximum
principle stress at the weld toe calculated using reference points and an extrapolation to the weld toe at
the hot-spot in consideration (Petershagen et al., 1991,
Fricke et al., 2007).
The hot-spot stress extrapolation procedures may
be based on a surface stress extrapolation, with the
assumption that the hot-spot stress varies linearly
through the plate thickness and the effect of the weld
notch is localized within a distance close to the weld,
where the locations of the reference points are nor
expressed as a function of plate thickness.
First, the stresses at the reference points are determined and secondly the hot-spot stresses HS are
calculated using an extrapolation equation (Niemi
et al., 2004):
where t is the plate thickness, HS is the structural

hot spot stress, 0.5t is the principal stress of reference
point which is 0.5t away from the weld toe (hot-spot)
and 1.5t is the principal stress at 1.5t.
STRUCTURE DESCRIPTION
The stiffener-frame structure analyzed here is composed of a ship side shell plate, a longitudinal stiffener,
three welded frames with stiffener and a lug as may be
seen in Figure 2. The frame structure is made of normal
shipbuilding steel, 235 MPa of yield stress.
The structural geometry is given in Figure 3,
where ship side shell plate is defined as 6400
800 16 mm; frames are of 1300 800 12 mm; the
lugs are of 180 149 12 mm. The longitudinal stiffener is made of HP 320 14, the frames are made of
HP 260 12. The material properties used for the FEA
are Young modulus of E = 206000 MPa and Poisson
302
Figure 5. Finite element model of stiffener-frame structure.

Figure 3. Stiffener-frame structure section.
Table 1. Welded toe description.
No
1
2
3
4
5
Welded
Joint
Position of
weld toe
Size of
weld toe
mm
Longitudinal Stiffener
Center Frame and side
shell
Longitudinal stiffener and
center frame
Longitudinal Stiffener and
lug
Center frame and lug
Figure 4a
Figure 4a
16
16
Figure 4a
14
Figure 4a
14
Figure 4b
12
Figure 6a. Welded toe between longitudinal stiffener and

side shell.
Figure 4. Types of welded joint.
ratio is 0.3. The load subjected to structure is defined

by two concentrated forces of 100 kN.
The sizes of weld toes in the stiffener-frame structure are determined based on (Niemi et al., 2004) and
are listed in Table 1 and Figure 4.
The complexity of the frame structural geometry and

the existence of many hot spots, defined by weld toes,
require a three dimensional finite element model and
stress analysis. The finite elements used in the model
are 20-node quadratic solid elements with mid-side

nodes at the edges. The 20-node elements tolerate
irregular shapes without much loss of accuracy. The
displacement shapes of elements are compatible to
model curved boundaries. The 20-node solid element
has three degrees of freedom per node: translations in
the nodal x, y, and z directions.
The finite element model of the stiffener-frame
structure is shown in Figure 5. The boundary conditions are defined in the way that the displacement and
rotation of nodes of bottom areas of frames and stiffeners are zero. The elements size, in the vicinity of
welded joint studied here, is defined as thickness size.
The entire stiffener-frame structure contains 24051
elements and 143206 nodes. The weld toe configuration of stiffener-frame structure may be seen in
Figure 6.
The finite element analyses performed here is linear and elastic one. The objective of the FE analysis
is to define the principal stresses in the vicinity of
the root and toe of the weld joints. The weld toe here
303
Figure 6b. Welded toe between center frame and side shell.
Figure 6e. Welded toe between lug and center frame.
Figure 6c. Welded toe between middle frame and longitudinal stiffener.
Figure 7a. Global stress distribution.
Figure 6d. Welded toe between lug and longitudinal

stiffener.
Figure 7b. Local stress distribution between lug and longitudinal stiffener.
304
Figure 7c. Local stress distribution between lug and center

frame.
Figure 8. Hot-spots of stiffener-frame structures.
The long-term distribution of the stress range of

welded details is assumed here to be a two-parameter
Weibull distribution:
Figure 7d. Local stress distribution between center frame

and shipside.
is assumed to be polished and fully penetrated butt

weld. No account for stress concentrations associated
with weld defects is taken. Figure 7 shows the stress
distribution of the entire stiffener-frame structures.
Figure 7(b)7(d) show the hot spots to be analyzed.
where: Q is probability of exceedance of the stress

range , is the shape parameter of the Weibull distribution and may be established from long term wave
load analysis and q is the scale parameter, defined as:
where n0 is total number of cycles associated with the

stress range level 0 .
A one-slope S-N curve is used for the fatigue
damage assessment (Nolte and Hansford, 1976):
FATIGUE DAMAGE ASSESSMENT
Fatigue damage assessment is carried out based on

the Palmgren-Miner (Palmgren, 1924) rule, which
accumulate individual fatigue damage ratios caused
by all stress range occurrences during the anticipated
service life of the structure. The basic assumption of
the S-N approach is that the structural damage per load
cycle is constant at a given stress range. The stress
range is defined as a difference between maximum
and minimum stress amplitude as:
where: 0 is long-term average response zero-crossing

frequency, assumed as 0.1 Hz; Td is design life
assumed in the fatigue assessment of ship, and is taken
as 25 years, which can be converted into 7.884E+8
seconds; K is a parameter of S-N fatigue curve, 1012.38 ,
m is a parameter of S-N fatigue curve, 3.0 (DnV,
2010). is considered as 1.0 and D represents the
accumulated fatigue damage.
The positions of all hot-spots estimated in stiffenerframe structures are marked as black dots in Figure 8,
where there are 16 hot-spots altogether.
305
Table 2.
each hot-spot of welded joint detail shows that the

accumulated fatigue damage of the hot spot No. 4
has the maximum and exceeds 1.0 during the anticipated service life of the structure and the fatigue
life is less than the design life. This reveals that the
welded joint between longitudinal stiffener and lug
(hot spot No. 4, see Figure 6(d)) cannot provide adequate fatigue strength and is the weakest hot spot of
the structure and a fatigue crack initiation is most
probable expected. The welded joint between the longitudinal stiffener and the lug control the fatigue life
of entire frame structure. Further structural redesign
may be needed. The fatigue strength of welded joint
between the longitudinal stiffener and the lug in the
stiffener frame structure should be reinforced in order
to strengthen the structure to ensure safety during the
design life.
The work uses 3-D finite element analysis with
20 nodes solid elements, which proved to be effective for structural fatigue damage assessment. The
results achieved here may be used for fatigue reliability
assessment.
Fatigue damage.
No
HS
MPa
Fatigue
damage
Fatigue life
Year
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
12.704
0.280
130.578
318.978
185.189
76.990
1.650
16.442
158.619
14.705
38.146
140.586
81.584
60.062
79.619
99.082
0.690
0.015
7.089
17.316
10.053
4.180
0.090
0.893
8.611
0.798
2.071
7.632
4.429
3.261
4.322
5.379
6.468E05
6.955E10
7.024E02
1.024
0.200
1.440E02
1.417E07
1.402E04
0.1259
1.003E04
1.751E03
8.766E02
1.713E02
6.836E03
1.592E02
3.069E02
3.8652E+05
3.5944E+10
355.9
24.4
124.7
1736.4
1.7647E+08
1.7826E+05
198.5
2.4921E+05
14276.1
285.1
1459.3
3657.2
1570.0
814.6
ACKNOWLEDGEMENTS
Risk-Based Expert System for Through-Life Ship
Structural Inspection and Maintenance and New-Build
Ship Structural Design (RISPECT) and has been partially funded by the European Union through the
REFERENCES
Figure 9. Fatigue damage of each hot-spot.
The cycling load acting on frame structure varies

from 100 KN to 100 KN, resulting in the hot-spot
stress range HS for each welded joint in the stiffenerframe structures as may be seen in Table 2.
Table 2 summaries the fatigue damage calculation
of hot spots of the stiffener-frame structures. The
fatigue damage of each hot-spot is performed using
Eqn (5) and the service life of each individual hot spot
is obtained as a ratio between design life and fatigue
damage index. The comparison of fatigue damage
between different hot spots is shown in Figure 9.
CONCLUSIONS
Fatigue damage assessment of stiffener-frame structure subjected to cyclic loading has been performed
based on the 3-D linear FE analysis and SN approach.
The comparison of fatigue damage assessment of
Almar-Naess, A., 1985. Fatigue Handbook, Offshore Steel

Structures. Tapir.
DnV, 2010. Fatigue Assessment of Ship Structures, Classification Notes No 30.7.
Dong, P., 2001. A Practical Stress Definition and Numerical Implementation for Fatigue Analyses. International
Journal of Fatigue 23, pp. 865876.
Fricke, W., Bollero, A., Chirica, I., Garbatov, Y., Jancart, F.,
Kahl,A., Remes, H., Rizzo, C.M., von Selle, H., Urban,A.,
Wei, L., 2007. Round Robin Study on Structural Hot-spot
and Effective Notch Stress Analysis, in: Guedes Soares,
C., Das, P. (Eds.), Advancements in Marine Structures.
Taylor & Francis Group, London, UK, pp. 159167.
Fricke, W., Petershagen, H., 1992. Detail Design of Welded
Ship Structures Based on Hot-Spot Stresses, in: Caldwell,
J.B., Ward, G. (Eds.), Proceedings of the Practical Design
of Ships and Mobile Units. Elsevier Science Limited,
pp. 10871100.
Garbatov, Y., Guedes Soares, C., 1998. Fatigue Reliability of
Maintained Welded Joints in the Side Shell of Tankers.
Journal of Offshore Mechanics and Arctic Engineering
120, pp. 29.
Garbatov, Y., Guedes Soares, C., 2002. Bayesian updating in the reliability assessment of maintained floating
structures. Journal of Offshore Mechanics and Arctic
Engineering 124, pp. 139145.
Garbatov, Y., Rudan, S., Guedes Soares, C., 2002a. Fatigue
Damage of Structural Joints Accounting for Non-linear
Corrosion. Journal of Ship Research 46, pp. 289298.
306
Garbatov, Y., Rudan, S., Guedes Soares, C., 2002b. Prediction of Fatigue Damage in Brackets of very Large
Scale Specimens of Ship Structure, Proceedings of the 8th
Potrtuguese Conference on Fracture, Vila Real, Portugal,
pp. 6169.
Garbatov, Y., Rudan, S., Guedes Soares, C., 2004. Assessment of geometry correction functions of tanker knuckle
details based on fatigue tests and finite-element analysis.
Journal of Offshore Mechanics and Arctic EngineeringTransactions of the Asme 126, pp. 220226.
Garbatov, Y., Rudan, S., Guedes Soares, C., 2010. Fatigue
Assessment of Welded Trapezoidal Joints of Very
Fast Ferry Subjected to Combined Load. Engineering
Structures 32, pp. 800807.
Gaspar, B., Garbatov, Y., Guedes Soares, C., 2011. Effect
of weld shape imperfections on the structural hot-spot
stress distribution. Ships and Offshore Structures 6,
pp. 145159.
Guedes Soares, C., Garbatov, Y., von Selle, H., 2003. Fatigue
Damage Assessment of Ship Structural Components
Based on the Long-term Distribution of Local Stresses.
International Shipbuilding Progress 50, pp. 3556.
Matoba, M., Kumano, A., Yamamoto, A., 1994. Based of
Design and Planing for Maintenance of Hull Planning of
Inspection and Life Design for Preventive. NK Technical
Bulletin, pp. 4763.
Niemi, E., Fricke, W., Maddox, S., 2004. Structural Stress
Approach to Fatigue Analysis of Welded Components
Designers Guide, IIW Doc. XIII-1819-00/XV-1090-01.
Nolte, K., Hansford, J., 1976. Closed-form expressions for

determining the fatigue damage of structures due to ocean
waves, Proceedings Offshore Technology Conference,
pp. 861870.
Palmgren, A., 1924. Die Lebensdauer von Kugellagern,
VDI-Zeitschrift 68, p. 339.
Petershagen, H., Fricke, W., Massel, T., 1991. Application of
the Local Approach to the Fatigue Strength Assessment
of Welded Structures in Ships, International Institute of
Welding.
Rudan, S., Garbatov, Y., Guedes Soares, C., 2000. Finite
Element Study of Stress Concentration Factors in ShipKnuckle Details, Proceedings of the 3rd International
Congress of Croatian Society of Mechanics, Dubrovnik,
Croatia, pp. 367374.
Saad-Eldeen, S., Nguyen, T., Garbatov,Y., Guedes Soares, C.,
2010. Fatigue Stress Assessment Accounting for Contact
Element, Proceedings of the International Conference on
Marine Science and Technology. Union of Scientists of
Varna, Varna, Bulgaria.
Xiao, Z.-G., Yamada, K., 2004. A Method of Determining Geometric Stress for Fatigue Strength Evaluation of
Steel Welded Joints. International Journal of Fatigue 26,
pp. 12771285.
Yuasa, M., Watanabe, T., 1999. Fatigue Strength of Bilge
Knuckle Structure, NK Technical Bulletin, pp. 1732.
307

Study on the ultimate torsional strength of ship hulls

B. Liu & C. Guedes Soares
ABSTRACT: A simple method based on plastic theory is proposed to study the ultimate torsional strength of
ship hulls. The plastic theory and a finite element analysis are used to analyze the collapse of a test specimen
in torsion so as to compare with the experimental results. A computer program using one-dimensional finite
element method is used to analyze the elastic behavior of the test specimen, and a nonlinear finite element
method is used to analyze in detail its plastic behavior. The study is mainly concentrated on the section that is
the transition from open to closed section. Special attention is paid on the comparisons of different calculation
results.
1
INTRODUCTION
Table 1.
The analysis of ultimate torsional strength is an important part of structural analysis for a ship hull with
large deck openings such as container ships and large
bulk carriers. However, there are not many studies
available on the ultimate strength of ship hulls under
torsion. Nowadays, ship structural design is mainly
based on the first yield of hull structures, with buckling
checks for structural components and ultimate strength
under vertical bending moment. For ship hulls with
large deck openings, especially for container ships,
the ultimate torsional strength is important.
Paik (2001) studied the ultimate strength of ship
hulls under torsion using the idealized structural unit
method (ISUM) to calculate the distribution and magnitude of warping and shear stresses for a typical
container vessel hull cross section under unit torsion.
Hu and Chen (2001) used the lower bound theorem
to study the ultimate torsional strength of ship hull
based on a section of the ship under free torsion. The
effect of constrained torsion on the strength of the joint
between the cargo hold and the engine room (or fore
peak) and the free torsion of the midship, needs to be
studied in some detail.
The ultimate torsional strength of ship hulls is a
very complex issue, including yielding, buckling and
rupture. Sun and Guedes Soares (2003) carried out
an experimental study of ultimate torsional strength
of a ship-type hull girder with a large deck opening
and presented the nonlinear finite element analysis to
compare with the experimental results.
Numerical and experimental methods are frequently used in the study on ultimate strength of
ship hulls. There are three kinds of numerical methods which are used to calculate ultimate strength of
ships: the idealized structural unit method (ISUM),
the method based on the Smith approach and the
nonlinear finite element method. The calculation of
Overall dimensions of test specimen.
Description
Value (mm)
Overall length, L
Breadth, B
Depth, H
Deck opening length, L0
Deck opening width, B0
Spacing of transverse frames
Spacing of stiffeners
6000
1250
500
4500
1000
750
250
ultimate strength is extremely complicated because of

the geometric and material nonlinearities. The general nonlinear finite element software around the
world includes ANSYS, MSC.Marc, MSC.Nastran,
ABAQUS, ADINA, which can also be used to calculate the ultimate torsional strength considered in this
paper.
In this paper, several methods are applied to calculate the strength of a torsion test specimen: onedimensional finite element method based on Vlasov
theory, three dimensional finite element method, simple method based on plastic theory and nonlinear finite
element method. Through the comparison with the
experimental results, advantages and disadvantages of
different calculation results are assessed.
ULTIMATE TORSIONAL STRENGTH

EXPERIMENT
Two test specimens with the same dimensions were

used to do ultimate strength experiments under pure
torsion, studying the possible failure modes of large
opening deck ships. Table 1 lists the main dimensions of the test specimen. The sketches of deck
plan, typical open and closed sections are shown in
309
thin-walled prismatic element is established according to Vlasov theory, and the global stiffness matrix
is established through the general procedure of finite
element method.
The stiffness matrices of individual elements of
the thin-walled beam could be assembled into a
global stiffness matrix after making coordinate transformation and considering warping compatibility
between adjacent elements. Then the matrix equation of one-dimensional finite element method can be
expressed to:
Figure 1. Deck plan sketch.
Figure 2. Typical sections of test specimen.
3.2 Geometrical properties of cross section

It is well known from Vlasov theory that normal
stresses z , shear stresses uniformly distributed
over the wall thickness and Saint Venant shear stresses
s are given by
Figure 3. Stress-strain curve of material.
Figures 12. The thickness of the plate is 3.0 mm,

and the sizes of stiffeners and transverse frames are
50 3 and 80 3 respectively. The stress-strain curve
is shown in Figure 3 and the nominal yield stress is
315 MPa.
The anti-symmetric torsional load was applied on
both sides of test specimen. The collapse took place
in the two regions nearby the discontinuous section
(from open section to closed section). The ultimate
torque was 48.75 kNm from the test (Sun and Guedes
Soares 2003).
in which:
N: normal force, Mx , My : bending moment with
respect to the x and y axis, B: bimoment, Tx , Ty : shear
force in the x and y direction, T : warping torsion
moment, and Ts : Saint Venant torsion moment, t: wall
thickness.
To obtain normal and shear stresses from Equation 2, the following parameters, which depend on
geometrical properties of cross section, must be calculated:
Area of cross section:
3 ELASTIC ANALYSIS OF TEST SPECIMEN

3.1 One-dimensional finite element method
The model of thin-walled beam is frequently used in
structural strength analysis of ship hulls with large
deck openings. The theory of thin-walled beams developed by Vlasov (1961) has many engineering applications. According to the theory, the cross-section would
warp when the thin-walled beam is under torsion,
and the warping stress would arise when the warping
deformation is restrained.
The theory of one-dimensional finite element
method proposed by Gunnlaugsson and Pedersen
(1982) is to divide the beam with varying crosssection into a number of thin-walled prismatic elements along the length. The stiffness matrix of the
Moment of inertia about principal centroidal x-axis:
Moment of inertia about principal centroidal y-axis:
Generalized warping function:
310
Table 2.
Section properties of open section, part 1.
The area of open section:

The coordinates of
centroid:
The coordinates of
torsional center:
Table 3.
Figure 4. Determination of generalized warping function.
Principal sectorial moment of inertia (warping

constant):
A = 8850 mm2
xc = 625 mm, yc = 142.54 mm
xp = 625 mm, yp = 223.34 mm
Section properties of open section, part 2.
The moment of inertia about

principal centroidal x-axis:
principal centroidal y-axis:
Torsion (Saint Venant) constant:
Principal sectorial moment of
inertia (warping constant):
Ixx = 2.1003 109 mm4

Iyy = 2.8315 108 mm4
J = 25650 mm4
I = 8.9763 1013 mm6
Static moment of area about the principal centroidal

x-axis:
Static moment of area about the principal centroidal

y-axis:
Figure 5. Coordinate axis of open section.
Static warping function moment:
The torsional center can be determined by satisfying

the conditions:
Figure 6. Warping function of open section.
The determination of the generalized warping function

is shown in Figure 4. Using the equations mentioned
above, a computer program written in Matlab is used to
calculate geometrical properties of thin-walled beams
with open-closed section.
3.2.1 Open section
The section properties are calculated with an arbitrary
set of axis xoy, as shown in Table 2. The coordinate
axis of open section is shown in Figure 4.
A new reference coordinate system xCy is applied to
calculate other section properties, as shown in Table 3.
The following results are based on the coordinate
system xCy.
The largest warping shown in Figure 6 is at the open

edge. The warping function is an important parameter
to analyze the warping stress under constrained torsion. The distribution of warping stress in the open
section of test specimen is the same with the distribution of warping function. It is convenient to analyze
the stress level in the elastic range.
The shear stresses shown in Figure 7 are calculated
by:
The largest shear stress is at the center of the side plate.

The distribution of shear stresses in open sections can
311
Figure 7. Shear stresses of open section due to torsion

moment: warping.
Figure 9. Warping function of closed section.
Figure 10. Shear stresses of closed section due to torsion

moment: Saint Venant.
Figure 8. Coordinate axis of closed section.

Table 4.
Section properties of the closed section, part 1.
The area of closed section:

The coordinates of centroid:
The coordinates of torsional
center:
Table 5.
A = 12300 mm2
xc = 625 mm, yc = 250 mm
xp = 625 mm, yp = 250 mm
Section properties of the closed section, part 2.

principal centroidal x-axis:
principal centroidal y-axis:
Principal sectorial moment
of inertia (warping constant):
Ixx = 2.4440 109 mm4
Figure 11. Shear stresses of closed section due to torsion

moment: warping.
Iyy = 6.0751 108 mm4
stiffness of an open section, so it is not necessary to

compare the stress level in the closed section with
that of open section. Shear stresses due to Saint
Venant torsion and warping torsion are shown in
Figures 1011.
J = 1.3393 109 mm4

I = 1.7 1013 mm6
be used to analyze the shear stress level, compared

with the warping stress. But for the test specimen, the
shear stress is much smaller than the warping stress.
So, the shear stress is not the main stress.
3.2.2 Closed section
The coordinate axis of closed section is shown in
Figure 8. The following results are based on coordinate
axis xoy, as shown in Table 4.
The results shown in Table 5 are based on the
coordinate system xCy.
The warping function shown in Figure 9 can be
used to analyze the warping stress of closed sections
under constrained torsion. The torsional stiffness of
the closed section is much larger than the torsional
3.3 Stress calculation results and yield load

Hu (1996) proposed a finite element model for static
and dynamic analysis of thin-walled beams with asymmetric cross-sections. One-dimensional finite element
method is used to calculate torsional angle, warping
function and bimoment. It is calculated and drawn
in Matlab. When applying a torque of 1 kNm at both
sides, the calculation results are shown in Figure 12.
The largest bimoment is 2.2494 109 N mm2 at the
discontinuous section (from open section to closed
section). The associated largest warping stress is:
312
Figure 12. Torsional angle bimoment, torsional function and torsional moment.
The test specimen would reach to the yield load when

reaches to 315 MPa. As the one-dimensional finite
element method is based on linear elastic theory, it is
easy to obtain the yield load based on the result above
mentioned.
This load calculated by one-dimensional finite element

method is the yield load.
The bimoment in the middle of test specimen
is zero, so the load of the constrained torsion is
0.9973 kNm when applying a torque of 1 kNm at both
sides with length of 2550 mm.
PLASTIC ANALYSIS OF TEST SPECIMEN
4.1 A simplified form of yield criterion

For this model, the warping stress is the only normal
stress, and the shear stress due to free torsion and warping torsion is the shear stress. The Von Mises stress
yield criterion is:
In this case, the load of free torsion is much smaller

than the load of constrained torsion, and the shear
stress is also much smaller than the normal stress. In
order to conveniently analyze the specimen, the normal
stress yield criterion under constrained torsion is used
in this analysis. The normal stress yield criterion is:
In the elastic range, the warping stress, the shear stress

due to free torsion and the shear stress due to warping
torsion are shown in above. The error is small compared with the Von Mises stress yield criterion. So, the
normal stress yield criterion is used to analyze the test
specimen in this paper.
The load of free torsion is:
The shear stress due to free torsion is:
4.2 Plastic torsional center of cross section

Under pure torsion, composite force and composite
moment of ultimate normal stress should be zero, so:
From the Figure 7, when the load is 0.9973 kNm, the
largest shear stress due to warping torsion is:
313
Figure 14. Plastic warping stress distribution.
Figure 13. Boundary conditions of simple model.
The three equations can be used to calculate the sectorial zero point and torsional center of cross section. For
symmetrical section, the sectorial zero point and the
torsional center are located at symmetry axis. Thus,
Equations 2223 are always satisfied because of symmetry, and the torsional center can be calculated only
by Equation 24.
4.3
Simplified load model and its boundary

conditions
For the specimen, the torsional stiffness of the open

section is much smaller than that of the closed section. If torsional moments are applied at both sides,
the open section would twist much more due to its low
torsional stiffness. In order to simplify the model, a
rigid plane is used to replace the closed section. The
boundary conditions of the simple model are shown in
Figure 13. This method only analyzes the open section,
not considering the influence of transverse frame.
4.4
Figure 15. Plastic warping function.
in the buckling of side plate and the collapse of

model.
Using the warping stress distribution, the plastic
torsional center can be calculated by Equation 24. The
coordinates of plastic torsional center are:
xp = 625 mm, yp = 220.1 mm
The plastic warping function can be calculated by
Equation 6 and the coordinates of plastic torsional
center, are shown in Figure 15. The bimoment can be
calculated by the following equation:
Ultimate strength calculation result
Trahair (1999) studied the plastic torsion analysis of

mono-symmetric beams. The method is only suited
for beams with small section, because the stress of all
members in a section must arrive at the plastic yield.
Actually, the main cause of the collapse of the specimen is buckling, and only parts of a section reach the
plastic yield. One side would be in compression and the
other side would be in tension when test specimen is
under pure torsion. So, when the test specimen arrived
at its ultimate strength, only the stress of partial members would reach to the yield stress. It can refer to the
warping function of open section. The larger the value
of warping function, the larger the value of warping
stress.
Suppose that the upper plate is buckling when the
stress in the half of upper plate arrives at yield stress.
The plastic warping stress distribution is shown in
Figure 14. The buckling of the upper plate results
The ultimate torsional moment can be calculated by

the following equation:
NONLINEAR FINITE ELEMENT ANALYSIS

OF TEST SPECIMEN
5.1 General
Nonlinear finite element method in ANSYS, is used
to analyze the ultimate strength of the specimen
considering geometric and material nonlinearity. The
arc-length method is used to analyze the finite element
models, and the solution includes large displacements,
314
Figure 16. Boundary conditions of the test specimen.

Figure 18. Von-Mises stress of test specimen under ultimate
load.
Figure 17. Boundary conditions of the simple model.
finite strains and followed forces. Elasticperfectly

plastic material with yield stress of 315 MPa, Youngs
modulus of 210 GPa, and Poissons ratio of 0.3 is used
in the finite element models.
5.2
Boundary conditions
All DOFs (degrees of freedom) are entirely constrained at the unloaded end of the numerical model,
and multi-point constraints are used to constrain all
DOFs of the loaded end, limiting the loaded end plane
nodes to follow the displacement of a reference node.
The nodes at the loaded end section are able to translate
freely in the sections plane, and the reference node is
the centre of the end section. The torsional moment is
applied on the reference node.
The finite element model of test specimen consists
of 8787 nodes, 7360 shell elements and 1542 beam
elements, and the boundary conditions of test specimen are shown in Figure 16. The finite element model
of the simple model consists of 5684 nodes, 4680 shell
elements and 1032 beam elements, and the boundary
conditions of simple model are shown in Figure 17.
5.3 Analytical results
Calculation results of the ultimate load for the simple
model are close to that for test specimen. The error
is 1.9%. The ultimate load is the total reaction forces
Figure 19. Von-Mises stress of simple model under ultimate

load.
of restraint nodes. The ultimate load of the specimen

and the results of the simple model are 54.89 kNm and
55.92 kNm, respectively.
The simple model is suitable to calculate the ultimate strength of the specimen, only analyzing the
cargo hold region. Von Mises stress distributions of the
specimen and of the simple model at the ultimate limit
state are shown in Figures 1819, respectively. This
method is adequate to calculate the ultimate strength
of container ship, only modeling the open section.
6
COMPARISON OF RESULTS
The failure region of the test specimen is nearby the

discontinuous section, and all the calculation results
are consistent in the experimental failure region. The
failure regions are in high stress and compressive
regions. The stress distribution of the finite element
model is basically consistent with the test result.
All calculation results are shown in Table 5. The
yield load calculated by one-dimensional finite element method is more than the yield load of test, and
the error is 7.7%. The yield load calculated by the
315
Table 6.
Comparison of all calculation results.
Yield load
Test result from paper (Sun):
One-dimensional FEM result:
FEM result:
Ultimate load
Test result from paper (Sun):
NFEM result from paper (Sun):
NFEM result of test specimen:
NFEM result of simple model:
Plastic result of simple model:
Error
44.32 kNm
47.74 kNm
41.56 kNm
7.7%
6.2%
48.75 kNm
54.58 kNm
54.89 kNm
55.92 kNm
55.81 kNm
12.0%
12.6%
14.7%
14.5%
finite element method is less than the yield load test

result, and the error is 6.2%. Table 5 shows the different results obtained with the different methods, and it
is very useful to assess the calculation results through
comparing with experimental results.
All the ultimate loads calculated by different methods are larger than the experimental ultimate load.
Without considering welding residual stress, the calculated value of the finite element model would be
slightly larger than the experimental value.
Sun and Guedes Soares (2003) have studied the
influence of welding on the ultimate strength. The
intermittent welding of the stiffeners to the plate could
reduce the compressive resistance of the test specimen
they tested. The ultimate torque with 50% effective
stiffness of the stiffeners is well in agreement with the
experimental value. So, all the calculation results of
ultimate load are well in keeping with the test results,
if the full weld is used in the test specimen.
The simple model is easier to calculate the ultimate strength, and it shows the distribution of warping
stress under ultimate load. All the analyses can well
describe the collapse of the test specimen, and the
analysis results could expand to ship hull with large
deck openings, such as container ship.
7
The ultimate strength of ship hulls subject to torsion

moment is governed by buckling of the compression
plates and yielding of the tension plates, which near
the joint between cargo hold and engine room (or fore
peak).
The hull would reach its ultimate limit state immediately after the two side plates nearby the discontinuous
section (from open section to closed section) collapse.
It is effective to use the simple plastic method to
analyze the discontinuous section of the specimen.
This method can analyze the stress distribution under
ultimate load, and it is good to analyze the ultimate
strength of container ship.
The comparisons of all results show the precision of
calculation methods. They are useful to study ultimate
strength under torsion and to design container ship.
REFERENCES
Gunnlaugsson G. A., Pedersen P. T. 1982. A finite element
formulation for beams with thin walled cross-sections.
Computer & Structures 15(6).
Hu Y. R., Chen B. Z. 2001. Limit State of Torsion of Ship
Hulls with Large Hatch Openings. Journal of Shanghai
Jiaotong University Vol. 35, No. 4.
Hu Y. R., Jin X. D, Chen B. Z.. 1996. A finite element model
for static and dynamic analysis of thin-walled beams
with asymmetric cross-sections. Computers & Structures.
Vol. 61, No. 5, pp. 897908.
Paik J. K., Thayamballi A. K., Pedersen P. T., Park Y. 2001.
Ultimate strength of ship hulls under torsion. Ocean
Engineering 28: 1097133.
Sun H. H., Guedes Soares C. 2003. An experimental study of
ultimate torsional strength of a ship-type hull girder with
a large deck opening. Marine Structures 16: 5167.
Trahair N. S. 1999. Plastic torsion analysis of monosymmetric and point-symmetric beams. Journal of Structural
Engineering Volume 125, Issue 2, pp. 175182
VlasovV. Z. 1961.Thin-walled elastic beams. Natural Science
Found. Washington. U.S.A.
CONCLUSIONS
The study suggests that it is convenient to only use

the open portion to calculate the ultimate strength
of container ship using the nonlinear finite element
method.
316

A new trigonometric shear deformation theory for sandwich and

composite laminated plates
ABSTRACT: A new trigonometric shear deformation theory for sandwich and composite plates is developed.
The proposed displacement field, which is m parameter dependent, is assessed by performing several computations of the plate governing equations. The theory accounts for adequate distribution of the transverse shear
strains through the plate thickness and tangential stress-free boundary conditions on the plate boundary surface,
thus a shear correction factor is not required. Plate governing equations and boundary conditions are derived by
employing the principle of virtual work. The exact solutions for bending analysis are presented for sinusoidally
and uniformly distributed loads. The accuracy of the present theory is ascertained by comparing it with various
available results in the literature.
INTRODUCTION
Multilayer structures are alternative materials used

extensively in many fields such as in aerospace, naval
industry and semiconductor technologies, for high
performance and reliability (Chamis 2007). Example
of multilayer structures are sandwich constructions,
composite structures made of orthotropic lamina, layered structures made of different isotropic layers. In
the last decades the use of sandwich construction
grew rapidly around the world. Its advantages, the
development of new materials and the necessity of
high performance under static, dynamic and thermal loads guarantee that the sandwich structures will
be in demand for many years (Vinson 1999, 2001,
2005). With the increased use of sandwich structures,
there is a tremendous need to develop efficient manufacturing techniques, economical and effective repair
techniques, and analysis methods to predict the short
and long-term behavior of the multilayer composite
materials under a variety of loading and environmental
conditions.
In the literature, different models were proposed
in order to study the multilayered composite structures such as equivalent single layer, quasi-layerwise
and layerwise models (Demasi 2009a, b). The heterogeneous laminated plates and shells are treated as a
statically equivalent single layer, thus reducing the
3D problem to a 2D one. Another important point
in the analysis of composite structures is the variational statement used in the analysis to derive the
necessary governing equations and the boundary conditions. Deformation theories can either be developed
using displacement-based theories (when the principle
of virtual displacement is used), stress based theories

or displacement-stress-based theories (when Reissner
mixed variational theorem is used, see Carrera (2000,
2001, 2002) and Demasi (2006, 2009a, b, c, d, e). The
well-described unified formulation, initially presented
by Carrera (2003) and recently extended by Demasi
(2006, 2008, 2009a, b, c, d, e), describes precisely and
clearly the models, types and class of theories.
According to the unified generalized formulation proposed by Demasi (2009a), among equivalent
single layer theories, there are a number of theories.
However, there are mainly three well-know major theories; namely the classical lamination theory (CLT),
which is based on the assumptions of Kirchhoffs
plate theory (Reissner 1961; Whitney 1970; Ashton
1970), which neglects the interlaminar shear deformation, the first order shear deformation theory (FSDT)
(Dong 1972; Chou 1973; Reissner 1975), assumes
constant transverse shear deformation through the
entire thickness of the laminate and violates stress
free boundary conditions at the top and bottom surfaces of the plate, and more accurate theories such
as higher order theories (HSDT) assume quadratic,
cubic or higher (also non-polynomial, e.g. trigonometric) variations of surface-parallel displacements
through the entire thickness of the laminates to model
the behavior of the structure for thick to thin regions
(Murthy 1981; Reddy 1985; Murakami 1986; Kant
2002; Swaminathan 2004). In fact, few higher order
shear deformation theories containing non-polynomial
shape strain functions were developed. For example,
the unknowns in generalized formulations presented
by Carreara (2003) and Demasi (2009e) are expanded
along the thickness by using chosen polynomial
317
functions. They recommend a series of trigonometric functions of the thickness coordinate z. However,
according to the authors knowledge, regarding to
the equivalent single layer theories, there is only
one trigonometric shear deformation theory developed
by Touratier (1991), in which the sinus function was
introduced. Therefore, it is still important to exploit
the behavior of other functions in the implementation
of new shear deformation theories.
The unified formulation of Carrera applied to
advanced composite plates by using meshless methods was recently extended to include non-polynomial
shape strain functions in their formulation (Neves
et al. 2011), and the need of new non-polynomial
shape strain functions, which can be adapted to this
compact generalized formulation, is demanding. The
present new HSDT can be also adapted to the compact
CUF or GUF in similar manner as in Ferreira et al.
(2011).
In the present work, a new trigonometric higher
order displacement field in which the displacement
of the middle surface expanded as tangential trigonometric functions of the thickness coordinates and the
transverse displacement taken to be constant through
the thickness is proposed. Necessary equilibrium
equations and boundary conditions are derived by
employing the principle of virtual work. The theory
accounts for adequate distribution of the transverse
shear strains through the plate thickness and the tangential stress-free boundary conditions on the plate
boundary surface, therefore a shear correction factor is not required. Exact solutions for deflections
and stresses of simply supported sandwich plates are
presented. The accuracy of the present theory is ascertained by comparing it with Pagano (1970), Pagano
and Hatfield (1972), Srinivas (1973), other analytical
solutions, and also with various numerical calculations such as finite element (Pandya & Kant 1988;
Ferreira & Barbosa 2000) and meshless methods
(Ferreira 2003, 2005; Xiang 2009).
2
Figure 1. Laminate geometry with positive set of lamina/

laminate reference axes, displacement components and fiber
orientation.
trigonometric shape strain functions are richer than

polynomial ones (Karama et al. 2009). Following similar procedures as presented by Reddy et al. (1985)
or the generalized procedure developed by Soldatos
(1992), the new displacement field is obtained as
follows:
HIGHER ORDER DISPLACEMENT THEORY
Lets assume with the following displacement field to

begin the derivation of the theory:
Therefore, by comparing the Equations (2ab) to
(1ab), f (z) becomes,
where u(x, y), v(x, y), w(x, y), 1 (x, y) and 2 (x, y) are
the five unknown functions of middle surface of the
plate as in Figure 1, while f (z) represents shape functions determining the distribution of the transverse
shear strains and stresses along the thickness.
In this paper, as mentioned above a new trigonometric shear deformation theory is introduced. Because
The function f (z) in Equation 3 is m dependent

and therefore it needs to be selected or calculated.
It is obtained after several computations of the plate
governing Equations (4ae), in order to obtain the
maximum shear stresses which give close results compared to 3D elasticity bending solutions. Selection of
the parameter will be omitted for the sake of brevity.
318
Considering the static version of the principle of

virtual work, the static governing equations of the plate
are derived and they are given as follows:
where
Substituting Equations (7ae) into Equations (4ae),

following equations are obtained,
Elements of Kij in Equation (9a) can be obtained by

using a unified compact formulation used by Mantari
(2011).
where q is the load term, Ni , Mi , Pi , Qi and Ki are the
resultants calculated based on the proposed displacement field. The derivation of the laminate stiffnesses
(Aij , Bij , Dij , Eij , Fij and Hij ), not presented here and it
can be found in Reddy et al. (1985) or Mantari et al.
(2011).
In what follows, the simply supported boundary
conditions are prescribed at all four edges:
where Qmn are the coefficients in the double Fourier

expansion of the transverse load,
NUMERICAL RESULTS AND DISCUSSION
4.1 Selection of the parameter m in f(z) shape

strain function
3
SOLUTION PROCEDURE
The Navier solution exists if the following stiffnesses

are zero:
Solution functions to the partial differential Equations

(4ae) of a cross-ply plate for simple supported boundary conditions given by Equations 5 are assumed as
follows:
In Section 2, the governing equations are formulated with y, and therefore m dependency. From
Equation 3, it can be noticed than m is directly
proportional to . For convenience, will be used
to get the f (z) shape strain function. The unknown
parameter of the present higher order theory is
obtained by providing the maximum shear stresses
for all values of a/h ratios after several computations
of the plate governing equations.
Following the determination of the shear strain
shape function f (z) as given in Equations (11ab),
some example problems for laminated composites are
solved in order to show the validity and accuracy of
the present deformation theory.
The exact solutions of symmetric and antisymmetric

cross-ply plates are calculated by the present and various higher order shear deformation theories under
sinusoidally and transversely distributed loads for a
simply supported plate on all edges. The results are
then discussed in the following sections.
319
Table 1. Non-dimensionalized deflections and stresses in

three-layer (0 /90 /0 ) square laminates under sinusoidal
load (b = a).
Method
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
4.2
a/h
4
10
20
50
100

three-layer (0 /90 /0 ) square laminates under sinusoidal
load (b = 3a).
xx
yz
Method
1.9222
1.9218
0.7131
0.7125
0.5049
0.4439
0.4351
0.4342
0.755
0.733
0.734
0.590
0.568
0.568
0.552
0.546
0.541
0.540
0.539
0.539
0.539
0.217
0.183
0.183
0.123
0.103
0.103
0.094
0.08
0.084
0.076
0.083
0.075
0.075
3D (Pagano)
Present
Touratier
Reddy et al.
3D (Pagano)
Present
Touratier
Reddy et al.
3D (Pagano)
Present
Touratier
Reddy et al.
3D (Pagano)
Present
3D (Pagano)
Present
Reddy et al.
Bending analysis of cross-ply laminated

composite plates
4.2.1 Three-layer symmetric cross-ply (0 /90 /0 )

plates, under sinusoidal load
Pagano (1970) provided exact solutions for one square
(b = a) and one rectangle (b = 3a) plate for threelayer symmetric cross-ply (0 /90 /0 ) lamination and
simply supported on all edges.
The mechanical properties of each layer are as
follows (Pagano, 1970):
10
20
50
100
xx
yz
2.820
2.6415
2.6657
2.6411
0.9190
0.8631
0.8698
0.8622
0.6100
0.5948
0.5958
0.5937
0.5200
0.5190
0.5080
0.5081
0.5070
1.100
1.034
1.067
1.036
0.725
0.692
0.698
0.692
0.650
0.641
0.642
0.641
0.628
0.626
0.624
0.624
0.624
0.033
0.035
0.036
0.035
0.015
0.017
0.017
0.017
0.012
0.014
0.014
0.014
0.011
0.013
0.011
0.013
0.013

four-layer (0 /90 /90 /0 ) square laminates under sinusoidal
load (b = a).
Method
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
3D (Pagano)
Present
Reddy et al.
The following normalized quantities are defined for

deflection and stresses:
The results of the present theory and other theories

such as the one proposed by Reddy et al. (1985)
and Touratier (1991) are compared with the threedimensional elasticity results given by Pagano (1970).
The results for square plates are presented inTable 1.
It is evident from Table 1 that the present method
gives more accurate results for normal stresses than
in shear stresses for thick and thin plates, because it
is an equivalent single layer theory. However, a layerwise description of this theory may be more accurate.
Results for the rectangular plates (b = 3a) are given in
Table 2. Similar conclusions, compared with the square
plate, can be inferred.
a/h
a/h
4
10
20
100
xx
yz
1.954
1.894
1.893
0.743
0.715
0.715
0.517
0.507
0.506
0.439
0.435
0.434
0.720
0.664
0.665
0.559
0.545
0.546
0.543
0.539
0.539
0.539
0.539
0.538
0.291
0.239
0.239
0.196
0.153
0.153
0.156
0.123
0.123
0.141
0.112
0.112
4.2.2 Four-layer symmetric cross-ply

(0 /90 /90 /0 ) plates under sinusoidal load
In this section, four layer symmetric (0 /90 /90 /0 )
cross-ply square plates (b = a) are considered. Exact
3D solution is also obtained from Pagano and Hatfield
(1972). The same material is used in the calculations. Results provided by Reddy et al. (1985) are
compared with the three-dimensional elasticity theory
given by Pagano and Hatfield (1972). Normally, the
efficiency of the different models can be checked in
several cases such as thick laminates (Karama 2009).
In this section, the results are presented in Table 3. It
shows that the present results are in good agreement
with 3D-elasticity solution in deflection, and normal
stresses. However, there is a considerable difference
with 3D-elasticy solution for xz stress.
320
theory is ascertained by comparing it with various

available results in the literature. The results show that
the present model performs as good as the Reddys
and Touratiers shear deformation theories for analyzing the static behavior of multilayered sandwich
and composite plates. It is believed that the present
theory may be extended to cover multilayered shell
structures, higher order layer-wise shear deformation
theories and advanced numerical calculations, such as
finite element and meshless methods.
ACKNOWLEDGEMENTS
Figure 2. Maximum central plane deflection in two-layer

(0 /90 ) square laminates under sinusoidal load as a function
of span-to-depth ratio (a = b).
The first and second authors have been financed

by the Portuguese Foundation of Science and Technology (Fundao para a Cincia e a Tecnologia)
under contract numbers SFRH/BD/66847/2009 and
SFRH/BPD/47687/2008 respectively.
REFERENCES
Figure 3. Maximum central plane deflection in two-layer

(0 /90 ) square laminates under uniform load as a function
of span-to-depth ratio (a = b).
4.2.3 Two-layer antisymmetric cross-ply (0 /90 )

plates under sinusoidal and uniform load
In this section, the two-layer antisymmetric cross-ply
(0 /90 ) plates under sinusoidal load are also calculated for a/h = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20,
25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90,
95, 100}. The deflections are in good agreement with
Pagano (1970) for sinusoidally distributed load. They
are also in good agreement with Reddy et al. (1985)
for uniformly distributed load as given in Figures 2
and 3, respectively.
CONCLUSIONS
A new trigonometric higher order shear deformation

theory of sandwich and composite plates is presented.
The theory accounts for adequate distribution of the
transverse shear strains through the plate thickness
and tangential stress-free boundary conditions on the
plate boundary surface, therefore a shear correction
factor is not required. The accuracy of the present
Ashton, J.E. 1970. Anisotropic plate analysis-boundary

conditions. J Compos Mater 4:18291.
Arya, H., Shimpi, R.P., Naik, N.K. 2002. A zigzag model
for laminated composite beams. Compos Struct 56:214.
Aydogdu, M. 2009. A new shear deformation theory for
laminated composite plates. Compos Struct 89:94101.
Carrera, E. 2000. An assessment of mixed and classical
theories on global and local response of multilayered
orthotropic plates. Compos Struct 50:183198.
Carrera, E. 2001. Developments, ideas, and evaluations
based upon Reissners mixed variational theorem in the
modeling of multilayered plates and shells. Appl Mech
Rev 54:30129.
Carrera, E. 2002. Theories and finite elements for multilayered, anisotropic, composite plates and shells. Arch
Comput Meth Engng 9(2):87140.
Carrera, E. 2003. Theories and finite elements for multilayered plates and shells: a unified compact formulation
with numerical assessment and benchmarks. Arch Comput
Methods Eng 10:215296.
Chamis, C.C. 2007. Probabilistic Design of Compos Struct.
ICCES 4(2):5966.
Chou, P.C., Carleone, J. 1973. Transverse shear in laminated
plate theories. AIAA J 11:133336.
Demasi, L. 2006. Treatment of stress variables in advanced
multilayered plate elements based upon Reissnerss mixed
variational theorem. Compos Struct 84:12151221.
Demasi, L. 2008. 3 Hierarchy plate theories for thick and
thin composite plates: The generalized unified formulation. Compos Struct 84:256270.
Demasi, L. 2009a. 6 Mixed plate theories based on
the Generalized Unified Formulation. Part I: Governing
equations. Compos Struct 87:111.
Demasi, L. 2009b. 6 Mixed plate theories based on the
Generalized Unified Formulation. Part II: Layerwise
theories. Compos Struct 87:1222.
Demasi, L. 2009c. 6 Mixed plate theories based on the
Generalized Unified Formulation. Part III: Advanced
mixed high order shear deformation theories. Compos
Struct 87:183194
Demasi, L. 2009d. 6 Mixed plate theories based on the
Generalized Unified Formulation. Part IV: Zig-zag
theories. Compos Struct 87:195205.
321
Demasi, L. 2009e. 6 Mixed plate theories based on the

Generalized Unified Formulation. Part V: Results.
Compos Struct 88:116
Dong, S.B., Tso, F.K.W. 1972. On a laminated orthotropic
shell theory including transverse shear deformation.
J Appl Mech 39:109196.
Ferreira, A.J.M., Barbosa, J.T. 2000. Buckling behavior of
composite shells. Compos Struct 50:938.
Ferreira, A.J.M., Roque, C.M.C., Martins, P.A.L.S. 2003.
Analysis of composite plates using higher-order shear
deformation theory and a finite point formulation
based on the multiquadric radial basis function method.
Composites: Part B 34:62736.
Ferreira, A.J.M., Roque, C.M.C., Jorge, R.M.N. 2005. Analysis of composite plates by trigonometric shear deformation
theory and multiquadrics. Compos Struct 83:222537.
Ferreira, A.J.M., Carrera, E., Cinefra, M., Roque, C.M.C.,
Polit, O., 2011. Analysis of laminated shells by a sinusoidal shear deformation theory and radial basis functions collocation, accounting for through-the-thickness
deformations. Composites: Part B 42:12761284.
Idibi, A., Karama, M., Touratier, M. 1997. Comparison of
various laminated plate theories. Compos Struct 37:
173184.
Kant, T., Swaminathan K. 2002. Analytical solutions for the
static analysis of laminated composite and sandwich plates
based on a higher order refined theory. Compos Struct
56:32944.
Karama, M., Afaq, K.S., Mistou, S. 2003. Mechanical behavior of laminated composite beam by the new multilayered
laminated Compos Struct model with transverse shear
stress continuity. Int J Solid Struct 40(6):15251546.
Karama, M., Afaq, K.S., Mistou, S. 2009. A new theory
for laminated composite plates. In: Proc. IMechE 2009
Vol. 223 Part L: J. Materials: Design and Applications.
Mantari, J.L., Oktem, A.S., Guedes Soares, C. 2012. A
new trigonometric shear deformation theory for isotropic,
laminated composite and sandwich plates 49:4353.
Murakami, H. 1986. Laminated composite plate theory with
improved in-plane responses. J Appl Mech 53:66166.
Murthy, M.V. 1981. An improved transverse shear deformation theory for laminated anisotropic plates. NASA
Technical Paper 1903.
Neves, A.M.A., Ferreira, A.J.M., Carrera, E., Roque, C.M.C.,
Cinefra, M., Jorge, R.M.N., Soares, C.M.M., 2011.
Bending of FGM plates by a sinusoidal plate formulation
and collocation with radial basis functions. Mechanics
Research Communications 38:368371.
Pagano, N.J. 1970. Exact solutions for rectangular
bidirectional composites and sandwich plates. Journal
of composite materials 4:2034.
Pagano, N.J., Hatfield, S.J. 1972. Elastic behavior of multilayered bidirectional composites. AIAA J 10:9313.
Pandya, B.N., Kant, T. 1988. Higher-order shear deformable
theories for flexure of sandwich plates-finite element
evaluations. Int J Solids Struct 24:41951.
Reddy, J.N., Liu, C.F. 1985. A higher-order shear deformation

theory of laminated elastic shells. Int J Eng Sci 23;31930.
Reissner, E., Stavsky, Y. 1961. Bending and stretching of
certain types of heterogeneous aelotropic elastic plates.
J of Appl Mech 28(3):40208.
Reissner, E. 1975. On transverse bending of plates, including
the effect of transverse shear deformation. Int J Solids and
Struct 11:56973.
Roque, C.M.C., Ferreira, A.J.M., Jorge, R.M.N. 2005.
Modeling of composite and sandwich plates by a trigonometric layerwise deformation theory and radial basis
functions. Composites: Part B 36:559572.
Soldatos, K.P. 1992. A transverse shear deformation
theory for homogeneous monoclinic plates. Acta
Mechanica 94:195220.
Shimpi, R.P., Ghugal, Y.M. 1999. A layerwise trigonometric shear deformation theory for two layered cross-ply
laminated beams. Journal of Reinforced Plastics and
Composites 18:151642.
Shimpi, R.P., Ghugal, Y.M. 2001. A new layerwise trigonometric shear deformation theory for two-layered cross-ply
beams. Composites Science and Technology 61:1271
1283.
Shimpi, R.P., Aynapure, A.V. 2001. A beam finite element based on layerwise trigonometric shear deformation
theory. Compos Struct 53:153162.
Srinivas, S. 1973. A refined analysis of composite laminates.
J Sound Vibr 30:495507.
Swaminathan, K., Ragounadin, D. 2004. Analytical solutions
using a higher-order refined theory for the static analysis of antisymmetric angle-ply composite and sandwich
plates. Compos Struct 64:40517.
Touratier, M. 1991. An efficient standard plate theory. Int J
Eng Sci 29(8):901916.
Vidal, P., Polit, O. 2008. A family of sinus finite elements
for the analysis of rectangular laminated beams. Compos
Struct 84:5672.
Vinson, J.R. 1999. The behavior of sandwich structures of
isotropic and composite materials. Lancaster, PA USA:
THECHNOMIC Publishing Company Inc.
Vinson, J.R. 2001. Sandwich structures. Appl Mech Rev
54(3):201214.
Vinson, J.R. 2005. Sandwich Structures: Past, Present, and
Future. In: Proceedings of the 7th International Conference on Sandwich Structures, Aalborg University,
Aalborg, Denmark, p.312.
Whitney, J.M., Leissa, A.W. 1970. Analysis of simply
supported laminated anisotropic plates. AIAA J 8: 2833.
Xiang, S., Wang, K., Ai, Y., Sha, Y., Shi, H. 2009. Analysis
of isotropic, sandwich and laminated plates by a meshless
method and various shear deformation theories. Compos
Struct 91:3137.
322

Layerwise finite element formulation for the analysis of laminates and

sandwich panels
ABSTRACT: A layerwise finite element formulation of a newly developed layerwise theory for the flexure of
thick multilayered panels is presented. The displacement model accounts for non-linear and constant variation of
in-plane and transverse displacement through the panel thickness. Reissner-Mindlin approach is used to model
the faces of the sandwich structure. The displacement field for the core is modeled as quadratic function of the
thickness coordinate. The three dimensional problem is reduced to the two-dimensional one by through-thickness
integration of the strain energy. The discrete element chosen is a four-nodded quadrilateral with nine-degrees-offreedom per node (when shear continuity is imposed). An example problem is provided for a well-know sandwich
composite plate. The accuracy of the present code is ascertained by comparing it with various available results
in the literature.
1
INTRODUCTION
Advanced composites materials have created a revolution in high performance structures. Their advantages
relative to conventional materials such as high strength
to weight and stiffness to weight ratios, superior resistance to environmental conditions, design flexibility,
make them attractive for a wide range of applications in marine, aerospace, automotive, and chemical,
industries (Swanson 1997; Jones 1999). Especially,
the recent development in the military ships shows
that composite structures can be used to increase the
operational performance to reduce maintenance and
fuel consumption costs (Mouritz et al. 2001). Recent
advances in laminated composite and sandwich materials make reliable to build naval ships with more than
100 m (Mouritz et al. 2001; Kontastinos 2001).
However, the analysis of composite structures
brings additional difficulties to the researchers such
as the transverse or inter-laminar shear stress due
to mismatch of material properties among layers,
bending-stretching coupling due to asymmetry of
lamination, and in-plane orthotropy. Therefore using
such advanced materials require an in-depth understanding of the response characteristics of these
structural components.
The finite element method (FEM) has become the
standard method of analysis of complex and realistic structures. Finite element numerical codes can be
implemented by using classical, first or higher order
shear deformation theories, and it can be done also
adopting equivalent single layer (ESL) or layerwise
description of the plate. Among ESL theories, there
are mainly three major theories. The classical lamination theory (CLT) is based on the assumptions
of Kirchhoffs plate theory (Reissner 1961; Whitney

1970; Ashton 1970), and neglects the interlaminar
shear deformation. The first order shear deformation
theory (FSDT) (Dong 1972; Chou 1973; Reissner
1975), assumes constant transverse shear deformation
through the entire thickness of the laminate and violates stress free boundary conditions at the top and
bottom surfaces of the panel.
More accurate theories such as higher order
theories (HSDT) assume quadratic, cubic or higher
variations of surface-parallel displacements through
the entire thickness of the laminates to model the
behavior of the structure for thick to thin regions
(Murthy 1981; Reddy 1985; Murakami 1986; Kant
2002; Swaminathan 2004). However, the abovementioned theories may not be sufficient if local effects are
important or accuracy in the calculation of transverse
stresses is required. For that purpose, more advanced
plate theories have been developed to include zigzag effects. This advanced method is an accuracy
refined method but with the disadvantage of expensive
computing time (Demasi 2009).
Analytical, closed form solutions are available in
very few special cases only, which are mainly restricted
to the boundary condition, simple geometry, loading
condition, etc (Carrera 2002). In fact, they can exactly
predict the displacement and stresses and can be used
to check the accuracy of other numerical methods such
as finite element models (Demasi 2007, 2010). Pagano
(1970) provided an exact solution for the problem of
rectangular orthotropic sandwich plate subjected to a
laterally distributed load. This solution has been used
as a benchmark to compare with various numerical,
finite differences, finite element, and other approximation solutions (Vinson 2001). Snirvas (1973) also
323
provided exact solutions, which are extensively used

as reference for various numerical calculations, such
as finite element (Ferreira et al. 2005).
Finite element methods applied to bending analysis by using FSDT and HSDT shear deformation
theories have been explored since 1960s. The FSDT
requires only displacement (C0 ) continuity across the
interelement boundaries, which is suitable for the
formulation of general finite elements. However, it
requires arbitrary shear correction factors and presents
shear locking problems for thin plates. Many HSDT
theories, among the ones developed by Reddy et al.
(1985), Touratier (1991), Soldatos (1992), Karama
(2009), etc. account for approximately parabolic transverse shear deformation and satisfy transverse shear
traction free conditions on the top and the bottom
surfaces of the plate, and do not require any shear
correction factor. However, these plate theories have
all positive features except one drawback, which is
found in a situation when finite element technique is
applied to this plate model, since these theories require
continuity of the transverse displacement as well
as its derivatives (C1 continuity) along the interelement boundary, which poses difficulties in developing
simple conforming quadrilateral element.
An advantage of using layerwise finite element
method adopting a new or existing shear deformation
theories (FSDT or HSDT) is that, they do not need
shear correction factors. Moreover, if shear continuity is superimposed in the layerwise description of the
theory, the number of degree of freedom (DOF) can be
fixed, i.e. the layerwise FEM code can be independent
of the number of layers.
The fewer DOFs the formulation has the faster
its computation is. In fact, few papers having fixed
DOF exist; see for example Roque et al. (2005).
Consequently, the aim of this paper is develop a new
polynomial layerwise finite element formulation for
the flexure analysis of thick multilayered panels having fixed DOF. Reissner-Mindlin approach is used to
model the faces of the sandwich structure. The displacement field for the core is modeled as quadratic
function of the thickness coordinate. Based on the
authors knowledge, the model proposed in this work
is not reported in the literature. Moreover, this model

can be also used for sandwich panels with different
outer and inner face thicknesses.
Based on new layerwise theory presented here
for the flexure of thick multilayered panels, a C0
Lagrangian isoparametric faceted quadrilateral element to analyze the static behavior of the general
laminated composite and sandwich plates is introduced. The discrete element chosen is a four-nodded
quadrilateral with: (a) nine-degrees-of-freedom per
node (when shear continuity is imposed), therefore,
the degrees of freedom (DOF) is layer independent;
(b) eleven-degrees-of-freedom per node (for 3 layers),
DOF is layer depended; and (c) seven-degrees-offreedom per node (for 3 layers), when fully linear
layerwise shear deformation theory is used, of course,
it is also layer depended. The accuracy of the present
code is ascertained by comparing it with Srinivas
(1973) exact solution and with various numerical calculations available in the literature such as the finite
element (Pandya & Kant 1988; Ferreira & Barbosa
2000) and meshless methods (Ferreira et al. 2005;
Roque et al. 2005).
2
STATEMENT OF THE PROBLEM
The sandwich laminated panel, composed of a finite

number of orthotropic layers of different thickness
is shown in Figure 1. The displacement field of the
present theory is determined by enforcing the continuity conditions of in-plane displacements at interfaces
and transverse shear stresses. Free surface conditions
of transverse shear stresses are neglected.
The new layerwise model proposed in this paper
accounts for non-linear and constant variation of
in-plane and transverse displacement respectively
through the panel thickness.
The displacement field for the inner layers (e.g. h1 ,
h2 , etc) is given as follows:
Figure 1. Mathematical terms in the modeling of the layerwise displacement field.
324
FINITE ELEMENT FORMULATION
For the present formulation, a four-nodded quadrilateral C0 continuous isoparametric element with
nine-degrees-of-freedom per node is employed. The
generalized displacements included in the present
theory can be expressed as follows:
where hin and hc are the thicknesses of an inner layer

and the core in Equations (1a-b), respectively
The displacement field for the middle layer, considered as the core in sandwich laminates, is given as:
x
where = {uo vo wo w/x w/y cb
cb ctx ct }T ,
i is the displacement vector corresponding to node i,
Ni is the shape function associated with the node i and
n is the number of nodes per element, which is
four in the present study. Considering the Equation
5, the strain vector {} can be expressed in terms of
containing nodal degrees of freedom as:
Finally, the displacement field for the outer layers

(e.g. h4 , h5 , etc), is given below:
hex and hc are the thicknesses of the outer layer and the
core in Equations (3ab), respectively.
In Equations (1ac), (2ac) and (3ac), uk and vk
are the inplane displacements at any point (xk , yk , zk )
in layer k {1,2, . . ., in, . . ., c, . . . ex, . . ., l-1, l}, uo
and vo denote the inplane displacement of the point (x,
y, 0) on the midplane of each layer, wo is the transverse
y
y
x
x
deflection, in ex and in
ex
are the rotations of the
normals to the midplane about the y and x axes, respecy
y
x
tively; while ctx , ct and cb
, cb are the top and bottom
core rotations about the x and y axes. In order to reduce
the degrees of freedom, the transverse shear stress continuity can be imposed in the interface between layers.
y
y
x
x
Therefore in
, in and ex
, ex are obtained as follows:
where [B] is the strain-displacement matrix in the

Cartesian coordinate system.
The potential energy e of the plate of mid-surface
area and volume V, loaded with an equivalent
load vector P corresponding to the nine-degreesof-freedom of a point on the mid-plane, can be
represented as
and the elements of [H ] are functions of z, which

can be taken from the displacement field model
abovementioned.
Considering the Equations (68), the elemental
potential energy can be rewritten as follows:
where
325
Figure 2. Laminate geometry with positive set of lamina/laminate reference axes, displacement components and fiber
orientation.
The equilibrium equation can be obtained by minimizing e with respect to {} as,
In-plane displacements, transverse displacement and

stresses presented in tables and figures are normalized
as given below:
where [Ke ] is the element stiffness matrix and {Pe } is

the nodal load vector.
Integrations of the equations (10ab) are carried out
numerically following the Gauss quadrature integration rule. A reduced integration technique is adopted
for the calculations of the shear stiffness matrix
in order to avoid numerical disturbances such as
shear locking, which may appear in a full integration
scheme. The stiffness matrix of all the elements was
calculated and assembled together to form the overall
stiffness matrix [K] of the panel and then the static
problem was solved.
4
NUMERICAL RESULTS AND DISCUSSION
A simply supported square sandwich plate under uniform load is considered, as used by Srinivas (1973).
The ratio of side to thickness, a/h is taken as 10. The
sandwich laminate considers two outside layers (skins)
of thickness h1 = h3 = 0.1 h and one inner layer (core)
of thickness h2 = 0.8 h. The skin orthotropic properties are obtained by multiplying an integer, R, by the
core orthotropic properties, given by
The skin properties are obtained by
Results of vertical deflection, normal stress and shear

stresses are presented in Tables 14. When shear
continuity condition is imposed, the results of shear
stresses are not accurate (see Table 1). This may be
due to the fact that in the present formulation the
free surface boundary conditions are not guaranteed,
and in such case the imposition of the shear continuity is unfavorable. Closer results are obtained, when
eleven-degrees-of-freedom are considered. Also the
results are in good agreement with the counterparts
by using only seven-degrees-of-freedom. In this paper,
the results given in Tables (24), are obtained by using
eleven-degrees-of-freedom, for three values of R, and
for different number of elements, N (mesh).
The present formulation produces results that are
in good agreement with all higher-order formulations
and in good agreement with the exact results, for the
values of R considered as in given in Tables (24) and
in Figure 3.
326
Table 1.
Maximum deflection and stresses of a square sandwich plate under uniform load (R = 15).
Method
w
(a/2, a/2, 0)
1xx
(a/2, a/2, h/2)
1xz
(0, a/2, 0)
2xz
(0, a/2, 2h/5)
Srinivas, exact solution

Present, 11 DOF
Present, 9DOF, shear continuity imposed
Present, 7DOF (linear disp. field)
121.72
121.285
121.168
121.285
66.787
66.554
66.517
66.554
3.964
3.747
3.782
3.747
3.577
2.034
0.447
2.034
Table 2.

Present, layerwise, DOF = 11, N = 21
Pandya & Kant HSDT
Pandya & Kant FSDT
Ferreira & Barbosa
Roque et al., layerwise, N = 21
Ferreira et al., N = 21
Table 3.
1xx
2xx
3xx
(a/2,
a/2, 0)
(a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (0, a/2, (0, a/2,
h/2)
2h/5)
2h/5)
h/2)
2h/5)
2h/5)
0)
2h/5)
258.970
257.560
256.336
254.191
258.740
236.100
258.740
259.120
258.720
257.120
257.523
255.920
252.084
60.353
60.154
60.066
59.946
62.380
61.870
59.210
60.338
60.262
59.985
59.968
59.650
58.863
46.623
46.429
46.346
46.174
46.910
49.500
45.610
46.570
46.511
46.305
46.291
46.037
45.423
9.340
9.286
9.269
9.235
9.382
9.899
9.122
9.314
9.302
9.261
9.258
9.207
9.885
1yy
38.491
38.448
38.416
38.372
38.930
36.650
37.880
38.547
38.501
38.330
38.321
38.141
37.690
2yy
30.097
30.073
30.031
29.949
30.330
29.320
29.590
30.148
30.112
29.981
29.974
29.830
29.477
3yy
6.161
6.015
6.006
5.990
6.065
5.864
5.918
6.030
6.022
5.996
5.995
5.966
5.895
1xz
2xz
4.364
3.898
3.807
3.695
3.089
3.313
3.593
4.539
4.461
4.289
4.046
3.977
3.831
3.268
1.742
1.682
1.618
2.600
2.444
3.593
3.384
3.331
3.217
2.390
2.538
2.532
1xz
2xz

Pandya & Kant HSDT
Pandya & Kant FSDT
Ferreira & Barbosa
1xx
2xx
3xx
(a/2,
a/2, 0)
(a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (0, a/2, (0, a/2,
h/2)
2h/5)
2h/5)
h/2)
2h/5)
2h/5)
0)
2h/5)
159.380
158.701
158.025
156.838
152.330
131.095
159.402
159.500
159.290
158.320
155.030
154.770
153.830
65.332
65.114
65.011
64.889
64.650
67.800
64.160
65.279
65.215
64.926
65.370
65.290
64.999
48.857
48.639
48.539
48.315
51.310
54.240
47.720
48.279
48.717
48.514
49.823
49.763
49.550
Table 2 shows that the present results are closer in

1
normal and shear stresses (xz
) than Pandya & Kant
(1988) and Ferreira & Barbosa (2000), but the oppo2
site occurs for xz
. From Table 3 it can be noticed
that the present model gives accurate results for
vertical deflection, normal and shear stresses, except
4.903
4.864
4.854
4.832
5.131
4.424
4.772
4.877
4.872
4.851
4.982
4.976
4.955
1yy
43.566
43.604
43.576
43.558
42.830
40.100
4.2970
43.682
43.641
43.451
43.273
43.220
43.030
2yy
33.413
33.466
33.420
33.317
33.970
32.080
42.900
33.523
33.493
33.353
33.601
33.564
33.421
3yy
3.500
3.347
3.342
3.332
3.397
3.208
3.290
3.352
3.349
3.335
3.361
3.356
3.342
4.096
3.809
3.722
3.613
3.147
3.152
3.518
4.290
4.217
4.056
4.284
4.213
4.049
3.515
1.942
1.866
1.788
2.600
2.676
3.518
3.672
3.617
3.491
3.675
3.632
3.509
2
for xz
stress. Similar comments as can be inferred for
the results in Table 4. But in this case the differences
are more pronounced, therefore, it can be inferred that
when R increases from 5 to 15, the present model gives
more accurate results when compared with Pandya &
Kant (1988) and Ferreira & Barbosa (2000) and they
327
Table 4.

Pandya & Kant HSDT
Pandya & Kant FSDT
Ferreira & Barbosa
1xx
2xx
3xx
(a/2,
a/2, 0)
(a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (a/2, a/2, (0, a/2, (0, a/2,
h/2)
2h/5)
2h/5)
h/2)
2h/5)
2h/5)
0)
2h/5)
121.720
121.285
120.814
119.987
110.43
90.85
121.82
121.880
121.680
121.880
115.460
115.270
114.570
66.787
66.554
66.441
66.314
66.62
70.04
65.65
66.730
66.659
66.730
66.870
66.799
66.510
48.299
48.065
47.959
47.706
51.97
56.03
47.09
48.204
48.148
48.204
50.041
49.987
49.783
3.238
3.204
3.197
3.180
3.47
3.75
3.14
3.214
3.210
3.214
3.336
3.333
3.319
1yy
46.424
46.504
46.480
46.483
44.92
41.39
45.85
46.586
46.535
46.586
45.724
45.669
45.471
2yy
34.955
35.047
35.002
34.886
35.41
33.11
34.42
35.109
35.069
35.109
35.150
35.107
34.961
3yy
2.494
2.336
2.334
2.326
2.36
2.21
2.29
2.341
2.338
2.341
2.343
2.341
2.331
1xz
3.964
3.747
3.663
3.556
3.04
3.09
3.47
4.167
4.105
4.167
4.177
4.111
3.953
2xz
3.577
2.034
1.948
1.862
2.70
2.76
3.47
3.742
3.705
3.742
3.771
3.728
3.602
Figure 3. Displacement and normalized stresses for a sandwich square plate (N = 21, a/h = 10, R = 15).
are in good agreement with Ferreira et al. (2005) and

Roque et al. (2005). This may be due to the fact that
the present formulation has a layerwise description of
the layers and some of the other HSDTs do not present
such advantage in terms of accuracy. Figure 3 shows
that the displacement and normalized stresses for a
sandwich square plate, and the differences in mechanical properties between the core and the faces at the
boundary interfaces between them are visible.
The mechanical behavior of the laminated composite structures can be improved further by considering
the continuity of the transverse shear stresses between
the layer interfaces. Additionally, an assessment of
the present theory can be done by including the normal stress component z . The present model can be
extended to cover different multilayered structures
such as shells. Dynamic and buckling calculations can
be also performed.
CONCLUSIONS
A layerwise finite element formulation of a newly

developed higher-order shear deformation theory for
the flexure of thick multilayered panels is presented.
328
The discrete element chosen is a four-nodded quadrilateral with nine-degrees-of-freedom per node (when
shear continuity is imposed). The accuracy of the
present theory is ascertained by comparing it with
various available results in the literature. The numerical results show that the present model performs
well as other existing higher order layerwise deformation theories for analyzing the global and interlaminar
mechanical behavior of multilayered sandwich and
composite plates.
ACKNOWLEDGEMENTS
The first and second authors have been financed by
the Portuguese Foundation of Science and Technology (Fundao para a Cincia e a Tecnologia) under
contract numbers SFRH/BD/66847/2009 and SFRH/
BPD/47687/2008 respectively.
REFERENCES
Ashton, J.E. 1970. Anisotropic plate analysis-boundary conditions. J Compos Mater 4:18291.
Carrera, E. 2002. Theories and Finite Elements for Multilayered, Anisotropic, Composite Plates and Shells. Arch.
Comput. Meth. Engng. 9(2):87140.
Chou, P.C., Carleone, J. 1973. Transverse shear in laminated
plate theories. AIAA J 11:133336.
Demasi, L. 2007. Three-dimensional closed form solutions
and exact thin plate theories for isotropic plates. Composite Structures 80: 183195.
Demasi, L. 2009. 6 Mixed plate theories based on the Generalized Unified Formulation. Part II: Layerwise theories.
Composite structures 87: 1222.
Demasi L. 2010. Three-dimensional closed form solution
and 3 theories for orthotropic plates. Mechanics of
Advanced Materials and Structures 17: 2039.
Dong, S.B., Tso, F.K.W. 1972. On a laminated orthotropic
shell theory including transverse shear deformation. J
Appl Mech 39:109196.
Ferreira, A.J.M., Barbosa, J.T. 2000. Buckling behavior of
composite shells. Compos Struct 50:938.
Ferreira, A.J.M., Roque, C.M.C., Jorge, R.M.N. 2005. Analysis of composite plates by trigonometric shear deformation theory and multiquadrics. Composite Structures
83:222537.
Jones, R.M. 1999. Mechanics of composite materials. Taylor
& Francis, Inc., Second Edition.
Kant, T., Swaminathan K. 2002. Analytical solutions for the

static analysis of laminated composite and sandwich plates
based on a higher order refined theory. Compos Struct
56:32944.
Karama, M., Afaq, K.S., Mistou, S. 2009. A new theory
for laminated composite plates. In: Proc. IMechE 2009
Vol. 223 Part L: J. Materials: Design and Applications.
Kontastinos G. 2001. Hull Construction with Composite
Materials for Ships over 100 m in Length. Master thesis,
Manchester institute and technology, MIT.
Mouritz A. P., Gellert E., Burchill P., Challis K. 2001. Review
of advanced composite structures for naval ships and
submarines. Composite Structures 53: 2141
Murakami, H. 1986. Laminated composite plate theory with
improved in-plane responses. J Appl Mech 53:66166.
Murthy, M.V. 1981. An improved transverse shear deformation theory for laminated anisotropic plates. NASA
Technical Paper 1903.
Pagano, N.J. 1970. Exact solutions for rectangular bidirectional composites and sandwich plates. Journal of
composite materials 4:2034.
Pandya, B.N., Kant, T. 1988. Higher-order shear deformable
theories for flexure of sandwich plates-finite element
evaluations. Int J Solids Struct 24:41951.
Reddy, J.N., Liu, C.F. 1985. A higher-order shear deformation
theory of laminated elastic shells. Int J Eng Sci 23;31930.
Reissner, E., Stavsky, Y. 1961. Bending and stretching of
certain types of heterogeneous aelotropic elastic plates.
J of Appl Mech 28(3):40208.
Reissner, E. 1975. On transverse bending of plates, including
the effect of transverse shear deformation. Int J Solids and
Struct 11:56973.
Roque, C.M.C., Ferreira, A.J.M., Jorge, R.M.N. 2005. Modeling of composite and sandwich plates by a trigonometric
layerwise deformation theory and radial basis functions.
Composites: Part B 36:559572.
Soldatos, K.P. 1992. A transverse shear deformation theory for homogeneous monoclinic plates. Acta Mechanica
94:195220.
Srinivas, S. 1973. A refined analysis of composite laminates.
J Sound Vibr 30:495507.
Swaminathan, K., Ragounadin, D. 2004. Analytical solutions
using a higher-order refined theory for the static analysis of antisymmetric angle-ply composite and sandwich
plates. Composite Structures 64:40517.
Swanson, S.R. 1997. Introduction to design and analysis with
advanced composite materials. Prentice Hall Inc.
Touratier, M. 1991. An efficient standard plate theory. Int J
Eng Sci 29(8):901916.
Whitney, J.M., Leissa, A.W. 1970. Analysis of simply supported laminated anisotropic plates. AIAA J 8: 2833.
329

FE parameters estimation and analysis of ultimate strength

of box girder
S. Saad-Eldeen, Y. Garbatov & C. Guedes Soares
ABSTRACT: This paper presents a finite element model and an analysis of large-scale specimen of box girders
subjected to four-point loading. Extensive nonlinear finite element analyses have been performed to find out
the best configuration of the model by varying the initial imperfection (shape and amplitude) and mesh size that
affect the ultimate strength and post-collapse behaviour. The results achieved by the use of the adopted finite
element models demonstrate a very good agreement with the three-linear trend of the bending moment-curvature
behaviour of an intact box girder. Furthermore, a comparison with different box girders having different material
properties and geometrical descriptions has been carried out.
INTRODUCTION
The longitudinal strength is the most important criterion for structural design of ship hulls, which is generally represented by the maximum bending moment
that the hull cross-section can withstand. The linear
elastic theory has been employed to predict the longitudinal strength of the ship hull for years. According
to this theory, the maximum bending moment that the
hull cross-section can withstand is equal to the bending moment corresponding to the first yield, that is, the
bending moment when the maximum stress on the hull
cross-section reaches the yield stress of the material.
In design practice, an allowable stress is used instead
of the yield stress, which corresponds to a safety factor
against yielding.
However, researches in the last decades have
revealed that the linear elastic theory is not adequate
in estimating the longitudinal strength of the ship
hull. It is necessary to take into account the following
factors: (1) various possible failure modes including
buckling, (2) progressive and interactive behaviour of
the failure of structural members, (3) redistribution
of the loads on the hull cross-section and (4) residual strength of structural members after buckling and
even after collapse. By considering these factors, the
maximum bending moment that the hull cross-section
can withstand is designated by the ultimate longitudinal strength, which represents the maximum loadcarrying capacity of the ship hull under longitudinal
bending.
Obviously, calculation of the ultimate longitudinal
strength is a non-linear problem in which both the
non-linearities related to material and geometry are
involved.
The test of simply supported box girders subjected to a pure bending moment may reproduce in
a simple manner the behaviour of the ship structure
under bending (Gordo and Guedes Soares, 2008, 2009,
Saad-Eldeen et al., 2010).
A series of finite element analysis (FEA) were conducted to simulate the behaviour of the tested box
girders. Hansen (1996) performed four finite element
models with different considerations for a crosssection selected from the model tested by Nishihara
(1984). The finite element model used in the analysis
was built up by 4-noded shell element with 5 layers
in the thickness. The model was loaded incrementally
with force vector at the end of the load section in order
to ensure the overall bending.
Qi et al. (2005) performed a series of non-linear
FEA for the ultimate strength of tested box girders
simulating a large surface ship, frigate and double hull
tanker. The calculation results showed a good agreement with the test results. Moreover, a comparison has
been done to calculate the ultimate hull girder strength
for a large double hull tanker using different methods.
Nikolov (2009) performed a collapse behaviour
analyses comparison between the experimental results
and different simplified methods of five different box
girders used for ultimate strength test. The stress-strain
and moment-curvature relationships were obtained,
and the comparison showed that there is significant difference between the numerical and some experimental
results.
The study presented here is part of long-term
project dealing with different levels of deteriorated box
girders representing the midship ship hull structures.
Through this project, three tests have been conducted
for box girders subjected to initial, moderate and
331
severe corrosion, as presented by Saad-Eldeen et al.

(2010, 2011c, a), for assessing the ultimate strength.
The aim of the analyses performed here is to calibrate the finite element model for a box girder without
corrosion through adjustment of different variables
that affect the post-collapse behaviour as well as
the ultimate bending moment-curvature relationship.
Defining the most appropriate FE model description,
the model will be used for analysing, later on, the deteriorated box girders used for the ultimate strength tests
performed by Saad-Eldeen et al. (2011b).
BOX GIRDER CONFIGURATIONS
The box girder consists of three bays; the deck panel is

stiffened with five longitudinal flat bars with a spacing
of 150 mm. The side panel is stiffened with two stiffeners on a distance of 300 and 500 mm respectively
and the bottom panel was stiffened with one stiffener
in the middle, as may be seen in Figure 1. The geometry configurations of the analysed box girder are given
in Table 1.
Because the experimental work, done by the
authors, only covers the deteriorated box girders, an
artificial intact box girder has been generated as a trend
based on the results achieved from corroded box girder
experimental data having the same structural configurations. A three-linear moment-curvature relationship,
as has been developed by Saad-Eldeen et al. (2011b),
simulates the behaviour of an intact box girder is used
as a base for the comparison with the FE calculations.
However, the three-linear moment-curvature relationships developed for different levels of corrosion
deterioration are based on the experimental results of
three corroded box girder testes as shown in Figure 2
and are used as base for the developed trend of smooth
moment-curvature relationships.
The mean value of the measured corrosion depth of
the deck plates of the box girder, as reported by SaadEldeen et al. (2011b) is 0.41 mm is compared with the
upper limit of the 95 percent confidence interval of the
corrosion depth defined by the regression equations
developed by Garbatov et al. (2007), for deck plate
of ballast tanks of real tanker ships, reveals that the
initially corroded box girder matches the 0.2 year of
deterioration without accounting for coating life (see
Figure 2).
The uniformly distributed corrosion depth of
0.41 mm leads to a small difference in the trend
of bending moment-curvature behaviour between the
intact box-girder and the initially corroded one, especially in the post-collapse part. In a case of the
application of coating protection, the only change will
be in the time dependency values leading to the delay
of the corrosion initiation until the failure of coating,
but the corrosion wastage dependency will be always
correct.
However, if a different corrosion environment is
present, the time-dependent model will be different
(Guedes Soares et al., 2008, 2009, Guedes Soares
Figure 1. Box girder geometry.

Table 1.
Principal characteristics of intact box girder
Item
Dimensions
units
Deck plating
Port Side plating
Starboard Side plating
Bottom plating
Stiffeners
Web frames
Brackets
4.5
4.5
4.5
4.5
25 4.5
50 50 6.5
80 100 4
mm
mm
mm
mm
mm
mm
mm
et al., 2011) but the corrosion wastage dependency

will remain the same.
FINITE ELEMENT MODEL
Numerical analyses of the ultimate strength for an

intact box girder are performed based on general
non-linear finite element commercial code ANSYS.
The FEA utilizes the full NewtonRaphson equilibrium iteration scheme, the large deformation option
was activated to solve the geometric and material
nonlinearities and pass through the extreme points.
332
Figure 2. Developed
relationship.
three-linear
Figure 4. Moment-curvature relationship as a function of

element size.
moment-curvature
heavy plates, which are located outside of the analysed box-girder and on the supporting arms connection
to avoid the presence of the shear effect (Figure 3).
The displacement load is applied by small increments
to ensure that the analysis would closely follow the
structures load-response curve.
The boundary conditions are modeled in the way
that the both ends are simply supported; both the ends
are constrained from translation in the vertical and
transverse direction and the translation in the longitudinal direction is constrained at the one of the ends,
also no rotation is prevented (see Figure 3).
4
Figure 3. Finite element model, including boundary

conditions.
The automatic time stepping features are exployed

allowing ANSYS to determine appropriate load steps.
The geometry of the box-girder structure is modeled in the same way as the real one used during
the ultimate strength test (see Figure 3). Shell element was used to generate the entire FE model. The
shell elements, SHELL 93 are defined by eight nodes,
four thicknesses, with six degrees of freedom at each
node: translations in the nodal x, y, and z directions
and rotations about the nodal x, y, and z axes. The
deformation shapes are quadratic in both in-plane
directions. The element has plasticity, stress stiffening,
large deflection, and large strain capabilities.
The material used is mild steel with a elasticperfectly plastic uniaxial stress-strain curve, where the
yield stress has a value of 235 [MPa] and the Young
modulus is 206 [GPa].
The box girder is subjected to a vertical loading
producing a constant pure vertical bending moment
along the box-girder length. The load is generated by
an imposed vertical displacement acting on the two
FINITE ELEMENT ANALYSES
The aim of these extensive non-linear analyses is to

calibrate the finite element model through varying
several governing parameters that affect the ultimate
strength and post-collapse behavior of the box-girder
structure and to find the best finite element model for
the intact box-girder.
4.1 Coarse mesh size
In the FE modelling, a finer mesh may generally result
in an accurate solution. However, with refining the
mesh the computation time increases. Therefore, it is
necessary to define the appropriate mesh size based
on the results accuracy and computational time effort
point of view.
The first variable to be analysed is the mesh size.
Four different element sizes, ES, have been adopted
(2.5, 5, 10 and 15 cm) to carry out series of finite
element analyses.
The analyses have been performed accounting for
several initial considerations. An initial imperfection
shape is generated, where the imperfection amplitude
is of 3 [mm].
The resulting moment-curvature relationships are
presented in Figure 4, and the ultimate bending
moments for different element size are summarized
in Table 2.
333
It can be observed that the slope of the momentcurvature relationship, representing the flexural rigidity, EI, coincide with the three-linear trend, and this
provides an evidence that the experimental results
may be well represented by the developed three-linear
trend relationship.
The gradient of the ultimate bending moment with
respect to the element size is calculated as:
bending moments for different imperfection amplitudes are summarized in Table 3. As may be seen
from Figure 6, when the initial imperfection amplitude is of 1 mm, the collapse behavior is too hard to
compare with the base solution. However, the small
initial imperfection leads to a bigger ultimate bending
moment capability as presented in Table 3.
From the gradient point of view, it is evident that,
as the larger the initial imperfection amplitude the
less the gradient of the ultimate bending moment
(see Figure 7). It has to be pointed out that there
where BMi is the ultimate bending moment at specified element size, ESi . By plotting the gradient vs.
the element size (see Figure 5) it can be observed that,
there is an inflection point at the level of the element
size of 5 [cm]. This point refers to the change of the
curve behavior, therefore, for the present finite element analysis the most appropriate element size seems
to be 5 [cm].
4.2
Initial imperfection amplitude
In the present study, plate initial imperfections are

modeled based on a Fourier series (Ueda and Yao,
1985). These imperfections have been generated in
ANSYS by changing the vertical position of element
nodes without inducing any additional stresses. The
non-linear analyses have been carried out with different initial imperfections amplitudes (1, 2, 3 and 4,
5 mm) with a pre-selected element size of 5 [cm].
The resulting moment-curvature relationships are
plotted as may be seen in Figure 6, and the ultimate
Table 3. Effect of imperfection amplitudes on the ultimate

bending moment.
Table 2.
Imperfection,
mm

different imperfections.
Effect of element size.
ES [cm]
2.5
10
15
FEM [kN m]
Exp [kN m]
Diff [kN m]
Gradient
582.53
601.55
3.16
1.60
586.53
601.55
2.49
2.23
597.66
601.55
0.65
1.36
604.48
601.55
0.49
FEM [kN m] 616.01 590.18 586.53 584.53 583.32

Exp [kN m] 601.55 601.55 601.55 601.55 601.55
Diff [kN m]
2.33 1.89 2.48 2.83 3.03
Gradient
25.83
3.65
1.99
1.22
Figure 5. Gradient of ultimate bending moment vs. element

size.
Figure 7. Gradient of ultimate bending moment vs.

imperfection.
334
is no any inflection point, which indicates that the

trend of the curve decreases with the increase of the
initial imperfection amplitude. There is also an indication from the gradient analysis that the difference
between each two consecutive initial imperfections
is decreasing, as tabulated in Table 3. Therefore, the
most appropriate initial imperfection amplitude is
defined as 4 [mm].
4.3
Initial imperfection shape
Two shapes of initial imperfections are considered in

the analyses: concave and convex. The two shapes are
based on the Fourier series, as given by Eqn 2, where
a is the length of plate (400 mm), b is the plate width
(150 mm), z is the longitudinal axis, x is the transversal
axis, m is the mode of the imperfections in z (i.e. the
number of half-sine waves within the length of one
plate), n is the mode of the plate in x (i.e. the number
of half-sine waves within the width of one plate) and
w0 is the amplitude of the imperfections. The values
of m and n used in this study correspond to the shapes
Figure 8. Initial imperfection shapes.
The upward and downward indications refer to the

shape of imperfection at the middle bay.
The global deformed shapes of the box girder with
initial imperfections shapes for the last sub-step is
shown in Figure 9. It is clear that the deformation
obey the normal behavior of the expected deformation due to the applied load as in sagging condition, but
the maximum displacement achieved by the two shape
models is different in its position, see Figure 9 and 10.
It has to be pointed out that at the ends (simply
supports), there is no vertical displacemnts, which satisfies the imposed boundary conditions, see Figure 9.
Figure 11 shows the stress distribution in longitudinal direction. There are some locations with plastic
deformation, mainly in the bottom part of the boxgirder and the neighbour part of the side plating near
the bottom.As may be seen the stress distribution in the
middle bay is different, for the downward imperfection
(left), the deck panel did not reach the yield therefore
there is no plasticity on the deck,on the contrary, for
the upward imperfection (left).
The moment-curvature relationship as a function
of imperfection shapes are plotted in Figure 12 and
the comparison between the calculated and experimentally registered maximum bending moments is given
in Table 4. As may be seen there is a small difference
in the post-collapse behaviour between the two models
and the upward shape gives a higher ultimate bending
moment rather than the downward one with a value of
4.72 [kN m] as given in Table 4. Because of that, the
upward shape is defined as the most appropriate initial
imperfection shape.
Finally the most appropriate characteristics of
the finite element model constructed for the present
analysis are 5 [cm] element sizes and 4 [mm] initial
imperfections with upward shape located at the middle
bay of the box-girder.
Figure 9. Vertical displacement, downward (left) and upward (right) imperfections.
335
Figure 10. Vertical displacement (middle bay), downward (left) and upward (right) imperfections.
Figure 11. Longitudinal stresses (middle bay), downward (left) and upward (right) imperfections.
Table 4. Effect of initial imperfection shape on the ultimate
bending moment.
Shape
Down
Up
Diff [kN m]
FEM [kN m]
Exp [kN m]
Diff [kN m]
584.53
601.55
2.83
589.25
601.55
2.04
4.72

different initial imperfection shape.
COMPARATIVE ANALYSIS
Many analyses of the ultimate strength of experimentally tested box-girders have been performed by many
researchers, as has been already described in Section 1, using many different numerical approaches.
Here the results of finite element analysis performed
for intact box-girder are compared with the ones published in literature.The box girders in consideration are
somehow similar, but have been built from different
steel class materials and geometrical configurations,
336
Table 5.
be seen form Table 5. On the contrary, the box girder

withstands an ultimate bending moment more than
the analysed one and this is as a result of its higher
material yield stress.
Principal characteristics of different box girders.
Item
MST-4 MST-3 N-200 M3-200
Present
box
Units
a
b
tp
ts
hs
y
540
540
200
180
180
150
4.35
3.05
4
4.35
3.05
4
50
50
20
264
287
270
400
150
4.5
4.5
25
235
mm
mm
mm
mm
mm
MPa
MPa
206
210
589
GPa
GPa
MNm2
MNm2
kN m
kN m
y% 112
122
114
E
E%
EI
EI%
Mu
Mu%
207
101
180
86
587
100
200
97
151
72
632
107
208
101
258
123
898
153
800
200
3
4
45
1833 mm
3104 mm
783 mm
1324 mm
210
102
130
62
316
54
CONCLUSIONS
A series of non-linear FEA for intact box girder

has been carried out using the commercial software
ANSYS. The principal objective was to define the
most appropriate parameter descriptions for the developed FE model by varying the initial imperfections
(shape and amplitude) and the mesh size that are
the most important governing parameters that affect
the ultimate strength and post-collapse behaviour. The
results achieved by the developed FE model demonstrated a very good agreement with the experimentally
defined three-linear trend of the moment-curvature
relationship of the intact box girder.
The progressive collapse behaviour of the four box
girders was compared with the analysed one showing
a good understanding of the effect of both flexural
rigidity and the material yield stress on the collapse
behaviour and the ultimate capacity of the box girder.
ACKNOWLEDGEMENTS
The work reported here is a contribution to the activities of the MARSTRUCT VIRTUAL INSTITUTE,
(www.marstruct-vi.com) in particular its Technical
Subcommittee 2.3 on Ultimate Strength.
The first author has been funded by the Portuguese
Foundation for Science and Technology (Fundao
para a Cincia e Tecnologia FCT) under contract
SFRH/BD/46790/2008.
Figure 13. Comparison with different box girders.
REFERENCES
MST-4, MST-3, N-200 and M3-200 and have been

analysed using FEA by Nikolov (2009). The principal
characteristics of analysed box-girders are presented
in Table 5.
Figure 13 summarises the resulting bendingmoment-curvature relationship. The curve outlined as
a solid line refers to the analyzed intact box girder, and
the other dashed lines belong to the other box girders.
It has to be pointed out that the the momentcurvature curve of the analysed intact box girder is
placed almost at the middle between the other boxes.
The inclination of the moment-curvature relationship,
representing the flexural rigidity, EI in the elastic zone
clearly supports that position.
The summary of the flexural rigidity of the boxes
presented in Figure 13 may be seen in Table 5. For
example, for the box girder N-200 is evident that the
inclination of the moment-curvature relationship is
less than the one of the analysed box girder as may
Garbatov, Y., Guedes Soares, C., Wang, G., 2007. Nonlinear time dependent corrosion wastage of deck plates of
ballast and cargo tanks of tankers. Journal of Offshore
Mechanics and Arctic Engineering 129, pp. 4855.
Gordo, J.M., Guedes Soares, C., 2008. Experimental Evaluation of the Behavior of a Mild Steel Box Girder
under Bending Moment. Ships and Offshore Structures
3, pp. 347358.
Gordo, J.M., Guedes Soares, C., 2009. Tests on Ultimate
Strength of Hull Box Girders Made of High-Tensile Steel.
Marine Structures 22, pp. 770790.
Guedes Soares, C., Garbatov, Y., Zayed, A., 2011. Effect of
Environmental Factors on Steel Plate Corrosion under
Marine Immersion Conditions. Corrosion Engineering,
Science and Technology 46, pp. 524541.
Guedes Soares, C., Garbatov, Y., Zayed, A., Wang, G., 2008.
Corrosion Wastage Model for Ship Crude Oil Tanks.
Corrosion Science 50, pp. 30953106.
Guedes Soares, C., Garbatov, Y., Zayed, A., Wang, G., 2009.
Influence of Environmental Factors on Corrosion of Ship
Structures in Marine Atmosphere. Corrosion Science 51,
pp. 20142026.
337
Hansen, A.M., 1996. Strength of Midship Sections. Marine

Nikolov, P.I., 2009. Box Girder Strength under Pure
Bending: Comparison of Experimental and Numerical
Results, IMAM, Istanbul, Turkey, pp. 7986.
Nishihara, S., 1984. Ultimate Longitudinal Strength of
Midship Cross Section. Naval Architecture & Ocean
Qi, E., Cui, W., Wan, Z., 2005. Comparative Study on the
Utimate Hull girder Strength of Large Double Hull
Tankres. Marine Structures 18, pp. 227249.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2010.
Experimental Assessment of the Ultimate Strength of a
Box Girder Subjected to four-point Bending Moment,
Proceedings of the 11th International Symposium on
Practical Design of Ships and other Floating Structures
(PRADS2010). ASME, New York.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2011a.

Compressive Strength Assessment of a Moderately
Corroded Box Girder. Marine System and Ocean Technology 5, pp. 515.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2011b.
Corrosion Dependent Ultimate Strength Assessment of
Aged Box Girders Based on Experimental Results.
Journal of Ship Research 55, pp. 289300.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2011c.
Experimental Assessment of the Ultimate Strength of
a Box Girder Subjected to Severe Corrosion. Marine
Ueda, Y., Yao, T., 1985. The influence of complex initial deflection modes on the behaviour and ultimate
strength of rectangular plates in compression. Journal of
Constructional Steel Research 5, pp. 265302.
338

Structural behaviour of a lightweight craft

M.S. Santos
Faculdade de Cincias e Tecnologia, Universidade Nova de Lisboa, UNIDEMI, Departamento de Engenharia
Mecnica e Industrial, Campus de Caparica, Caparica, Portugal
R.F. Martins
Faculdade de Cincias e Tecnologia, Universidade Nova de Lisboa, UNIDEMI, Departamento de Engenharia
Mecnica e Industrial, Campus de Caparica, Caparica, Portugal
ICEMS, Insituto Superior Tcnico, Technical University of Lisbon, Portugal
P.P. Silva
Misso de Acompanhamento e Fiscalizao Lanchas de Fiscalizao Costeira, Marinha Portuguesa,
Base Naval de Lisboa, Alfeite, Portugal
A.R. Mateus
Marinha Portuguesa, Direco de Navios, Base Naval de Lisboa, Alfeite, Portugal
ABSTRACT: The main purpose of this paper is to present both the global and local structural response of
a fast craft, when subjected to the effect of slamming and flooding of watertight adjacent compartments. The
studies were accomplished using the Finite Element Method (FEM) applied to the vessel model, giving special
attention to the bow and stern of the craft when submitted to slamming effects and jet thrust loads, respectively.
The modes and frequencies of vibration were determined to check for structural resonance problems due to
slamming induced loads, considering the sea conditions where the craft operates. The forced frequency response
of a specific region on the bow was also determined when subjected to slamming loads applied in a frequency
range from 25 to 80 Hz. The results revealed that the maximum stress induced in the structure was lower than
the yield strength of the materials used in the production of the craft, the aluminium alloys 5083-H111 e 6082T6. Hypothetical flooding of watertight compartments was also simulated in the ships section comprehended
between bulkheads no. 17 and 20 and in the engine and waterjets compartments, separately.
1
INTRODUCTION
The vessel under study (Fig. 1) is a motor lifeboat

which was designed to operate in a sea state 8
with significant wave height between 9 and 14
meters, wind force 10 and self-righting ability (www.
revistademarinha.com).
The hull was built using two aluminium alloys,
belonging to series 5xxx and 6xxx (Hydro Aluminium
Vekst 2002), namely 5083-H111 and 6082.
The alloy 5083-H111 has a medium mechanical

strength and good resistance to corrosion in marine
environment and is therefore used in the hull, which
is constantly in contact with seawater. Heat-treated
aluminium alloy 6082-T6 is considered a medium/
high resistance alloy, having been used in the reinforcements of the hull, in order to obtain lightweight
stiffened panels. Figures 2 and 3 show stress/strain
curves of the 5083-H111 and 6082-T6 alloys, obtained
from axial tensile tests (Matos et al. 2009).
Figure 1. Lifeboat under study.
LOADINGS
The craft, when in service, may be subjected to slamming pressures, due to the slam of the hull on the sea
surface upon re-entry into the water. This phenomenon
induces high forces on the hull and on its structural
elements (Hermundstad & Moan 2007), as well as the
possibility of excitation of the first vibration modes
of the structure.
The magnitude of the slamming load is dependent
on the hulls design, service conditions, operation area
339
Figure 4. Areas considered in the FEM for the application

of the slamming pressure.
Figure 2. True stress-strain curve for 5083-H111 aluminium
alloy obtained from axial tensile tests (Matos et al. 2009).
Table 1. Significant wave high and wave period for the

North Atlantic Ocean (Journe & Pinkster 2002).
Tz (s)
Hs (m)
3.5
1.5
4.5
4.5
5.5
9.5
6.5
13.5
7.5
14.5
8.5
14.5
angle of the waves direction relative to the ships

speed vector (Perez 2005).
Varying U between 1 and 15 m/s (0,51 e 29,2

knots) and between 0 (stern seas) and 180 (bow
seas), encounter frequencies, e , between 0 rad/s
and 3,12 rad/s (0 and 0,50 Hz) were calculated. The
maximum value obtained corresponds to a speed of
29 knots and an encounter angle of rad (180 ).
Taking into account previous information, it was
possible to model the dynamic loading on a single
harmonic pulsed load (2), according to equation (3).
Figure 3. True stress-strain curve for 6082-T6 aluminium

alloy obtained from axial tensile tests (Matos et al. 2009).
of the vessel and the pressure applied at the bottom of

the hull. For the craft under study, the maximum value
calculated for the slamming pressure, PS (Lloyds
Register of Shipping 1996), was 196.3 kN/m2 . This
was the load value applied on the outside shells of
the hull, near bulkhead no. 20, to perform the structural analysis (Figure 4).The areas where the slamming
pressure was applied are much smaller than in reality or than what is predicted by the Classification
Societies, but was intended to maximize the stresses
induced in the structure analysed.
Based on Journe & Pinkster (2002) the significant wave height (Hs) is related with the wave period
(Tz ) for the North Atlantic Ocean, where the lifeboat
operates (Table 1). According to Table 1, for the wave
height considered (Hs 10 m), the wave period, Tz ,
was considered equal to 5.5 s and the frequency, ,
equal to 1.142 rad/s.
Pressure due to the slamming effect depends on the
encounter frequency (e ) of the waves on the hull,
expressed by equation (1), where U defines the speed
of the craft, g is the acceleration of gravity (9.81 m/s2 ),
corresponds to the wave frequency and is the
The study of the watertight flooding compartments

was done based on equation (4), simulating an internal hydrostatic pressure applied on the bulkheads and
on the hull of the craft, with the maximum pressure
occurrence at h = hmax , and a minimum pressure
verified at the deck (h = 0).
where h is the height of the water column, is the

density of seawater at 25 and g is the acceleration of
gravity.
3
MODELLING
The secondary and terciary structural behaviour of

the craft can be modelled analitically by replacing
340
Figure 8. Symmetry restraint applied to the bow and stern

models (yellow lines).
Figure 5. Overall view of the lifeboat finite model, meshed
with plate and shell elements.
Figure 9. Definition of fixed edges in the bow and stern

models (red lines).
Figure 6. Bow model of the lifeboat under study.
Figure 10. Motor support zones (in red).
Figure 7. Stern model of the lifeboat under study.
the reinforced panels for equivalent thin plates, once

they are planar elements, in which the thickness has a
very small size compared with the other dimensions
(Kirchhoff Plate). Prior to the analysis of the vessel
and taking as reference analytical expressions (Szilard
1974) for some plates and shells geometries, loading
and boundary conditions), a validation of the results
obtained using SolidWorks Simulation and ANSYS ,
was carried out (Santos 2011).
For the prupose of Finite Element Analysis (FEM)
the global craft structure (Fig. 5) was modelled from
its technical drawings. However, due to the high
complexity of the structure, reflected in large files
size, for this study only results for the bow and stern
(Figs. 6 and 7) were analised.
To study the structures indicated in Figures 6 and 7,
it was necessary to apply boundary restraints and loads
that simulated the service conditions. Thus, acceleration due to gravity was applied, as well as symmetry
(yellow lines) and fixed edges (red lines), as shown in
Figures 8 and 9, respectively.
The weight of each motor (1250 kg), was taken into

account and distributed through four areas according
to the existing supports (Fig. 10). The value of force
applied in each support is given by equation (5).
The impulse of the water-jet propulsion was

determined through the jets performance curve
(Fig. 11) (www.scania.pt). According to the graph
(Fig. 11), for a craft speed of 25 knots and a power of
478 kW/motor, a thrust of 18 000 N is expected for each
jet. Considering each jet supported by eight points
(Fig. 12), a force of 2250 N per node was applied.
Stresses induced by slamming, self-weight, weight
of the equipments (motors and jets) and hydrostatic
pressure applied on the outside of the hull, which
simulates the effect of seawater pressure on the hull
(Fig. 13), were calculated assuming the superposition
of the loadings referred.
341
Figure 11. Jets performance curves.
Figure 14. Structural analysis of the bow when subjected

to slamming pressure and external hydrostatic pressure.
a) Displacements [m]; b) von Mises stresses [Pa].
Figure 12. Jets support points in the stern.
Figure 13. Surface affected by the hydrostatic pressure.
Figure 15. Structural analysis of the stern when subjected

to the jets propulsion and external hydrostatic pressure.
a) von Mises stresses [Pa]; b) Displacements [m].
CALCULATION OF STRESSES INDUCED

ON STRUCTURES UNDER STUDY
Non-linear FE analyses were performed, which

allowed for considering large deformations, hardening and plasticity, through the use of two mechanical behaviour material models, (Figs. 2 and 3). The
application of typical in-service loads slamming
pressure, seawater hydrostatic pressure, jet thrust,
self-weight and/or weight of the equipments resulted
in displacements and von Mises stresses, which can
be observed in Figures 14 and 15.
The analysis of figure 14a shows that the maximum displacement expected in the bow is 1 mm
and that was obtained in the region designated by
Point 1. The maximum von Mises stresses induced in
the bow by the superposition of the slamming pressure and the external hydrostatic pressure, Figure 14b,
do not exceeded 65 MPa (half of the yield strength of
the 5083-H111 alloy), although in a global perspective, stresses induced were much lower 15 MPa (on
average).
Figure 15a shows that the von Mises stresses

induced in the jets supported areas, due to the jets
thrust, do not exceed 12 MPa, while the stresses
induced in the remaining structure, combining the
jets thrust and the external hydrostatic pressure is
less than 30 MPa (23% of the yield strength of alloy
5083-H111). The maximum displacement calculated,
Figure 15b, was 2 mm, and obtained in bulkhead
no. 0 (zero), near the circular hole where the jets
are supported. However, once the structural contribution of jets to the local stiffness increase was not
taken into account, one can expect a smaller maximum displacement value in this area. In a global
view of the stern, the von Mises stresses were higher
on bulkhead no. 7, with 28 MPa (21.5% of the
materials yield strength) near stress concentration
zones (Table 2).
Table 2 shows the displacements and the von Mises
stresses calculated on bulkheads no. 0, 3 and 7.
342
Table 2. Displacements [m] and von Mises stresses [Pa]

on bulkheads no. 0, 3 and 7 due to jets thrust and external
seawater hydrostatic pressure.
Figure 16. Displacements [m] due to flooding of compartment comprehended between bulkheads no. 17 and 20.
Figure 17. Displacements [m] in the engines compartment.
Figure 18. Displacements [m] in the jets compartment.
Table 3. Displacements [m] and von Mises stresses [Pa]

induced on bulkheads no. 17 and 20 due to slamming and
external hydrostatic pressures.
HYPOTHETICAL FLOODING OF
WATERTIGHT COMPARTMENTS
The resulting displacement obtained from non-linear

FE analysis due to the hypothetical flooding of watertight compartments located in the bow, jets and engines
is visible in Figures 16, 17 and 18, respectively.
In the bow, predicted displacements (Fig. 16) are
higher in bulkheads no. 17 and 20, being of 5.6 mm
and 3.8 mm, respectively. Additionally, flooding of
the engine compartment induces a maximum displacement in the hull of 12.8 mm (Fig. 17), since the span
between bulkheads no. 3 and 7 is larger (1800 mm),
resulting in greater deflection. The flooding of the
jets compartment induces a maximum displacement
on bulkheads no. 3 of approximately 7 mm (Fig. 18).
From numerical analyses, displacements and the
von Mises stresses calculated in bulkheads no. 17 and
20, located in the bow (Fig. 14), are shown in Table 3.
In the bow, the von Mises stresses induced in the
structure are low (20 MPa, Figure 19). However, in the
connection of the hulls reinforcements to bulkhead
no. 17 and in the connection between decks reinforcements with longitudinal stiffeners of bulkhead no. 17,
the local stress values were about 180 MPa, verified
on the bulkheads plate (Fig. 19), indicating the existence of high local stresses, due to the existence of
geometric stress concentration factors. However, as
the modelling of the brackets responsible for connection between the structural elements referred was not
343
Figure 22. Modes of vibration of the bow.

Figure 19. Von Mises stresses [Pa] on the watertight compartment between bulkheads no. 17 and 20.
Figure 20. Von Mises stresses [Pa] in the jets compartment,

excluding the circular hole in bulkhead no. 0 (zero).
Figure 23. Modes of vibration of the stern.
Table 4.
Bow
Stern
Frequencies of vibration of the bow and stern [Hz].

1st mode
2nd mode
3rd mode
68.49
44.36
78.17
45.69
48.16
Excluding stress concentration hot spots detected

and described previously, the maximum stress calculated in the analysed sections occurred in the extreme
fibres of the bulkheads reinforcements, which was
about 60 MPa (20% of the yield strength of 6082-T6
alloy).
Figure 21. Von Mises stresses [Pa] in the engines

compartment.
made with the exact radius of curvature of the brackets

installed on the craft, one can assume that the stress
values calculated by the numerical methods may be
higher than those existing in the real structure.
Nevertheless, it appears that high local stresses,
close to the aluminium alloy yield strength, could
be induced in the bulkhead plate, which is envisaged
in the structures design, in the flooding case, where
the structures could deformed till the plasticization of
building material (Chalmers 1993).
In the jets compartment, flooding induces higher
stresses in the connection between the hulls reinforcements and the reinforcements of bulkhead
no. 3, reaching 182 MPa (60% of the yield strength
of 6082-T6 alloy) (Fig. 20), similarly to what happens
in the same bulkhead, in case of flooding of engine
compartment (Fig. 21).
MODES AND FREQUENCIES OF

VIBRATION OF THE BOW AND STERN
The finite element analysis of the modes and frequencies of vibration of the bow and stern resulted
in Figures 22 and 23.
The frequencies of vibration corresponding to the
modes illustrated in Figures 22 and 23 are presented
in Table 4.
According to what was previously seen, the
encounter frequency (less than 0.50 Hz) imposed by
slamming, and the natural frequencies of vibration of
the craft (higher than 44 Hz) are far away, thus eliminating the possibility of resonance due to the demands
of slamming as the craft operates.
344
area near bulkhead no. 17, when in reality the load

due to slamming is applied in a higher area that is
most aft this bulkhead. Thus, the real values may differ
from those calculated, but sought to maximize induced
stresses in the structure.
Analyzing the graphs included in Figure 24, one
observes that the displacement of point 1, due to the
slamming pressure, has peaks that coincide with the
natural frequencies of the bow. The points of instability
observed at frequencies of 26 Hz, 57 Hz and 74 Hz
are indicative of other vibration modes (not shown in
Table 4), which deform the non-reinforced plates of
the deck at the tip of the bow, which are less critical
and were not included in the analysis of modes and
frequencies of vibration of the bow (Table 4).
Through the observation of Figure 24 is possible
to calculate the total displacement near the first vibration frequency (68.49 Hz), T68 (combining x, y and z),
as shown in equation (6). Similarly, in the instability point located near the second vibration frequency
(78.17 Hz), the total displacement value T78 is given
by equation (7).
The maximum displacement obtained for the first

vibration frequency (68.49 Hz), 1.08 103 m, and
at the instability point near the second frequency of
vibration (78.17 Hz), 1.2 103 m, which reveals displacement values similar to that obtained for point 1
(0.99 103 m) in the static analysis of the bow when
subjected to the slamming loads. This study indicates
that the displacement values calculated in a frequency
response of the bow are very close to those obtained
statically. The von Mises stresses at point 1 are shown
in Figure 25.
By observation of Figure 25, one concludes that
the maximum stress values calculated were 33 MPa
(25.4% of the 5083-H111 alloys yield strength), which
is much below the yield strength of the material.
As occurred in the displacement analyses, the stress
distributions show spikes that coincide with the natural
frequencies of vibration of the bow.
Figure 24. Displacements in point 1 along x direction (a),

y direction (b) and z direction (c).
BOW FREQUENCY RESPONSE WHEN

SUBJECTED TO SLAMMING
A constant pressure of 196.3 kN/m2 was applied, in a

frequency range between 25 Hz and 80 Hz, in the area
shown in Figure 4, to obtain the frequency response
of the bow when subjected to slamming. Figure 24
shows the displacement versus frequency graph for
point 1 (Fig. 14), in the x, y and z directions. Values
were obtained assuming a narrow slamming impact
CONCLUSIONS
The structural response of the bow, resulting from

superimposed loading of slamming and hydrostatic
pressure, showed von Mises stress values below
65 MPa, which occurred on the plate of bulkhead 17,
in the vicinity of complex geometrical continuity, thus
stress concentration area.
345
Figure 25. Von Mises stresses [Pa], at point 1, in two

frequency ranges (a) 25 and 75 Hz; (b) 70 and 80 Hz.
This value is about 50% of the yield stress of the

aluminium alloy 5083-H111 from which the bulkhead
plate is made.
The maximum displacement, of about 1 mm, is
observed on the hull plate.
The dynamic structural response of the bow, on the
chosen point for analysis, showed a maximum displacement of 1.2 mm, and a von Mises stress of 33 MPa
(25.4% of the 5083-H111 alloy yield stress).
From the analysis of flooding simulation on the
bow compartment, it was verified that bulkhead 17
is the most demanded structure, with average von
Mises stresses of 20 MPa (15% of the 5083-H111 alloy
yield stress), but with high stress concentration areas
(180 MPa) and maximum deformation of 5.6 mm.
However, in these stress concentration areas, a submodeling FEA, with more accurate modeling of the
structural elements responsible for structural continuity and stress releasing, would give a more accurate
evaluation of the stress concentration factors and guide
the structural designer on possible measures to avoid
or minimize it.
Nevertheless, for flooding structural design it may
be assumed permanent deformation (material working
on the plastic zone) (Chalmers 1993).
The study of the free vibration of the bow, showed
that the 1st and 2nd modes are more relevant on the
deformation of bulkhead 17. That could be an issue
in the case of the craft structure being excited on
those modes, thus causing resonance.
At the stern, the 3rd mode is relevant on bulkhead
3, while the 1st and 2nd modes are relevant on the hull
behavior.
From the analysis of the frequencies of excitation
its possible to conclude that the craft operates on a
range far from the natural frequencies of the structural
models studied. It is expected that it will not suffer
from resonance issues related with slamming loads,
since the lowest natural frequency, both on the bow
and on the stern is 44,36 Hz and the encounter frequency expected in the crafts area of operation is at
most 0.50 Hz.
Considering the load due to the waterjet impulse,
it was observed deformations of about 2 mm and von
Mises stresses of 28 MPa (21.5% of the 5083-H111
alloy yield stress).
engine room, it was verified that maximum displacement occurs on the hull near the keel (about
13 mm), maybe due to the highest span of the craft
compartments occurring in this compartment.
waterjet room, it was verified that bulkhead 3 holds
the maximum displacement about 7 mm. The highest stress values occur in the connection elements
between the hull and bulkhead 3 reinforcements
(about 182 MPa, thus 60% of the 6082-T6 alloy yield
stress). This is similar to what is verified on the same
bulkhead when the flood occurs in the engine room.
Excluding the areas of stress concentration, the
highest stress values occur on the external fibres of
the bulkhead reinforcements and are of 60 MPa (20%
of the 6082-T6 alloy yield stress).
All in all, the stress values of the bow and stern sections of the high speed lifeboat show enough strength
for the expected service loads and incident flooding
occurrence, the craft is not expected to suffer from
resonance phenomena due to slamming loads.
Further work should be done:
346
on hot spot analysis and maybe some redesign

for safe-life or life-safe operation could be
performed,
on modeling analysis of the global lifeboat structure, mainly for study of the castaway compartment,
increasing the area of slamming loads impact on
crafts bottom,
on obtaining and applying slamming loads as a
spectrum.
REFERENCES
Chalmers, D.W.; 1993; Design of shipsstructures; HMSO,
London
Hermundstad, Ole A; Moan, Torgeir; 2007; Efficient calculations of slamming pressures on ships in irregular seas;
Journal of Marine Science and Technology; Springer;
Norway.
J.M.J. Journe and Jakob Pinkster; 2002; Introduction In
Ship Hydromechanics, Lecture MT519; Draft Edition;
Delft University of Technology, Nederland.
Hydro Aluminium Vekst; 2002; Aluminium in the Marine
Environment; 1st Edition, Norway.
Lloyds Register of Shipping; 1996; Rules and Regulations
for the Classification of Special Service Craft.
Matos, J., Silva, P., Martins, R. F., Mateus, A.R.; 2009;

Influncia dos defeitos de produo na resistncia estrutural de painis reforados de lanchas de alumnio;
Encontro Nacional de Engenharia Naval09 O Transporte Martimo e Fluvial de Passageiros; Ordem dos
Engenheiros Colgio de Engenharia Naval; Lisboa.
Perez, Tristan; 2005; Ship Motion Control: Cruise Keeping
and roll stabilization using rudder and fins; Springer.
Santos, Marco; 2011; Anlise do Comportamento Estrutural
de uma Lancha Salva-Vidas, Tese de Mestrado, Faculdade
de Cincias e Tecnologia, Universidade Nova de Lisboa,
Caparica.
Szilard, Rudolph; 1974; Theory and Analysis of Plates
Classical and Numerical Methods; Prentice-Hall INC;
New Jersey. www.revistademarinha.com, (22-10-2010).
347

Comparison of numerical and experimental results of

the modal analysis of a ship deck panel
Goranka timac
Faculty of Engineering, University of Rijeka, Croatia
Hanbing Luo & C. Guedes Soares

ABSTRACT: In order to avoid resonances and consequently excessive vibrations, special attention is dedicated
to modal analysis, with the aim of determination of natural frequencies and mode shapes of the structure. The
objective of this paper is to perform numerical modal analysis of a ferry deck panel as well as to compare the
obtained results with previously available experimental data. Finite element analysis of the stiffened plate is
performed using finite element analysis with shell elements for the plate and stiffeners. Final comparisons show
good agreement of numerical and experimental data.
INTRODUCTION
Plates are in extensive use as one of the important

structural elements in marine, civil, aeronautical and
mechanical engineering. In the construction of marine
vessels, stiffened plates are the most commonly used
structural elements, forming the deck, bottom hull,
side shells and bulkheads, which can be subjected to
static and/or dynamic loads. Since the resulting vibrations often have an undesirable influence on passenger
comfort or, even more dangerously, can cause the
damage of the structure, its reduction is an important
part of the design process in shipbuilding industry.
In order to avoid resonances and consequently
excessive vibrations, special attention is dedicated to
modal analysis, with the aim of determination of natural frequencies and mode shapes of the structure (Den
Hartog, 1947). Nowadays, to perform this analysis it
is common to use the finite element method (FEM),
which has significant advantages in comparison to
analytical solutions, or perform experimental testing
on a physical structure.
There are many published works which directly or
indirectly include modal analysis of the stiffened plate.
Troitsky (1976) has extensively reviewed the literature related to analytical procedures used for analysis
of rectangular stiffened plates. In Ferrari and Rizzuto
(2003a) experimental modal analysis of the stiffened
panel of a ferry deck is performed, while in Ferrari
and Rizzuto (2003b) comparison of available experimental results with numerical data from finite element
software Nastran is carried out. A good correspondence is found regarding to the modes, while less
satisfactory is the agreement of absolute values of

predicted frequencies.
The influence of residual stresses induced from spot
welding of stiffeners was studied in Nacy et al. (2009),
using numerical and experimental modal analysis.
It was observed that it has small influence on natural
frequencies and can be neglected in the further
numerical simulations.
The modal analysis of stiffened plate has been
applied as an initial study in many works concerning
vibration behavior, such as transient dynamic analysis of vehicle-deck interactions induced by harmonic
excitation (Jia and Ulfvarson, 2005), sound radiation
from a plate (Skoblar et al., 2007), identification of
damage on a cross stiffened plate of a ship model
(Budipriyanto et al., 2007) and fatigue assessment of
a welded joint subjected to combined load (Garbatov
et al., 2010).
A recent and very comprehensive review of analytical, numerical, and orthotropic plate procedures that
were developed for analysis of stiffened plates and
shells is given in Bedair (2009).
This paper presents a modal analysis of a ferry deck
panel that was tested by Ferrari and Rizzuto (2003a).
The object of the study is the stiffened panel, which
consists of the horizontal plate and vertical stiffeners
in longitudinal and transversal directions. The numerical model of the plate is built in the finite element
software ANSYS, using the shell elements for the
plate and stiffeners. The comparison of the numerically obtained natural frequencies and mode shapes
with the experimental data available from structural
laboratory of the Department of Naval Architecture
349
Table 1.
Dimensions of the plate and stiffeners.
Plate, m
Longitudinal stiffeners, m
Length of transversal
stiffeners, m
Dimensions of T profile, m
Table 2.
2 2.49 0.006
2.1 0.1 0.006
2.54
Local Mode
Numerical Numerical
Experimental,
mode Number (Ansys), Hz (Nastran), Hz Hz
1.1
1.1
1.1
1.1
1.1
2.1
2.1
2.1
2.1
2.1
3.1
3.1
3.1
3.1
3.1
4.1
4.1
4.1
4.1
4.1
5.1
5.1
5.1
5.1
5.1
0.2 0.06 / 0.05 0.006
Figure 1. Finite element model of a stiffened plate.
and Marine Technology of the University of Genova

(Ferrari and Rizzuto, 2003a) is carried out. Final
comparisons show good agreement of numerical and
experimental data.
3
2
Comparison of natural frequencies.
FINITE ELEMENT MODEL OF THE PANEL
The object of the study is a ferry deck panel which

consists of the plate reinforced with two transversal
(T profile) and six longitudinal (beam) stiffeners.
The dimensions of the plate and stiffeners are given in
Table 1. The distance between longitudinal stiffeners
is 0.415 m and between transversal stiffeners is 1.8 m.
The length of the stiffeners is such that they protrude
from the plate in longitudinal and transversal direction
for 0.05 m and 0.025 m, respectively.
A numerical analysis of the plate is carried out in
finite element software ANSYS (Figure 1). According to the experimental model by Ferrari and Rizzuto
(2003a), as boundary conditions, four simple supports
are placed in the vertical direction at the intersection of transversal with second and fifth longitudinal
stiffeners, while relating to the shape of the observed
structure, as the most appropriate elements for the
plate and stiffeners (longitudinal and transversal) shell
elements with 4 nodes are chosen.
Using the convergence criterion of the natural frequencies in the frequency range up to 250 Hz, depending on the mesh refinement, an average shell edge
size is chosen as 50 mm since the further reduction of
element size does not lead to the significant changes
in natural frequencies.
The material is the standard steel with the modulus of elasticity, E = 210 GPa, Poisson coefficient,
= 0.28 and the density = 7700 kg/m3 .
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
61.58
70.19
81.35
113.10
124.83
130.87
144.02
173.70
142.75
156.33
168.56
191.63
203.85
177.70
191.26
193.29
221.53
235.49
219.83
231.88
258.32
259.06
271.36
71
79
87
106
110
121
136
147
179
194
154
169
181
210
224
190
204
211
237
63
68
71
101
104
93
114
131
167
180
107
118
125
188
198
141
153
159
214
221
181
199
207
248
251
COMPARISON OF NUMERICAL AND

EXPERIMENTAL RESULTS
Modes of the plate can be classified as global,

where the plate behaves as whole and local which
are defined by the dynamics of subpanels between
stiffeners. In the experimental modal analysis (Ferrari
and Rizzuto, 2003a), it was found out that only local
modes exist. Thus, only the equivalent mode shapes
from numerical simulation are compared with experimentally obtained natural frequencies. In Table 2
comparison of experimental data (Ferrari and Rizzuto,
2003a), numerical data from Nastran (Ferrari and
Rizzuto, 2003b) and numerical data presented in this
work is given.
For simplicity of the result presentation local mode
shape of the subpanel is designed as (m, n), where m
is the number of lobes in longitudinal direction and
n is the number of lobes in transversal direction, i.e.
local mode (2, 1) indicates the existence of 2 lobes in
longitudinal direction and 1 in transversal, in at least
one of the central subareas of the panel.
Good agreement between numerical results from
Ansys and experimental results (Ferrari and Rizzuto,
2003a) is observed, especially in the lower frequency
range (Table 2). Also, in comparison with the previously available numerical results from Nastran,
deviation of natural frequencies is reduced for up to
15%. Since the simulation was performed without
scallops at the intersection between longitudinal and
transversal stiffeners, fourth and fifth modes of the
350
Figure 2. Local mode 2, 1 (113.1 Hz).

and stiffeners. Furthermore, comparison of the natural frequencies with the experimental data previously
obtained by Ferrari and Rizzuto is carried out and
satisfactory agreement between numerical results and
experimental results is observed.
Also, in comparison with the previously available
numerical results from Ferrari & Rizzuto (2033b), the
deviation of natural frequencies is reduced. Since the
validation of the model is satisfactory, future investigations on the verified model, such as transient dynamic
analysis can be performed.
ACKNOWLEDGEMENTS
local mode (1, 1) were not observed. In Fig. 26 the
series of panel modes of the local mode (2,1), ordered
with increasing modal frequency are shown.
The difference between the modes of the same
local shape is observed in the magnitude of relative
amplitudes of subpanel modes and the way they are
positioned.
This work has been performed during the visit that the
first author made to the Centre for Marine Technology
and Engineering of Instituto Superior Tcnico,
Technical University of Lisbon, as part of her PhD
studies at the University of Rijeka.
REFERENCES
CONCLUSIONS
In this work modal analysis of a ferry deck panel

using a finite element code is carried out. The simulation is performed with shell elements for the plate
Bedair, O., 2009,Analysis and Limit State Design of Stiffened

Plates and Shells: A World View, Transactions of ASME,
Vol. 62, pp. 116.
Budipriyanto, A., Haddara, M.R., Swamidas, A.S.J., 2007,
Identification of Damage on Ships Cross Stiffened Plate
351
Panels Using Vibration Response, Ocean Engineering,

Vol. 34, pp. 709716.
Den Hartog, J. P., 1947, Mechanical Vibrations, 3rd edition,
Mc Graw-Hill Book Company, Inc., New York.
Ferrari, A., Rizzuto, E., 2003a, Experimental Modal Analysis of a Ship Deck Structure, Proceedings of NAV 2003,
Palermo, Italy, June.
Ferrari, A., Rizzuto, E., 2003b, Vibrational Behavior of
a Deck Panel of a Fast Ferry, Proceedings of the 7th
International Conference on Fast Sea Transportation,
Napoli, Italy, 710 October.
Garbatov, Y., Rudan, S., Guedes Soares, C., 2010, Fatigue
assessment of welded trapezoidal joints of a very fast
ferry subjected to combined load, Engineering Structures,
Vol. 32, pp. 800807.
Jia, J., Ulfvarson, A., 2005, Structural Behavior of a

high Tensile Steel Deck Using Trapezoidal Stiffeners
and Dynamics of VehicleDeck Interactions, Marine
Structures, Vol. 18, pp. 124.
Nacy, S.M., Alsahib, N. K., Mustafa, F. F., 2009, Vibration
Analysis of Plates With Spot Welded Stiffeners, Jordan
Journal of Mechanical and Industrial Engineering,
Vol. 3, pp. 272279.
Skoblar, A., timac, G., Braut, S., igulic, R., 2007,
Experimental and Numerical Analysis of the Sound
Radiation From aVibrating Plate, Proceedings of 6thYouth
Symposium on Experimental Solid Mechanics, Vrnjacka
Banja, Serbia, 912 May.
Troitsky, M. S., 1976, Stiffened Plates, Bending, Stability and
Vibrations, Elsevier, New York.
352

Ultimate strength assessment accounting for the effect

of finite element modelling
M. Tekgoz, Y. Garbatov & C. Guedes Soares
ABSTRACT: The objective of this work is to analyze the effect of different finite element models on the
ultimate strength assessment of plates and stiffened panels. The analysis is performed by the finite element
method using commercial software, where the effect of element size and type, boundary conditions, shape of
initial imperfection, thickness and net sectional configuration on the ultimate strength are evaluated. The ultimate
strength estimated for four different finite element models and different structural configurations are compared
to the solution defined by the new Common Structural Rules.
1
INTRODUCTION
Ships are built of plates, stiffeners, girders, bulkheads

and frames that contribute to its strength. The plates
are supported by the stiffeners, which keep them in
position to absorb the applied loadings. The frames
that prevent the stiffeners to buckle and support them
are stiffer and deeper than the stiffeners, which eliminate the overall grillage buckling. Ship structures are
assembled of stiffened panels and an estimation of
the maximum load carrying capacity or the ultimate
strength of these elements is of high importance for
the safety assessment and economical design of the
structure.
Ultimate strength of structural components and systems is a measure in strength assessment in a sense that
the ultimate strength is the maximum carrying capacity that they can withstand. No additional load can
be carried beyond the ultimate strength (Paik et al.,
2009). Achieving the ultimate strength implies that the
residual capacity deteriorates due to plasticity.
Guedes Soares and Kmiecik (1995) found that edge
boundary conditions have no significant effect on ultimate strength for stocky plates with large deflections.
They found that it increases the strength by % 10
for a practically perfect plate due to the restraining
of the edges and it has been found that when initial
imperfections are significant, the strength becomes
insensitive to edge conditions. Due to the fact that
strength reduction can arise from the edge supports,
when the ultimate strength of the ship is to be evaluated, the possible weakest structural element must be
selected.
Guedes Soares and Kmiecik (1993) have also
shown that assessing the plate strength with a more
accurate non-linear finite element code would lead
to the same general type of results although with
different numerical values.
Xu and Guedes Soares (2011) investigated the

effect of geometry and boundary conditions on the
strength behaviour of short stiffened panel and it has
been found that unconstrained longitudinal edges lead
to strength reduction and as the structure is weakened,
the dimensional influences on the ultimate strength
become more evident.
Ultimate strength assessment involves a large
amount of uncertainties and many factors may affect
it. While Matthies et al. (1997) discussed the general problem of uncertainty analysis of structures,
including detailed methods of analysis, Guedes Soares
(1988) analysed the more specific problem of uncertainty in the collapse strength of plate elements and
concluded that plate slenderness was a critical variable
for strength uncertainty.
Garbatov et al. (2011) implemented a Monte
Carlo simulation in order to find the most influential parameters on the ultimate strength. It has been
found that slenderness ratio and plate thickness have
the most significant effect on the ultimate strength
of the stiffened panels. In addition to that, residual stresses and different loading conditions, imperfection magnitude and material properties can be
added.
The objective of this work is to analyse the
effect of finite element modelling on the ultimate
strength assessment of different structural configuration of plates and stiffened panels cut from box
girders and prepared to be tested against compressive load. The analysis is performed by the finite
element method using commercial software, where the
element size and type, shape of initial imperfection
and boundary conditions sensitivities have been performed. The ultimate strength estimated for different
finite element models is compared with the solution
defined by the new Common Structural Rules (IACS,
2010).
353
Figure 3. Model 1 geometry.
Figure 1. Structural models 1 to 3.
Figure 4. Model 4 geometry.
Figure 2. Structural model 4.
2
2.1
FINITE ELEMENT MODELLING

Structural description
The four structural components analysed here are part

of box girders that have been tested and analysed by
Saad-Eldeen et al. (2011b, a, 2012). Models 1 to 3
are shown in Figure 1. Model 1 (see also Figure 3)
is a stiffened panel located between two neighbouring
transverse frames. Model 2 is also a stiffened panel
with a transverse frame at the middle and Model 3 is a
plate limited by two neighbouring transverse and two
longitudinal frames stiffeners respectively. Model 4,
which can be seen on Figure 2 and 4, is composed by
+ 1 + plate bays and two transverse frames.
The ultimate strength of panels and plate are analysed based on the finite element method using commercial softwareANSYS (2009).The software enables
modelling of elastic plastic material properties and
large deformations. Eight-node and four-node quadrilateral shell elements have been used to model the
plates and stiffeners.
The kinematic assumption of finite element analysis calculation is large displacement and rotation, but
small strain. The material modelling is assumed to
be bilinear elastic-perfectly-plastic without hardening
with the yield stress of y = 235 MPa, the elastic modulus is E = 2.1 E11 Pa and the Poison coefficient is
= 0.3. Both a load-displacement control method and
an automatic displacement control method are used

in the solution scheme. The applied load is uni-axial
compression.
The length of the stiffened panels is l = 400 mm and
the breadth is b = 150 mm respectively (see Figure 3).
The web thickness is 3.6 mm and the height is 25 mm
and the plate thickness is 2.7 mm. Plate slenderness
is = 1.87. The aspect ratio 1/b of the plate is 2.66.
The stiffener has a cross section of a standard type flat
bar. The area of the stiffener, Ast is approximately 22
per cent of the area of the plate, Apl = bl. The nondimensional column slenderness ratio of the stiffened
panel calculated with a full plate width is = 0.69.
The thickness, length, breadth of plates and the
height of stiffeners are equal for all finite element
models.
The initial geometry imperfection of plates and
stiffeners are generated by a pre deformed surface.
Faulkner (1975), Smith et al. (1988) have reported
that the maximum imperfections in a plate can be
assumed to be proportional to 2 and they suggested
that the maximum initial deformation for an average
imperfection can be calculated as
where t is the plate thickness and is the plate

slenderness.
The maximum permissible camber tolerance, for
a standard shape is usually assumed to be 0.2% of
the length of the plate. The initial geometry surface
imperfection is modelled as:
where l is the length of the panel and b is the breath of

the panel, x and y are the Cartesian coordinates of any
354
Table 1.
Case study of Mesh Analysis of Model 1 and 2.
Case
l, mm
b, mm
h, mm
No of half
waves
1
2
400
400
150
150
25
50
1
1
Table 2.
Element size study cases, Model 3.
Case
l, mm
b, mm
No of half
waves
1
2
3
4
400
400
800
400
150
150
150
75
1
2
1
1
Figure 5. Element quantity, Case 1, Model 2.
location on the plate and m and n are number of half

waves assumed.
The transverse frame of Model 2 is not modelled by
finite elements and its effect is accounted for by the
respective boundary conditions applied to the nodes
associated with the connection between the plate and
the attached stiffener.
A buckling analysis with the prescribed surface of
plates and stiffeners is performed before the incremental load-displacement analysis.
2.2 Element size effect
The element size of the finite element models is a
very important parameter, which highly influences
the calculated load capacity of structures. To analyse this effect four different study cases are checked
here associated with different model lengths, breadths
and number of half waves of imperfection shape (see
Table 1 and 2).
The element size of the finite element model of the
web of the stiffener has a big influence on the estimated
ultimate strength of the models 1 and 2. As can be seen
from Figure 5, the load carrying capacity of the stiffener panel, Model 2 achieves its minimum when the
web has 5 elements and it is decreasing or increasing
with the number of element used to model the web.
This point, the inflection point, may change its location (number of elements) accounting for the structural
configuration of the analysed stiffened panel.
The trend of the estimated ultimate strength is stabilized, once the web is modelled by 15 finite elements
for Case 2, Model 2 (see also Figure 6), which results
in the lowest load carrying capacity and less potential energy absorbed by the constructed finite element
model. It can be concluded that the use of coarse
meshes results in overestimation of the load carrying capacity and for any specific configuration of
the analysed structure an appropriate finite element
size should be identified by performing element size
analysis.
Figure 6. Element quantity, Case 2, Model 2.
Figure 7. Element size, Case 4, Plate.
A very similar solution is found out about Case 4

(see Figure 7) where instead of the number of elements
the admissible element size is studied and for this particular case it approaches to 4 mm. Figure 8 shows
the respective element sizes when Model 3 exhibits
the minimum load carrying capacity according to the
cases presented in Table 1 and 2.
In order to find out the most appropriate element
size, first the dimensions and shape imperfection of
355
Figure 8. Element sizes as a function of the minimum load

carrying capacity, Model 3.
Figure 9. Element types, Model 1.
the plate have been varied. It has been found that the
mesh size is not affected by the plate failure mode
in contrary to what has been observed in the case of
stiffener tripping.
2.3
Element type effect
The element type effect analysis on ultimate strength

is performed here. The problem is to find the most
suitable type of elements that may be used to model
the physical behaviour of the analysed structures. The
element type chosen should have capacity to cover geometrical and material non-linearity in any condition
without leading to non-convergence problems.
Shell 93 and Shell 181, which are eight-nodes and
four-node quadrilateral element types are analysed
here for finite element modelling of Model 1 and
Model 2 and the resulting ultimate strength is compared to the closed-form solution stipulated by IACS
(2010).
Stress stiffening-strains are assumed to be small,
and stiffening (or softening) of the structure due to
the stress state is taken into account. For Shell 181
element, stress-stiffening is not available independent
of large deflection. For other elements, the decision
to include stress-stiffening with large deflection is
generally based on ease of convergence, since large
deflection and stress-stiffening are redundant.
The finite element Shell 181 yields larger carrying capacity than the one of Shell 91. Shell 181 better
captures the behaviour of the stiffened panels (see Figure 9 and Figure 10). Shell 181 is used for Model 2
and Model 4 and Shell 93 is used for Model 1 and
Model 3 which have been considered less vulnerable
to compressive load.
2.4
Structural cross-section effect
Figure 10. Element types, Model 2.
sizes and imperfection shapes have been applied. It

has been found that the structural cross-section effect
is insignificant on the element size definition (see
Figure 11). However, it has a significant importance
on the ultimate strength.
It can be observed from Figure 12, that the crosssection structural configuration has significant influence on the strength. Transverse configuration has
more influence than longitudinal configuration on the
ultimate strength of this particular case. Garbatov et al.
(2011) reported that it does not always reflect in the
strength reduction in the case when the number of half
waves related to the shape of initial imperfection are
increasing and they found that as the plate surface is
modelled by 4 longitudinal half waves, the estimated
ultimate strength is larger than in the case of a surface
moddeled by one longitudinal half wave.
2.5 Thickness effect
The effect of the shape of the cross-section of stiffened panel, accounting for the shape of the initial
imperfection and element size on the ultimate strength
is investigated here. For this purpose, Model 2 has
been selected to be analysed and different element
The thickness effect on the admitted element size and

the ultimate strength has been investigated here. It
has been observed that there is no difference on the
admissible element size as a function of the minimum estimated ultimate strength as either increasing
356
Figure 11. Structural imperfection shape and structural configurations, Model 2.
Figure 13. Thickness effect, Model 2.
Figure 12. Structural shape effect, Model 2.
or decreasing the plate and stiffener thicknesses. However the plate thickness variation has a significant
influence on the ultimate strength (see Figure 13) has
to be pointed out that the web height, for this particular
case, has the most significant influence on the element
size.
3
3.1
ULTIMATE STRENGTH ANALYSIS

Model 1
Model 1 is made up of a stiffened panel, which is

supported by two neighbouring transverse frames (see
Figure 1 and Figure 14). The different boundary conditions analysed here are shown in Table 3. As can be
seen in Figure 14, symmetry boundary conditions are
applied, where the maximum imperfection is considered to occur. The coupling conditions are employed
on the longitudinal edges in order to keep the section
plane.
The transverse edges have been subjected to different rotational effects and appropriate translations as
given in Table 3. The coupling effect is implemented
Figure 14. Model 1.
considering that the existing transverse frames are stiff

enough to keep the sections plane.
Applying BCs 1 and 2, the effect of rotational constrains on the ultimate strength has been investigated.
In these particular cases, there is no significant difference in the strength behaviour as can be seen from
Figure 15, which can be explained with the fact that
the transverse edges are stiff enough.
357
Table 3.
Boundary conditions, Model 1.
Lines constrained to
L1-5: Uy, Uz, Rx
L1-5: Uy, Uz
L1-5: Uy, Uz, Ry
L1-5: C- Ux
L6-7: Uy, Uz
L6-7: Uy, Uz, Rx
L6-7: Uy, Uz, Ry
L6-7: C-Ux
L2-4: Ux, Uy, Uz
L2-4: Ux, Uy, Uz, Rx
L2-4: Ux, Uy, Uz, Ry
L3-8: Ux, Uy, Uz
L1-2: Uy, Rx
L1-2: Rx
L5-4: C-Uy
L5-4: Uy, Rx
L5-4: Rx
BC1 BC2 BC3 BC4 BC5 BC6 BC7

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Figure 16. Model 1, BCs 3, 4, 5 and CSR.
x
x
x
x
x
x
x
x
x
x
x
*U-Displacement, R-Rotation, C-Coupling.
Figure 17. Model 1, slenderness ratio effect.
Figure 15. Model 1, BCs 1, 2, 6, 7 and CSR.
The coupling effect on the ultimate strength has

been also analysed for BCs 1 to 7 and in the case where
the loaded edges (transverse edges) are coupled in longitudinal direction, the post-collapse regime is better
captured (see Figure 15). This is due to the fact that
the longitudinal edges have been considered imperfect; therefore the edge effectiveness is low. The post
collapse load carrying capacity is decreasing with less
plasticity and makes it smoother.
The coupling effect is also investigated for BCs 4
and 5. In the case when the longitudinal edge is coupled
in the transverse direction, this leads to a higher load
carrying capacity, but sharper post-collapse regime
(see Figure 16). The restrained transverse edge leads to
higher load carrying capacity and also the stress strain
curve behaviour changes to a plate failure, which is
observed for BCs 3 and 4 (see Figure 16).
Applying BCs 1 to 6 show that the transverse edge
restrained in terms of rotations gives larger load carrying capacity. It can also be observed that the post
collapse regime is more gradual due to the plasticity
propagation (see Figure 15).
Figure 18. Model 1, slenderness ratio effect.
As can be seen from Figure 17 and Figure 18, when

the slenderness ratio increases, the structure becomes
more sensitive to edge boundary changes. This kind
of structural response, for an example, is characteristic of a structure subjected to uniformly distributed
corrosion deterioration.
The normalized ultimate strength as a function of
the applied BC is shown on Figure 19. It can be
observed that the estimated ultimate strength based
on CSR is the lowest.
3.2 Model 2
Model 2 is made up of a stiffened panel between two
half bays, + . The boundary conditions for the
studied cases may be seen in Table 4. The longitudinal
358
Table 4.
Line constrains
L3-5: Ry
L3-5: Ry, C-Ux
L4-12: Rz
L4-12: Rz, C-Ux
L2-6: Uz
L10-11:Uz, C-Uy
L1-7: Ux, Ry
L8-9: Ux, Rz
L1-3: Rx, Uy
L1-3: Rx
L5-7: Rx
L5-7: Rx, C-Uy
Point 10-Uy
Figure 19. Model 1, BCs 1 7 and CSR.
BC1
BC2
BC3
BC4
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
*U-Displacement, R-Rotation, C-Coupling
Figure 20. Model 2.
Figure 22. Model 2, BCs 1 to 4 and CSR.
Figure 21. Model 2, BCs 1 to 4 and CSR.
and transverse edges (see Figure 20) are subjected

to symmetry boundary conditions, since the initial
maximum imperfection has been generated in the
longitudinal and transverse edges and also coupling
conditions are employed in order to keep the section
plane during loading.
The effect of four boundary conditions for the ultimate strength assessment is analysed here. BCs 1 and
2 show that using coupling conditions, on the longitudinal edges in the transverse direction, give larger
ultimate strength and because of the sufficient plasticity it makes the post-collapse regime smother (see
Figure 21 and 22). In addition to that, BCs 2 and 3
show that in the case when the loaded transverse edges
are not coupled in longitudinal direction, the structural
response yields to a significant load carrying capacity
reduction (see Figure 21 and 22).
Figure 23. Model 2, BCs 1 and 2, plate slenderness ratio.
BCS 2 and 4 show that the longitudinal edge

restraining leads to a larger load carrying capacity and
also to a smoother post-collapse regime (see Figure
21). On the other hand, when the model 1 is observed
closely (see Figure 16), the transverse restraining
does not result in significant behaviour change in
terms of ultimate point but has a similar effect in the
post-collapse regime.
As the plate slenderness ratio increases, the structure becomes more sensitive to boundary edges as can
be seen in Figure 23, which has been also observed for
Model 1.
359
Figure 24. Model 4.

Figure 25. Model 4, BCs 1 to 4.
Table 5.
Line Number
L1-3: Ry
L1-3: Ry, C-Ux
L2-11: Rz
L2-11: Rz, C-Ux
L4-10: Uz
L12-13: Uz, C-Uy
L14-15: Uz, C-Uy
L5-9: Uz
L8-6: Ux, Ry
L7-16: Ux, Rz
L1-8: Rx, Uy
L1-8: Rx
L3-6: Rx
L3-6: Rx ,C-Uy
Point 13, 15: Uy
BC1
BC2
BC3
BC4
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Figure 26. Model 2 and 4 with thickness reduction, BC2.
* U-Displacement, R-Rotation, C-Coupling
3.3
Model 4
Model 4 is made up of a stiffened panel between two

half and one entire bays, + 1 + . The analysed
boundary conditions are presented in Table 5.
The longitudinal and transverse edges are subjected to the symmetry boundary condition, since the
maximum imperfection has been considered in the longitudinal and transverse edges (see Figure 2 and 24).
Coupling conditions are employed in order to keep the
section plane during loading.
It has been seen that the Model 4 has larger carrying capacity than Model 2 as can be seen from Figure
25 to Figure 28. However the Model 4 has larger ultimate strength but the Model 2 has larger post-collapse
load carrying capacity. It has to be pointed out that
the ultimate strength capacity is more important and
therefore the Model 2 is more appropriate as a model
to evaluate the ultimate strength of the ship structure.
Figure 27 and Figure 28 show models M-2 and M-4
respectively. Model 2 and 4 exhibit similar behaviour
in terms of ultimate strength against the edge boundary
changes (see Figure 27). In addition, as the plate and
stiffener thicknesses are decreased, they behave closer
in terms of strength reduction.
Figure 27. Model 2 and 4 with no thickness reduction.
3.4 Model 3
Model 3 is made up of a single plate bordered by two
neighbouring transverse frames and longitudinal stiffeners as may be seen in Figure 1 and 29. The studied
boundary conditions are presented in Table 6.
The displacements of longitudinal and transverse
edges are all constrained in vertical direction. The
longitudinal edges are not subjected to symmetry
360
Figure 30. Model 3, BCs 1 to 3.

Figure 28. Model 2 and 4 with 1 mm thickness reduction.
Figure 29. Model 3.

Figure 31. Model 1 to 4 and CSR.
Table 6.
Line Number
BC1
L1-2: Uy, Uz
L1-2: Uy, Uz, Ry
L1-2: Uy, Uz, Rx
L2-3: Uy, Uz
L3-4: Ux, Uy, Uz
L4-1: Uy, Uz
BC2
BC3
x
x
x
x
x
x
x
x
x
*U-Displacement, R-Rotation
boundary condition. Since the plate has been considered bounded by the stiff transverse frames and
longitudinals, coupling was not employed.
Changing the boundary conditions of the transverse
edges does not change the behaviour of pre-collapse
region and results in a larger post-collapse loading
capacity. As the loaded transverse edge is constrained
against rotation, this leads to an earlier first yielding
occurrence (see Figure 30).
COMPARATIVE ANALYSIS
Three finite element models of stiffened panels have

been compared here. Model 1 and Model 2 are selected
due to close match behaviour to the closed form CSR

solution for the ultimate strength assessment. The element size for all the models are defined based on
admissible element size sensitivity analysis in order
to avoid misinterpretation.
Model 3 has larger ultimate strength than Model
1. Model 1 has a larger post collapse load carrying
capacity than Model 3. As expected, Model 2 has a
larger load carrying capacity than Model 1. Model 2
has larger post-collapse load carrying capacity than
Model 4. Model 1 exhibits weaker behaviour in contrary to Model 2 in terms of boundary conditions
change since Model l edge boundary conditions show
better performance as compared to Model 2.
Figure 31 and 32 shows a comparison between the
ultimate strength of the four different finite element
models and CSR solution. It can be seen that the CSR
solution of the ultimate strength is the lowest.
CONCLUSIONS
In this work the boundary condition effect has been

investigated on the ultimate strength of three stiffened
panels and one plate. It has been found that in the
case when the boundary conditions of the edges are
not fully constrained they lead to a sudden and smaller
load carrying capacity and to non-convergence problems. The post-collapse regime is highly affected by
361
Figure 32. Model 1 to 4 and CSR.
the edge boundary conditions. The restrained boundary conditions give a larger post-collapse loading
capacity. The boundary conditions may change the
failure mode too. As the plate slenderness is increased,
its sensitivity to boundary changes increases. The
structure cross-section configuration has a significant
influence on the ultimate strength. Model 2 and Model
4 are similarly affected by the boundary condition
changes with respect to the ultimate strength. Model 2
has larger post-collapse loading capacity than Model
4. Although Model 1 is stiffer, it has similar ultimate strength and larger post collapse load carrying
capacity than Model 3. As the thickness is decreasing,
Model 2 and Model 4 behave closely. Model 2 is the
most appropriate finite element model to calculate the
ultimate strength of stiffened panels.
ACKNOWLEDGMENTS
The work reported here is a contribution to
the MARSTRUCT VIRTUAL INSTITUTE, (www.
marstruct-v.com) in particular its Technical Subcommittee 2.3 on Ultimate Collapse Strength.
REFERENCES
ANSYS, 2009. Online Manuals, Release 11.
Faulkner, D., 1975. A Review of Effective Plating for use
in the Analysis of Stiffened Plating in Bending and
Compression. Journal of Ship Research 19, pp. 117.
Garbatov, Y., Tekgoz, M., Guedes Soares, C., 2011. Uncertainty Assessment of the Ultimate Strength of a Stiffened
Panel, in: Guedes Soares, C., Fricke, W. (Eds.), Advances
in Marine Structures. Taylor & Francis Group, London,
UK, pp. 659668.
Guedes Soares, C., 1988. Uncertainty Modelling in Plate
Buckling. Structural Safety 5, pp. 1734.
Guedes Soares, C., Kmiecik, M., 1993. Simulation of the
Ultimate Compressive Strength of Unstiffened Rectangular Plates. Marine Structures 6, pp. 553569.
Guedes Soares, C., Kmiecik, M., 1995. Influence of the
Boundary Conditions on the Collapse Strength of Square
Plates with Initial Imperfections. Graczyk, T.; Jastrzebski,
T.; Brebbia C. A., and Burns, R. (Eds), Marine Technology and Transportation. Computational Mechanics
Publications, Southampton, pp. 227235.
IACS, 2010. Common Structure Rules for Double Hull Oil
Tankers, Consolidated version, July 2010.
Matthies, H., Brenner, C., Bucher, C., Guedes Soares, C.,
1997. Uncertainties in probabilistic numerical analysis of structures and solids-Stochastic finite elements.
Structural Safety 19, pp. 283336.
Paik, J., Branner, K., Choo, J., Czujko, J., Fujikubo, M.,
Gordo, J.M., Parmentier, G., Iaccarino, R., ONeil,
S., Pasqualino, I., Wang, D., Wang, X., Zhang, S.,
2009. Committee III.1 Ultimate Strength, in: Jang, C.,
Hong, S. (Eds.), 17th Int. Ship and Offshore Structures
Cong. (ISSC2009). University of Seoul, South Korea,
pp. 375475.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2011a.
Corrosion Dependent Ultimate Strength Assessment of
Aged Box Girders Based on Experimental Results.
Journal of Ship Research 55, pp. 289300.
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2011b.
Experimental Assessment of the Ultimate Strength of
a Box Girder Subjected to Severe Corrosion. Marine
Saad-Eldeen, S., Garbatov, Y., Guedes Soares, C., 2012.
Effect of Corrosion Degradation on the Ultimate Strength
of Steel Box Girders. Corrosion Engineering, Science and
Technology 47, pp. 272283.
Smith, C.S., Davidson, P.C., Chapman, J.C., Dowling, J.P.,
1988. Strength and Stiffness of Ships Plating under
In-plane Compression and Tension. Transactions RINA
130, pp. 277296.
Xu, M.C., Guedes Soares, C., 2011. Numerical study of the
effect of geometry and boundary conditions on the collapse behavior of short stiffened panels, Guedes Soares,
C., Fricke, W. (Eds.), Advances in Marine Structures.
Taylor & Francis Group, London, UK, pp. 229237.
362

Fatigue damage assessment of a tanker structural detail based on the

effective notch stress approach
K. Tran Nguyen, Y. Garbatov & C. Guedes Soares
ABSTRACT: This work deals with a fatigue damage assessment of a double hull oil tanker structural detail
based on a local structural finite element model. The wave-induced vertical and horizontal bending moments,
as well as local pressure loads have been considered in the fatigue damage calculations. Stress analyses are
performed based on hot spot and notch stress approaches. The stress distributions at the notch surface of weld
toe are studied. Two hot spots are analyzed at the weld toe located between the flat bar stiffener of a transverse
web frame and the flange of a longitudinal stiffener at the side shell of a tanker ship hull.
1
INTRODUCTION
Fatigue assessment of welded joints and components in

ship structural details has been performed by using the
nominal stress based S-N curves (Hobbacher, 2007,
IACS, 2010). The nominal stress is easily calculated
directly from the beam theory on empirical formulae
(DnV, 2005), as well from coarse finite element analysis. However, for obtaining an accurate assessment
of stress response of the hull structure, the hot spot
stress approach is applied based on the finite element
(FE) method and its characteristic instead of the nominal stress approach. The hot spot stress approach for
structural details has been included for fatigue design
by Hobbacher (2007). The hot spot stress is defined
as local stress obtained by extrapolating stresses at
certain distances away from a geometrically discontinuous area such as weld toe. Some of the limitations
of the hot spot stress approach are that the notch
stresses cause by weld beads are excluded from the
total stresses (Figure 1).
Fricke and Kahl (2005) applied three different
structural stress approaches to fatigue strength
assessment of three welded structural details. Fatigue
lives were predicted using the design S-N curves
recommended within the different approaches and
compared with the results of fatigue tests evaluated
for a corresponding probability of survival.
Chakarov et al. (2008a) have analyzed symmetrical and unsymmetrical longitudinally stiffened panels, determining hot spot stress distribution and
stress concentration factors, in view of thickness
change misalignment, angular imperfections, rotation
of transverse weld toe and residual deformations. Also,
polynomial regression equations assessing the stress
concentration factors have been derived.
Radaj et al. (2009) developed the effective notch
stress approach by introducing fictitious effective
Figure 1. Hot spot stress approach.
Figure 2. Effective notch stresses approach.
notches of radius 1 mm to weld toes or weld roots

(Figure 2). Further studies have been performed and
the effective notch stress approach has been included in
the IIW fatigue design recommendations (Hobbacher,
2007).
Saad-Eldeen et al. (2010) analyzed the hot spot
and effective notch stresses accounting for the effect
of contact elements of two welded specimens (doubler plates and double lap joint) subjected to fatigue
load. Based on the results of the FE models analyzed,
polynomial regression equations assessing the stress
concentration factors at the weld toe and weld root
were derived.
363
Figure 3. Hot spot A and B locations.
Gaspar et al. (2011) proposed a simulation procedure using the structural hot spot stress approach
combined with the FE analysis and the Monte-Carlo
simulation method to examine the effect of the weld
shape uncertainties (the geometrical imperfections
induced by the welding process commonly used in
the shipbuilding and repair industries) on the structural hot spot stress distribution. A structural detail of
containerships deck structures was analyzed.
Moreover, Fricke et al. (2002) performed the fatigue
strength assessment based on different classification
societies for a hatch cover bearing pad in the ISSC
comparative study.
Garbatov and Guedes Soares (1998, 2002) presented a formulation for the assessment of the reliability of a local ship hull structure regarding to fatigue
failure. The potential cracks are considered to occur
in the side shell, in the connections between longitudinal stiffeners and transverse web frame. The analysis is
performed for T shaped stiffeners. The model accounts
for the crack growth process applying the linear elastic
fracture mechanics. The long-term stress range acting
on the elements is defined as a function of the local
transverse pressure of internal cargo and outside water,
combining with the stresses resulting from the longitudinal bending of the hull. The global hull loading
is a combination of horizontal and vertical bending
moments.
Normally, fatigue cracks may initiate at hot spot
stress points of the weld toe and grow though the thickness of plate such as the connection between the flat
bar stiffener attached to a transverse web frame (webstiffener) and the flange of a longitudinal stiffener at
the side shell (see Figure 3). During the inspection,
fatigue cracks may be detected by the surveyor at an
earlier stage in ship structures, or they may be detected
after penetrating the side shell. In the majority of cases,
there are two hot spot points, one is at the toe of the
web-stiffener (HSA), and the second is at the flange of
the flat bar close to the cut-out (HSB), as can be seen
from Figure 3 and 5.
To define the local stresses distribution at the intersection between the longitudinal stiffeners and the
transverse web frame, beside a procedure based on
the simplified analysis, a finite element analysis of
a stiffener panel at side shell connections is presented
Figure 4. Midship section of the oil tanker.

Table 1. Vessel hull characteristics.
Length overall, Loa
Length between perpendiculars, Lpp
Breadth moulded, B
Depth moulded, D
Design draft (state of full loaded), Tf
Ballast draft (state of ballast loaded), Tb
Block coefficient, CB
Maximum service speed, v
Web frame spacing, ls
Moment of inertia about the vertical
axis, IC
Moment of inertia about the horizontal
axis, IN
Height of neutral axis above base line, no
Vertical wave hogging bending moment
Vertical wave sagging bending moment
Horizontal wave bending moment
(full loaded)
Horizontal wave bending moment
(ballast)
332 m
320 m
58 m
31 m
20.8 m
14 m
0.815
16 kn
5.12 m
4207.925 m4
1479.421 m4
13.525 m
4404303 kNm
4739923 kNm
2642960 kNm
2172486 kNm
here. The effective notch stress approach is applied for

investigating the fatigue damage of the weld joint at
the two hot spot positions. The hot spot stress approach
is also applied and the results are compared with the
ones obtained from the effective notch stress approach.
A solid element with 20 nodes and a shell element
with 8 nodes are used in the finite element analysis
model for the notch and hot spot stress approaches
respectively by employing the commercial software
ANSYS (2009).
Typical connections of longitudinal to transverse
structural elements at double side of an oil tanker
(VLCC) are considered (see Figure 4). The oil tanker
has already been designed, successfully built and
spent some years in-service. The principal particulars
of the oil tanker and the midship section properties are
given in Table 1.
364
Figure 6. Global bending moment load.

Figure 5. Structural plan of the connection detail.
The structural detail considered for the fatigue

analysis is a weld connection between the flat bar stiffener on a transverse web frame and the flange of a
longitudinal stiffener at side shell (see Figure 5).
SIMPLIFIED NORMAL STRESS ANALYSIS
The basis of the simplified procedure for fatigue

assessment of ship structural details is described by
DnV (2005). The stresses are divided into two levels:
global and local stresses.
The global stresses are composed of vertical and
horizontal bending moment (see Figure 6). The local
stresses are defined from the local stiffener bending,
local relative deflection effect of stiffener between
different transverse displacement of the transverse
bulkhead and the adjacent frame and the double
bottom/side bending because of external or internal
pressure.
The local and global bending stresses are calculated
using beam theory. The double side normal stresses
are calculated using parametric formulas (Storhaug
and Berstad, 2001).
The stresses induced by global and local load are
correlated. The long-term stress range distribution is
approximated by the Weibull two-parameter distribution. The Weibull distribution is used together with
the relevant S-N curve to calculate the accumulated
fatigue damage according to the Palmgren-Miner rule.
The fatigue analysis is performed according to the
current rules of DnV (2005). Two loading conditions
are considered: full and ballast loadings. The loading
conditions relate to the probable variations in loading
that the hull structure may experience during its service
life. The dynamic external and internal pressure acting
on the side shell in the two loaded conditions can be
seen in Figure 7 to 9.
The approach presented by DnV (2005) is based
on the hot spot stress approach. The stress concentration factor (SCF) is defined from a list of standard
details. However, the stress range is multiplied by a
stress concentration factor accounting for the presence of the weld to obtain the notch stress range. The
calculations include the stresses component from the
wave-induced vertical, horizontal bending moments
Figure 7. External pressure at the side shell (full loading).
Figure 8. External pressure at the side shell (ballast

loading).
Figure 9. Internal pressure at the side shell (ballast loading).
365
Figure 12. Finite element model.

Figure 10. Global stress range of longitudinal elements at
side shell (full loading).
Figure 13. Define loading condition.

Table 2.
Figure 11. Global stress range of longitudinal elements at
side shell (ballast loading).
and dynamic external/internal pressure based on the

load distributions at the region of the structural detail
with the Weibull shape distribution parameter of
0.819 and 0.9 corresponding to the probability of
occurrence of 104 for full and ballast loading condition respectively.
The global stress range (primary stresses) is
defined as nominal stresses in longitudinal hull girder
elements (IACS, 2010) at the side shell, as shown in
Figure 10, 11 for full and ballast loading conditions
respectively.
3
3.1

Finite element modeling
The developed model for a FE analysis is shown in

Figure 12. It is extended to 2 transverse web frame
spacing and 3 longitudinal stiffener spacing. The
symmetry conditions are defined along both the sides
of the model. The side longitudinal stiffeners are
connected to the transverse web-frame by flat bar
stiffeners (web-stiffener).
The load is applied in the way that a similar stress
distribution occurs in the real continuous side longitudinal elements subjected to the global and local loads
as defined by DnV (2005). It is assumed that either
the external/ internal water pressure loading or the
Loading cases.
Loading case
Full loaded
Ballast loaded
Internal pressure [kPa]

External pressure [kPa]
Axial loading [MPa]
81.50
37.25
31.00
57.30
30.85
axial loading are separated for each of the ship loading cases as shown in Figure 13. The loading cases are
summarized in Table 2.
The lateral external and internal pressure is applied
on the side shell plate. The lateral displacement is
restrained on the top of the transverse web frame and
all the edge ends of models are clamped. A constant
axial force is applied at the one end of models while
keeping the other end constrained about axial displacement and the lateral displacement at the position of the
transverse web frame. The modulus of elasticity and
the Poisson ratio are 206 [GPa] and 0.3 respectively.
FE models are created based on the net scantlings.
3.2 The effective notch stress approach

The accuracy of finite element analysis of welded
joints based on the effective notch stress approach
depends very much on the element types and the
mesh size. Different modeling techniques are used
for the global analysis and sub model techniques for
the local areas around the flange of the middle longitudinal stiffener and the attached web-stiffener as is
366
Figure 14. Global model and 1st sub-model.
Figure 16. Local FE coarse mesh, Shell 93 element.
The effective notch radius tips are defined to touch

the roots of real notches with the element size not
more than 1/6 of the radius in case of linear elements
and 1/4 of the radius in case of the higher order elements. This can be observed in the curved parts as well
as in the beginning of the straight part of the notch
surfaces in both directions tangential and normal to
the surface respectively (see Figure 2). Relatively fine
meshes are set up with an element size of 0.2 [mm]
along the circumference of effective notch in the third
sub-models.
3.3 The hot spot stress approach

Figure 15. 2nd and 3rd sub-models for hot spot A and B.
shown in Figure 14 & 15. The nodal displacements

from the global structural analysis are transferred to
the local sub model prescribed as boundary conditions. Although both shell elements and solid finite
elements are used for the assessment of welded joints.
The solid elements can provide better stress distribution because they may model precisely the shapes of
the weld connections.
The three-dimensional solid element Solid 95 is
used for the FE models. The Solid 95 element is a
higher order element with 20 nodes, three degrees
of freedom per node and translation in x, y and z
directions. This element may tolerate irregular shapes
without loss of the accuracy. It is used to comply
with the recommendation presented by Hobbacher
(2007). The Solid 95 element has been also used as
for an example by Chakarov et al. (2008b), SaadEldeen et al. (2010), Gaspar et al. (2011), Fricke et al.
(2012) for studying welded joints structural details.
For this study, the sub-models are analyzed by using
the tetrahedral-shaped solid elements.
The model of the weld shapes for the connection
between the flange plate and the flat bar plate is with
a leg length l = 4.5 [mm], as well as a gap is modeled
of 0.1 [mm] by introducing a clearance in the global
and in the sub-models. The fictitious effective notch
radius of 1 [mm] is included around the welded toes
as recommended by Hobbacher (2007).
To define structural hot spot stresses as has been shown

in Figure 1, a selection of the reference points associated with the element sizes are required. The hot spot
stresses are calculated from the surface linear stress
extrapolation for the weld toes on plates of thickness
t over the reference points located 0.4 t and 1.0 t away
from the weld toe in the fine mesh cases, as proposed
by Hobbacher (2007). For the relatively coarse mesh
having an element size txt and mid-side nodes, the
linear extrapolation can be also performed over 0.5 t
and 1.5 t, as recommended by Hobbacher (2007) and
DnV (2005).
According to DnV (2005), acceptable results may be
achieved by using 8-node shell elements with a mesh
size equals the thickness. The shell elements with 8
nodes (Shell 93) are used in the FE analysis to obtain
the hot spot stress. Figure 16 shows the coarser mesh
with quadrilateral-shape shell elements for the linear
stress extrapolation over the reference points at 0.5 t
and 1.5 t from the considered hot spot. Finer mesh
with an element size of 0.2 t 0.2 t is also used for a
similar analysis and for comparison.
FEM STRESS ANALYSIS
For the fatigue damage assessment of welded joints,

the principal stresses are used which are perpendicular
to the weld toe. For the present analysis, the longitudinal stress (x ) that is coinciding with the principal
stress (1 ) is used. The stress distribution in the front
367
of as well as around the weld toe and the stress gradient over the flange of the longitudinal stiffener are
analyzed. Figure 17 shows the effective notch surfaces
radius along the weld toe line along the axes TR and
TL (through the mid-points of cross sections).
4.1
Effective notch stresses resulting from external

pressure load
The effective notch stress distribution for different

cross sections along the weld toe line at HSA and
HSB for full loading condition can be seen in Figure 18 (a, b). It is found, from the FE analysis, that
the absolute values of stresses decrease from section
(a-a) to section (e-e). It can be also observed that the
stresses increase from the first points to mid-points and
then fall down rapidly to the end points at the sections
(a-a to c-c). The stress distribution is almost uniform at
the sections (d-d) and (e-e). However, the peak stress is
not the mid-point of the cross section (b-b) at HSA and
HSB. This is explained with the shape of the elements
at the corner of the weld toe. There are the similar stress
distributions at HSA and HSB for ballast loading.
The notch stresses at HSA are always less than HSB
in the two loading cases as can be seen in Figure 19.
It is also shown that the effective notch stress distribution decreases from the section (a-a) to (e-e) along the
weld toe line. It has to be pointed out that the stresses
are nearly equal at the section (e-e) in all cases. The
distributions of the effective notch stresses are close
together at HSA in full loading and HSB in ballast
loading.
Figure 20a and 20b show the stress distributions
in the front of the hot spot points at 10 and 30 [mm]
away from the weld toe over the half of flange width
Figure 18b. Stress distribution, HSB, external pressure, full

loading.
Figure 17. Effective notch radius along the weld toe line TR
and TL.
Figure 19. Effective notch stress distributions, HSA & HSB,

external pressure.
Figure 18a. Stress distribution, HSA, external pressure,

full loading.
Figure 20a. Stress gradient over the half of flange width for
external pressure in full loading condition.
368
for external pressure in full and ballast loading conditions, respectively. The stress gradients at the weld
toe and at the heel of web-stiffener are almost similar
in the two loading cases, but the stress level is higher
in the full load condition. Close to HSA and HSB at
10 [mm], can be observed that the stresses increase
significantly towards the web of the stiffener (from
z = 0 to z = 90 [mm]), whereas the stresses increase
slightly at the distance of 30 [mm] in the two loading
conditions.
Figure 21 shows the longitudinal stress distribution on the flange of longitudinal stiffener for external pressure in full and ballast loading conditions.
It is found that the distributions of longitudinal stresses
are almost the same in the two loading cases. The
structural hot spot stresses are also determined from
the FE shell model according to the recommendation
presented by DnV (2005) and Hobbacher (2007).
The result of effective notch, structural hot spot and
nominal stress at two hot spot points in the two loading conditions are summarized in Table 3. It can be
found from the hot spot stress analysis, the elements
size has an influence on the FE results as shown in
this table. The difference is about 3% and 8% between
coarse and fine meshes at HSA and HSB, respectively. It can be also observed that the stresses increase
significantly if the effective notch stress is applied.
4.2 Effective notch stresses resulting from internal
pressure load
Figure 20b. Stress gradient over the half of flange width for
external pressure in ballast loading condition.
The local model is only subjected to internal pressure acting on outer side in ballast loading condition.
Figure 22 (a, b) shows the effective notch stresses
distribution of different cross-sections along the weld
toe line in ballast loading condition at HSA and
HSB respectively. It can be also observed that the
Figure 21. Longitudinal stress distributions on the flange

of the longitudinal stiffener subjected to external pressure in
full and ballast loading condition.
Figure 22a. Stress distribution, HSA, for internal pressure

in ballast loading condition.
Table 3.
Stresses at HSA, HSB subjected to external pressure in full and ballast loading conditions.
hs
[MPa]
FEA (Shell)
[MPa]
notch
Khs = hs /normal
Coarse
Fine
FEA
(Solid)
Coarse
Fine
Kn = notch /normal
Full loading
A
36.8
B
36.8
49.1
62.9
50.7
68.9
300.1
412.2
1.34
1.71
1.38
1.88
8.16
11.2
Ballast loading
A
30.3
B
30.3
34.5
44.3
35.6
48.5
211.0
289.7
1.14
1.46
1.18
1.60
6.96
9.56
HS
normal [MPa]
Taken the absolute value.
369
Figure 22b. Stress distribution, HSB, for internal pressure

in ballast loading condition.
Figure 24. Stress gradient over the half of the flange width
for internal pressure in ballast loading condition.
Figure 25. Longitudinal stress distributions on the flange

of the longitudinal stiffener subjected to internal pressure in
ballast loading condition.
Figure 23. Effective notch stress distributions along the

weld toe line at HSA and HSB for internal pressure in ballast
loading condition.
distributions of stress for internal pressure are almost

similar to the distributions for external pressure.
Figure 23 clearly shows that the effective notch
stress distribution decrease from the section (a-a) to
(e-e) along the weld toe line exposed to internal pressure in ballast load condition. It should be noted that
the notch stress at HSA is smaller than the one at HSB
and nearly the same at the section (e-e).
Stress gradient along the half of flange width in the
front of the hot spot points at 10 and 30 [mm] away
from the weld toe, for internal pressure in ballast load
condition, can be seen in Figure 24. It can be also
observed that the stresses increase towards the web of
stiffener similar to the external pressure case. However, close to HSA at 30 [mm], the stress distributions
seem to be uniform along the flange.
The longitudinal stress distribution on the flange
of the longitudinal stiffener exposed to internal pressure in ballast loading condition is shown in Figure 25.
The effective notch stresses increase significantly in
comparison with the stresses calculated based on the
simplified approach, as can be seen in Table 4.
4.3
for full loading condition is presented in Figure 26 (a,

b). It seems that the stress at the first part of the section
(b-b) is higher than the one at section (a-a), However,
the distribution of stresses is changed from the midpoint to the end point at this cross-section. Similar
results can be found for ballast loading condition.
Figure 27 shows the effective notch stress distribution at HSA, HSB along the axes TL of the weld toe for
axial load in the two loading cases as the distribution
for external and internal pressure loads in Figure 18
and 23 respectively.
It can be observed from Figure 28 (a, b) that the
stress distribution over the half of the flange width for
axial load in the two loading conditions have a pronounced tendency to increase significantly towards the
web of the stiffener. Stress gradient in ballast loading
condition is steeper than in full loading condition. It
is also found that the distributions of stresses close
to HSB at 10 [mm] are always highest in all loading
conditions including the distribution for external and
internal pressure loading cases (see Figure 20 and 24).
The effective notch stress, nominal stress calculated
by FE and simplified approach are listed in the Table 5.
FATIGUE DAMAGE ASSESSMENT
The fatigue life assessment is based on the S-N

approach. The design S-N curve is represented by DnV
(2005), Hobbacher (2007), IACS (2010):
Effective notch stresses resulting from

axial load
The effective notch stress distributions for different

cross sections along the weld toe line at HSA and HSB
370
Table 4.
Stresses at HSA and HSB, internal pressure in ballast loading condition.
[MPa]
hs
Loading
HS
normal [MPa]
Coarse
Fine
[MPa]
notch
FEA
(Solid)
Ballast
A
B
30.3
30.3
21.7
28.0
22.6
31.0
135.8
187.5
FEA (Shell)
Figure 26a. Stress distribution, HSA, axial load in full

loading condition.
Figure 26b. Stress distribution, HSA, axial load in full

loading condition.
where N is predicted number of cycles to failure

for a given stress range, , m is the negative inverse
slope of the S-N curve, K2 is the constant depending
on material property, weld joint and log K2 is the
intercept of log N-axis of the S-N curve.
For the simplified notch stress approach, the S-N
curve parameters are for the one-slope in air curve Ib,
m = 3 and log K2 = 12.76 (DnV, 2005).
According to the IIW recommendation (Hobbacher,
2007), the FAT 225 S-N curve is used for the effective
notch stress approach and for plate thickness 5 [mm]
with the curve slope m = 3 and log K2 = 13.36.
For ocean structures the probability density function of long-term stress range is represented by the
two-parameter Weibull distribution, and the fatigue
Khs = hs /normal
Coarse
Fine
Kn = notch /normal
0.72
0.92
0.74
1.02
4.48
6.19
Figure 27. Effective notch stresses along the weld toe line
at HSA and HSB for axial load.
Figure 28a. Stress gradient along the half of flange width

for axial force in full loading condition.
Figure 28b. Stress gradient along the half of flange width

for axial force in ballast loading condition.
371
Table 5. Stress at HSA and HSB subjected to axial load in

full and ballast loading conditions.
Kn = notch /normal
Full loading
A
36.8
B
36.8
165.3
179.8
4.50
4.89
Ballast loading
A
30.3
B
30.3
136.9
148.9
4.52
4.91
normal [MPa]
Table 6.
and B.
Fatigue damage and life.

Damage
notch [MPa]
FEA (Solid)
HS
Table 7.
Total
Life
[years]
Simplified notch stress approach (SNS)

(1)
A
0.808
0.458
B
0.653
0.377
(2)
A
0.601
0.345
B
0.856
0.416
(3)
A
0.630
0.358
B
1.019
0.466
1.266
1.030
0.946
1.272
0.988
1.485
15.20
18.67
20.32
15.12
19.46
12.95
Effective notch stress approach (ENS)

A
1.278
0.514
B
2.595
0.936
1.792
3.531
10.73
5.45
Hot spot
Long-term stress range descriptors at points, A
(1)
qn [MPa]
(2)
Load case
pn
hn
HSA
HSB
Full
Ballast
0.45
0.40
0.819
0.900
20.77
20.82
19.35
19.52
(3)
Full
Ballast
SCF is taken from the list of standard details, (DnV, 2005).

SCF, coarse mesh.
SCF, fine mesh.
damage ratio based on Palmgren-Miner approach may

be calculated as (Nolte and Hansford, 1976):
where Nload is the total number of the considered loading conditions, o is the long-term average response
zero-crossing frequency, Td is the ship service life,
is the Gamma function, pn is a part of the lifetime
spent in each of loading condition and hn , qn are the
Weibull stress range shape and scale factor for the nth
loading condition, respectively. The long-term stress
range distribution at the two hot spots, considered for
full and ballast loading condition is shown in Table 6.
The total fatigue damage, during the design service
life of 25 years, is found by summing the part damage
from each of the loading, and increased the damage by
a factor related to the corrosive environment.
Fatigue damage and fatigue life for the two hot spot
points A and B are given in Table 7. For the simplified notch stress approach, the differences of SCF due
to geometry effect are also listed in Table 7. It can
be seen from Figure 29 that the total fatigue damage obtained from the simplified approach is smaller
than the one obtained from the effective notch stress
approach which is based on the FE analysis. It is also
found from the effective notch stress approach that the
fatigue damage is bigger at point B.
It should be noted, as observed in Figure 29, that
for the simplified notch stress approach, the SCFs due
to geometry effect, which are analyzed by coarse and
fine mesh in FE analysis, comparing with the SCFs
in the standard list details (DnV, 2005) have an influence on the fatigue damage. At the hot spot point A,
the fatigue damage obtained from the coarse and fine
Figure 29. Fatigue damage.
mesh is smaller than the one from SCF of DnV (2005).

Moreover, the results of the damage are nearly equal
between coarse and fine mesh. At the hot spot point B,
the fatigue damage of the fine mesh is higher than the
two remaining cases. It is also found that the fatigue
damage at point B is smaller than at point A.
CONCLUSION
Fatigue assessment of a side-longitudinal stiffener

considering two most probable crack initiation points
in a double hull oil tanker based on the three stress
approaches has been done here. The analyses are performed for two basic loading conditions accounting for
the expected operation time in each of the considered
conditions. The finite element models were composed
by shell and solid elements using ANSYS software
and applying advanced sub-modeling techniques.
The stress distributions at the notch surface of weld
toe of the two critical points were analyzed. The FE
analysis for the effective notch stress approach showed
that the stress at hot spot B is always highest in all the
loading cases. The simplified notch stress approach
372
was applied and compared with the effective notch

stress approach for the fatigue damage assessment.
In this study, significant differences between the
three stress approaches were found with respect to
calculated fatigue damage. The total fatigue damage,
based on the effective notch stress approach, is higher
than the one from the simplified approach. It has been
found that the fatigue damage at point B is higher than
the one at point A.
Considering that the design service life of ship is
25 years, the predicted fatigue life indicated clear
difference between the simplified and the effective
notch stress approaches.
ACKNOWLEDGEMENTS
The first author has been financed by a PhD scholarship given by ABS, American Bureau of Shipping to
IST. The authors are indebted to Dr G Wang for having
promoted such scholarship.
REFERENCES
ANSYS, 2009. Online Manuals, Release 11.
Chakarov, K., Garbatov,Y., Guedes Soares, C., 2008a. Fatigue
analysis of ship deck structure accounting for imperfections. International Journal of Fatigue 30, pp. 18811897.
Chakarov, K., Garbatov, Y., Guedes Soares, C., 2008b. Hot
Spot Stress and Stress Concentration Factors due to
Different Fabrication Imperfections in Deck Structures.
International Shipbuilding Progress 55, pp. 4762.
DnV, 2005. Fatigue assessment of ship structures, Classification notes 30.7. Det Norske Veritas, Hovik, Norway.
Fricke, W., Codda, M., Feltz, O., Garbatov, Y., Remes, H.,
Risso, G., Rizzo, C., Romanoff, J., 2012. Round-Robin
on Local Stress Determination and Fatigue Assessment of
Load-Carrying Fillet-Welded Joints. Ships and Offshore
Structures.
Fricke, W., Cui, W., Kierkegaard, H., Kihl, D., Koval, M.,
Mikkola, T., Parmentier, G., Toyosada, M., Yoon, J., 2002.
Comparative fatigue strength assessment of a structural
detail in a containership using various approaches of
classification societies. Marine Structures 15, pp. 113.
Fricke, W., Kahl, A., 2005. Comparison of different structural
stress approaches for fatigue assessment of welded ship
structures. Marine Structures 18, pp. 473488.
Garbatov, Y., Guedes Soares, C., 1998. Fatigue Reliability of
Maintained Welded Joints in the Side Shell of Tankers.
Journal of Offshore Mechanics and Arctic Engineering
120, pp. 29.
Garbatov, Y., Guedes Soares, C., 2002. Bayesian updating in the reliability assessment of maintained floating
structures. Journal of Offshore Mechanics and Arctic
Gaspar, B., Garbatov, Y., Guedes Soares, C., 2011. Effect
of weld shape imperfections on the structural hot-spot
stress distribution. Ships and Offshore Structures 6,
pp. 145159.
Hobbacher, A., 2007. A recommendations for fatigue design
of welded joints and components, International Institute
of Welding.
IACS, 2010. Common Structure Rules for Double Hull Oil
Tankers, Consolidated version, July 2010.
Nolte, K., Hansford, J., 1976. Closed-form expressions for
determining the fatigue damage of structures due to ocean
waves, Proceedings Offshore Technology Conference,
pp. 861870.
Radaj, D., Sonsino, C.M., Fricke, W., 2009. Recent developments in local concepts of fatigue assessment of welded
joints. International Journal of Fatigue 31, pp. 211.
Saad-Eldeen, S., Nguyen, T., Garbatov,Y., Guedes Soares, C.,
2010. Fatigue Stress Assessment Accounting for Contact
Element, Proceedings of the International Conference on
Marine Science and Technology. Union of Scientists of
Varna, Varna, Bulgaria.
Storhaug, S., Berstad, A.J., 2001. Fatigue in the side shell
of ship shaped structures, Proceeding of the 20th International Conference on Offshore Mechanics and Arctic
Engineering, Rio de Jeneiro, Brazil.
373

Strength of a tanker double bottom in various grounding scenarios

R. Villavicencio
J. Amdahl
Marine Technology Department, Norwegian University of Technology and Science,
Trondheim, Norway
C. Guedes Soares
ABSTRACT: The paper presents finite element simulations of a damaged tanker double bottom subjected to
bending moment. An important issue is the influence of size and shape of the obstruction on the sea floor during
minor grounding events. The initial damage is induced by rigid indenters at various locations and using different
sea bottom topologies. The study found an important reduction in the ultimate capacity of the double bottom
structure when the initial indentation deforms stiff structural elements such as side girders, whereas more local
damage produces similar shortening response independently of the location or shape of the indenter.
INTRODUCTION
In the design of the midship section it is necessary

to know the ultimate longitudinal hull girder strength.
The ship structure is exposed to different types of loads
related to extreme environmental conditions or accidental events such as collision and grounding. In case
of structural damage, due to collision or grounding,
the hull girder strength may be reduced compared
with the intact condition. If a fracture occurs, the
oil spill will be the most severe consequence. Thus,
improvements in the design of tanker double bottoms
are essential to obtain a high level of safety at sea and
protection of marine ecosystems.
The ultimate hull girder capacity of a ship can
be estimated by simplified analytical expressions or
by fully nonlinear finite element analysis. Simplified
structural analysis methods are widely accepted as reliable and fast tools to obtain the longitudinal strength
of an intact ship (e.g. Smith 1977; Yao and Nikolov
1991, 1992; Gordo et al. 1996). In minor ship collision or grounding scenarios the damaged panel could
contribute to the hull girder residual strength. Thus,
the neglect of the damaged structure in the estimation of the residual strength is not a straightforward
decision. Guedes Soares et al. (2008b) benchmarked
the ability of simplified structural methods based
on the Smiths formulation (Smith 1977) to predict
the ultimate strength of a damaged ship and compared the results with finite elements analysis, finding
good agreement for intact and damaged conditions.
However, due to the complexity of the problem, it is

recommended to study the collapse response of ship
hulls by numerical procedures. On the other hand, the
nonlinear finite element analysis of the hull girder
requires huge computational sources. Thus, simplified finite element models of double bottom structures
can be used to analyze its resistance capacity in the first
stage of design.
Several analyses of the intact hull girder capacity
subjected to bending moment have been conducted.
For example, Bai et al. (1993) presented a numerical procedure for ship hull under complicated loads
where a set of finite elements were defined accounting
geometrical and material nonlinearities. Gordo et al.
(1996) presented a simplified method to represent the
behaviour of stiffened plate columns allowing the prediction of the strength degradation due to corrosion
and residual stresses. Qi et al. (2005) studied theoretical methods for large tankers which were based on
experimental works comparing the results with various
finite element and theoretical methods and also with
the common structural rules for double hull tankers.
Luis et al. (2007) studied the effect of bottom and side
damage on the ultimate strength of double hull tankers.
Amlashi and Moan (2008) summarized the definitions
for the nonlinear finite element modelling to estimate
the ultimate hull girder capacity of a bulk carrier under
alternate hold loading condition.
Some studies have been conducted on stiffened panels and hull girder capacity after denting damage such
as collision or grounding. For example, Smith and Dow
375
(1981) reviewed various forms of structural damage

on ships and offshore structures, suggesting analytical methods for evaluation of the residual stiffness
and strength. Pedersen (1994) presented a mathematical model for analysis of ship hull loading due to
grounding where the grounding forces, the sectional
shear forces and the bending moments were determined and related to the ultimate capacity of the
hull girder. Wang and Chen (2002) derived analytical
equations for the residual strength of ships with damages due to grounding or collision which were verified
with direct calculations of sample commercial ships
for a broad spectrum of accidents. A recent study of
similar type of stiffened panels under localized damage
was presented by Guedes Soares et al. (2008a).
The finite element analysis of damaged double bottoms has been focused on grounding scenarios where
the structure is penetrated until fracture in the inner
plate (Naar et al. 2002; Alsos and Amdahl 2007;
etc.). After such a major damage, the residual strength
of the double bottom loses importance. However, in
minor grounding scenarios, the ultimate capacity of
the double bottom structure to sustain compressive
loads plays an important role to ensure the safety at
sea and avoid environmental catastrophes.
Being interested in the response of a damaged
double bottom structure, Liu (2011) presented numerical simulations and simplified methods of estimation of the residual strength. Following their work,
Villavicencio et al. (2011) proposed a method for
estimating the displacement of the neutral axis of a
damaged ship subjected to bending moment. They
used the load-end shortening relationship and established force equilibrium over the whole transverse
section of the ship.
The present study is a continuation of the previous
numerical investigations (Liu 2011;Villavicencio et al.
2011). It aims at estimating the associated residual
strength in intact and minor damaged double bottom
structures. In the previous works, the double bottom was indented in one location and the width of
the damage was concentrated between the side girders affecting only the bottom platting and stiffeners
while the other structural elements remain intact. Here,
various minor grounding scenarios are numerically
simulated in order to understand the influence of the
extent of damage and location of the obstruction on
the sea floor during minor grounding incidents.
2
DOUBLE BOTTOM IN STUDY
The double bottom used in the present work was

previously analyzed (Liu 2011) performing numerical simulations yielding the residual strength in the
intact and damaged conditions. In the present Chapter,
the finite element model and some of the results are
summarized to understand the actual calculations.
The structure of the double bottom corresponds to a
Shuttle Tanker 140.000 m3 . The main dimensions of
the ship are: length between perpendiculars 256.5 m,
Figure 1. Scantling of the double bottom structure (FE

model).
Figure 2. Finite element model (fore floor is hidden).
moulded breadth 42.5 m, depth to upper deck 22.0 m

and design draft 15.0 m. The double bottom was modelled between four floors located near the midship
section of the ship. Only the starboard side was considered in the analysis. The main dimensions of the double
bottom model are: length 12.0 m, width 21.25 m and
height 2.65 m. The space between frames is 4.0 m.
Figure 1 represents the scantling used in the finite
element model, which was simplified with respect to
the original design and Figure 2 shows the isometric
view of the finite element model.
The computations were carried out using the finite
element package LS-DYNA Version 971 (Hallquist
2005). The plating and longitudinals were modelled
with 4-node shell elements with 5-integration points.
The mesh size was approximately 100 mm. The magnitude of the defined initial imperfections corresponded
to fabrication tolerances and the shape was defined
using sinusoidal expressions.
The true stress-strain relationship of the material
was defined by an isotropic plasticity model using
a power law hardening rule (Mat.18 Power Law
Plasticity in LS-DYNA). The inputs of the material were: yield stress 275 MPa, strength coefficient
740 and hardening exponent 0.24; strain rate effects
were ignored. Since only large plastic deformations
were studied, failure was not required in the material
definition.
Symmetry boundary conditions were defined at
the center line and the fore and aft floors were fully
clamped. Both boundary conditions were properly
modified to induce a bending moment. As the tanker
was considered in full load condition, water and oil
pressure were applied on the bottom and tank top,
respectively. In the intact case the sequence of the
loads was pressure and a bending moment whereas in
the damaged case the sequence was pressure, indentation damage and a bending moment. In both intact
and damaged cases the bending moment was induced
376
Figure 3. Ultimate strength of double bottom. Previous

intact and damaged cases.
by defining prescribed displacements and rotations on

the fore and aft floors.
The hogging bending moment applied on the double bottom structure induced axial strains and stresses
which were represented by mean strains and resistance forces. The mean strain was obtained by
the ratio between the longitudinal displacement of a
node located in the central fore floor and the span of
the stiffeners. The resistance forces were obtained
from elements located forward the central aft floor. It
must be noted that the resistance forces were given for
only one side (starboard) of the double bottom, which
represented all the thickness of the centre vertical keel.
Figure 3 shows the resistance force-mean strain
relationships of the intact and damaged cases. The
total resistance force is a summation of the resistance force of the structural elements located below
and above the local neutral axis of the double bottom
(1340 at BL; BL is the base line) denoted in this study
by outer and inner elements, respectively. Due to
the rotation applied on the fore and aft floor to induce a
bending moment, the inner and outer elements buckled
at different induced strains, see Fig. 3.
The damage on the bottom panel reduced its capacity to sustain a compressive load. This reduction is
more evident in the outer than in the inner elements.
The post-buckling response of the inner elements is
less influenced by the indentation of the central bottom panel. The intact ultimate strength of the double
bottom occurred when the outer elements buckled,
whereas in the damaged case it was observed at the
buckling of the inner elements.
3
GROUNDING SCENARIOS
Ship grounding is a complex process which involves

large contact forces, crushing of bottom structure
and shell plating fracture, affecting the global ship
motions.Although the sea bed profile is one of the governing factors within the damage process, information
on sea floor topology is very limited. In general, the
finite element analysis adopts indenter shape based on
information from grounding accidents.
Studies show that small indenters perforate the

bottom plating with relative ease while larger indenters
damage floors and side girders before the plating ruptures (Wang et al. 2000). In other words, the difference
in structural damage can be attributed to the shape and
size of the obstruction on the sea floor. Based on the
extent of damage, three indenter topologies have been
defined (Alsos and Amdahl 2007).
Indenters much smaller than the ship itself which
create local damage while the overall structure remain
intact, are called rock. When the surface of the
obstruction is relatively large (about half the ship
breadth) and wide parts of the bottom are damaged
the indenter is denoted by shoal. The intermediate
obstruction is defined as reef and in this case is
not easy to see in advance the response of the double
bottom in terms of local or global deformation.
In order to study the double bottom, four groups
of minor grounding scenarios were numerically simulated allowing to understand the influence of the shape
and size of the obstruction on the sea floor. After the
induced damage, the double bottom is subjected to
bending moment, in order to obtain the reduction in its
ultimate resistance during minor grounding incidents.
Figure 4 illustrates the groups of grounding scenarios. In all cases the double bottom is indented between
the fore and aft central floors, as shown in the profile
and plan views of the ship. The size and location of the
indenters are sketched in the body views. The dimension and shape of the indenters are shown in details
in Fig. 5, and a brief description of the grounding
incidents is given in the following paragraphs.
Group A: The triangular indenters cause damage
to three stiffeners. The bottom is indented between the
side girders causing damage only to the bottom plating
and longitudinal while the side girders remain intact.
This type of contact is similar to the local indentation
provoked by a rock.
Group B: The triangular indenters cause damage
to seven stiffeners. Location 1 implies crushing of the
side girder (second) near the center vertical keel; Location 2 represents the contact between the second and
third side girder and Location 3 considers grounding
of the side girder (third) away from the center vertical
keel.
Group C: The flat indenters cause damage to 7, 13
and 19 stiffeners in Locations 1, 2 and 3, respectively.
Locations 2 and 3 represent the most severe damage
crushing the second and the third side girders at the
same time.
Group D: The indenters cause damage to seven
stiffeners and the contact is between the second
and the third side girder. The influence of the indenter shape is studied defining flat, round and square
conditions (Fig. 5).
Some of the grounding incidents are shown in Figure 6 representing the double bottom deformation and
the Von Mises stress distribution due to indentation
damage. It is observed that small indenters cause local
damage on the mid-span of the stiffeners (Fig. 6a). On
the other hand, wider indenters cause local damage
377
Figure 4. Grounding scenarios.
Figure 5. Shape and design of the indenters. The number of damaged indenters is denoted by type 3, 7, 9, 19.
on the lower edges of the side girders while the stress

distribution on the stiffeners is more uniform.
The focus of this work is purely on the double bottom capacity to sustain bending moment action during
minor grounding scenarios. The resulting resistance
force-mean strain relationships are shown in Fig. 7.
Group A (Fig. 7a): As the indenters are located
between the side girders, the contact is carried only
through a small part of the bottom plating and longitudinal. Consequently, the difference in the double
bottom resistant capacity appears to be insignificant
when considering local damage between side girders independently of the contact location, i.e. in all
cases, the double bottom shows almost the same

end-shortening relationships.
Group B (Fig. 7b): Contrary to the local penetration of the previous group, here the indenters cause
wider bottom damage. The double bottom resistance
is similar when structural elements between side girders are damaged being imperceptible the effect of the
width of the indenter (number of damaged stiffeners).
For example, Group A-Location 2 and Group BLocation 2 show similar ultimate resistance. On the
other hand, when the damage is imposed on the side
girders (Location 2 or 3) an important reduction in the
double bottom capacity is noticed, reproducing similar
378
Figure 6. Double bottom deformation and Von Mises stress distribution due to indentation. (a): Group A, Location 2.
(b): Group B, Location 1. (c): Group C, Location 3. (d): Group D, Location 3.
Figure 7. Ultimate strength of double bottom at various grounding scenarios. (a): Group A. (b): Group B. (c) Group C.
(d): Group D. (1), (2) and (3) denotes locations 1, 2 and 3, respectively.
shortening responses in both locations. It is important

to mention that the damage to the side girders reduces
also the capacity of the inner structural elements.
The indentation in the side girders captures the intersection point between the local damage and the global
double bottom resistance. Moreover, it adds an aspect
to possible crashworthiness design. For example, in
order to reduce initial deformations of the side girders
in minor grounding incidents, the inclusion of bottom
brackets (Fig. 9) could be adopted.
Group C (Fig. 7c): The modified shape of the indenter covers 3/8 of the stiffeners span at the first instant
of impact (Fig. 5). The effect of this relatively more
severe damage to the stiffeners is insignificant using

the same width and location of the indenter.
A very important reduction in the double bottom
capacity is observed when the width of the indenter is increased, causing severe bottom damage and
deforming the side girders.
Certainly, the most severe effect is produced by
indenter type 19-flat due to its wider structural
damage. This indentation deformed most of the bottom structural components reducing their contribution
prior to inducing bending moments. This is observed
in the stiffness response of the end-shortening curves,
showing that once the both side girders collapse the
379
Figure 8. Bottom brackets.
double bottom resistance falls dramatically, which is

noticed in the outer and also in the inner elements.
Group D (Fig. 7d): It was mentioned that in relatively local damage between side girders the influence
of the size and shape of the indenter is insignificant.
The ultimate resistance depends on the area of provoked damage to the stiffeners during the grounding
incident. For example, indenter type 7 (triangular
shape) causes damage mainly at the mid-span which
is relatively similar to the one induced by indenters type 7-flat and type 7-round. However, when
the indentation damages exceed the span length, the
double bottom capacity is considerably reduced, as
seen with indenter type 7-square, which cause severe
damage to of the stiffenersspan. This damage effect
does not decrease the resistance of the inner elements.
4
CONCLUSIONS
The influence of the shape and size of the obstruction

on the sea floor is reproduced in various minor grounding scenarios. The residual strength of the damaged
double bottom shows that independently of the contact location or shape of the indenter, the relatively
local damage between the side girders has insignificant
effect on the strength of the double bottom in sustain
bending effects. On the other hand, initial large deformations of secondary structural elements such as side
girders must reduce severely the global hull bending
strength during minor grounding incidents.
ACKNOWLEDGEMENTS
FLAGSHIP (European Framework for Safe, Efficient
and Environmentally-friendly Ship Operations) that
is partially financed by the European Union through
contract TIP5-CT-2006-031406.
REFERENCES
Alsos HS, Amdahl J. 2007. On the resistance of tanker bottom structures during stranding. Marine Structures; 20:
218237.
Amlashi H, MoanT. 2008. Ultimate strength analysis of a bulk

carrier hull girder under alternate hold loading condition
A case study Part 1: Nonlinear finite element modelling
and ultimate hull girder capacity. Marine Structures; 21:
327352.
Bai Y, Bendiksen E, Pedersen PT. 1993. Collapse analysis of
ship hulls. Marine Structures; 6: 485507.
Gordo JM, Guedes Soares C, Faulkner D. 1996. Approximate
assessment of the ultimate longitudinal strength of the hull
girder. Journal of Ship Research; 4 (1): 6069.
Guedes Soares C, Lus RM, Teixeira AP, Quesnel T,
Nikolov PI, Steen E, Khan IA, Toderan C, Olaru VD,
Bollero A, Taczala M. 2008a. Parametric study on the
collapse strength of rectangular plates with localized
imperfections under inplane compression. International
Shipbuilding Progress; 55: 6385.
Guedes Soares C, Lus RM, Nikolov P, Downes J, Taczala M,
Modiga M, Quesnel T, Toderan C, Samuelides M. 2008b.
Benchmark study on the use of simplified structural codes
to predict the ultimate strength of a damaged ship hull.
International Shipbuilding Progress; 55: 87107.
Hallquist JO. 2005. LS-DYNA Keyword users manual,
Version 971. Livermore Software Technology Corporation, Livermore, California.
Liu Z. 2011. Analytical and numerical analysis of iceberg
collision with ship structures. Ph.D. Thesis, Department
of Marine Technology, Norwegian University of Science
and Technology.
Lus RM, Hussein AW, Guedes Soares C. 2007. On the effect
of damage to the ultimate longitudinal strength of double
hull tankers. Proc. 10th Int. Symp. on Practical Design
of Ships and Other Floating Structures (PRADS07);
ABS, Houston, Texas, USA.
Naar H, Kujala P, Simonsen BC, Ludolphy H. 2002. Comparison of the crashworthiness of various bottom and side
structures. Marine Structures; 15: 443460.
Pedersen P. 1994. Ship grounding and hull-girder strength.
Marine Structures; 7: 129.
Qi E, Cui W, Wan Z. 2005. Comparative study of ultimate
hull girder strength of large double hull tankers. Marine
Structures; 18: 227249.
Smith CS, Dow RS. 1981. Residual strength of damaged
steel ship structures and offshore platforms. Journal of
Constructional Steel Research; 1 (4): 215.
Smith CS. 1977. Influence of local compressive failure
on ultimate longitudinal strength of a ships hull. In:
Proc. Int. Symposium on Practical Design in Shipbuilding
(PRADS); N. Zosengakkai ed., Society of NavalArchitects
of Japan, Tokyo, Japan, 1820 October 1977, pp. 7379.
Villavicencio R, Liu Z, Amdahl J, Guedes Soares C. 2011.
Influence of the neutral axis displacement on the residual
strength of a damaged tanker double bottom structure.
Advances in Marine Structures; Guedes Soares & Fricke
(eds). Taylor & Francis Group: London. pp. 411418.
Wang G, Arita K, Liu D. 2000. Behavior of a double hull
in a variety of stranding or collision scenarios. Marine
Structures; 15: 443460.
Wang G, Chen Y. 2002. Longitudinal strength of ships with
accidental damages. Marine Structures; 15 (2): 119138.
Yao T, Nikolov PI. 1991. Progressive collapse analysis of a
ships hull under longitudinal bending. J. Soc. Naval Arch.
of Japan; 170: 449461.
Yao T, Nikolov PI. 1992. Progressive collapse analysis of a
ships hull under longitudinal bending (2nd report). J. Soc.
Naval Arch. of Japan; 172: 437446.
380

Review of material relations for marine structures collision simulations

R. Villavicencio & C. Guedes Soares
ABSTRACT: The finite element method is widely used to simulate marine structures subjected to lateral
impact. These simulations need to define the nonlinear material behaviour and include an appropriate failure
criterion to account fracture within the structure. However, the influence of the true stress-strain relationship
on the results is not presented in detail in some cases. Therefore, this paper reviews mathematical expressions
for the flow curve and the failure criteria, commonly used in collision simulations. Additionally, a procedure
is summarized to conduct numerical simulations of tension tests, as a practical method to evaluate the material
relation and the critical failure strain of the elements. Moreover, representative simulations of marine structures
under the action of impact loads are reviewed. In these references, the emphasis is put on the influence of the
material relation and the failure criterion on the numerical predictions.
INTRODUCTION
Finite element simulations of ship and offshore structures subjected to lateral impact are performed to
evaluate preliminary designs. These simulations need
to define the nonlinear behaviour of the material
including strain hardening and failure. Commonly,
the mechanical properties of the material are determined with tensile tests. Hence, the true stress-strain
relationships are obtained from the recorded engineering stress-strain state in the form of a power relation
or using instantaneous measurements of the crosssectional area. However, the material relationship
adopted, affects the necking and rupture behaviour
obtained by the finite element simulations. Moreover,
the failure due to material rupture is still not well
resolved numerically, because the fracture length is
much smaller than the side length of the elements in
a finite element model. Thus, it is difficult to establish a procedure suitable for prediction of failure in the
engineering practice.
Since experimental tests on full-scale prototypes are
extremely expensive and thus rarely performed, it is
necessary to verify experimental-numerical models of
the dynamic large deformation of small-scale structural elements, in order to predict the overall response
to lateral collision of complex marine structures. Material relations and failure criteria have been derived
to predict the plastic response and the critical failure
strain in finite element models of small-scale marine
structures subjected to lateral impact. Typically, this
is done through metal-forming tests and simulations.
As a result, the accuracy of the simulations depends
on the material relation itself. A true stress-strain
curve is affected by the manufacturing history, metallurgical treatments, and chemical composition of the
material. Therefore, the metal-forming simulations

require flow curves that reflect the special conditions
of the material.
Although significant research is conducted to
describe the criteria that determine the failure strain in
a collision simulation, in most of these references the
impact response is examined by penetrating the specimens using quasi-static lateral loads, which do not
consider the extremely high impact forces or vibrations
developed in a real collision scenario. Therefore, it is
necessary to develop material relations that account
these dynamic effects. However, considerable uncertainty remains as regards the magnitude of the material
dynamic variables.This is explained by the insufficient
data describing the behaviour of the materials under
dynamic biaxial loads and the influence of the generalized stresses, such as bending moments and membrane
forces on the response of structures subjected to impact
loads.
This paper gives an insight into the influence of the
material relation on the accuracy of collision simulations. Therefore, first are reviewed the mathematical
expressions for the flow curve and the failure criteria
that have been adopted in marine structures collision
simulations. Additionally, a procedure is summarized
to conduct simulations of tension tests as a practical
method to evaluate the true stress-strain relationship
and the critical failure strain defined in the finite element simulation. Finally, the different contributions
arising from the material relations and failure criteria
are summarized for different finite element simulations of structures under the action of impact loads.
These definitions are of considerable practical importance when it is necessary to assess the safety of the
structural elements in offshore engineering and naval
architecture, since these calculations require accurate
381
estimates of the large deformations produced by the

impact.
power relation shows that the flow curve of many metals in the region of uniform plastic deformation can be
expressed as:
2 TRUE STRESS-STRAIN RELATIONSHIPS

The mechanical behaviour of materials is described
by their deformation and fracture characteristics under
applied tensile, compressive or multiaxial stresses.The
engineering tension test is widely used to provide basic
design information on the strength of materials (Dieter
1986, ASTM E8). However, the engineering stressstrain curve does not give a true indication of the
deformation characteristics of a metal, because it is
based entirely on the initial cross-sectional area A0
and initial gage length L0 of the tensile specimen. The
true stress t and true strain t are based on the actual
cross-sectional area of the specimen, thus this curve
increases continuously until fracture (strain hardening). The true stress and true strain are expressed in
terms of engineering stress e and engineering strain
e by:
These equations should be used only until the onset

of necking. Beyond maximum load, the true stress
and the true strain should be determined from actual
measurements of the load P and the cross-sectional
area A:
Since highly accurate measuring systems are needed

for obtaining actual measurements of the crosssectional area, is common practice the use of
mathematical expressions to define the plastic flow
characteristics of the material. Among the various
types of mathematical expressions for the true stressstrain relationship that have been proposed, this paper
reviews representative expressions that have been used
for collision simulations of marine structures.
One of the first material relations is the Swift hardening law (Eq. 5), which is reviewed in Choung et al.
(2010). In this expression 0 is the initial yield stress,
0 is the initial yield strain and n is the plastic hardening exponent. Choung et al. recommend n = 0.2, which
correspond to a 95.5% probability level of survival.
The derivation of this relation is relatively simple since
requires only the elastic limit of the material for its
definition.
The most common expression is the simple power

curve relation as described in Dieter (1986). The
where n is the strain hardening exponent and K is the

strength coefficient, determined from the engineering stress-strain curve. The exponent n is the linear
slope of a log-log plot of the logarithmic true stress
and true strain up to the maximum load, expressed
in terms of the engineering stress and engineering
strain (Eqs. 1 and 2). The coefficient K is the true
stress at t = 1.0 and corresponds to the reduced area
Af = 0.37A0 , where Af is the cross-sectional area at
failure. This material relation has been included in
most of the material libraries of the finite element
solvers (Hallquist 2010).
Since this relation is derived from experimental
data, it is common to observe deviations from the mentioned log-log plot, often at low strains. In particular,
this plot can results in two straight lines with different
slopes. Hence, it will be difficult to estimate accurately the strain hardening exponent. As the simple
power relation does not account for the yield point
elongation (or yield plateau) observed in most of the
low-carbon steels, this relation should be modified and
represented in the following form to account for the
yield plateau (Dieter 1986, Ehlers 2010a):
where m can be considered to be the amount of strain

that the material received prior to the tension test, 0
is the yield stress and plateau is the strain at plateau
exist. In Eq. (7) is assumed that t = 0 at t plateau .
Thus, the simple power law relation intersect with the
modified power relation at plateau and 0 .
Both power relations require the engineering curve
until maximum load for its definition. However, the
engineering curves of each particular material are not
available in the preliminary design of marine crashworthy structures. In general tensile tests to define
the material relation of laboratory specimens are
conducted only for research purposes, while in the
industry the mechanical properties of the material are
limited to those given by standards (ASTM A36). This
information generally includes minimum yield stress,
minimum and maximum ultimate tensile strength and
rupture strain. Unfortunately, the difference between
the minimum and the maximum ultimate stress is
about 30% (see for example the mechanical properties of carbon steel ASTM A36, which is widely used
for shipbuilding).
Zhang et al. (2004) presented a simplified power
relation in an approval procedure to estimate the critical deformation energy in case of side collision of
382
double hull cargo vessels. This relation can be derived

only with the ultimate tensile strength:
where Ag is the maximal uniform strain related to the

ultimate tensile stress Rm and e is the natural logarithmic constant. If only the ultimate stress Rm (MPa) is
available, the following approximation can be used to
obtain the proper Ag :
Although both Eqs. (6) and (9) represent a power relation, they differ in the derivation of their constants.
For consistence, the present paper keeps the notation
used in (Dieter 1986, Zhang et al. 2004). This simplified relation is widely used by structural engineers
to perform analyses of marine structures subjected to
impact loads since it is derived only with the ultimate tensile strength obtained from standards. This
is accepted since the use of true material relations
obtained from tensile tests carried out with the same
material from which laboratory specimens are made is
a convenient assumption for numerical crashworthy
structural analysis. In practice, however, the properties
of the material vary from one structural component
to another, so using a particular material relationship
could bring underestimation of the impact response in
a real collision.
Villavicencio and Guedes Soares (2012) proposed
a combined material relation in which the true stressstain curve is divided in two parts with respect to the
onset of necking. Thus, the logarithmic true stressstrain up to the maximum load (Eqs. 1 and 2) define
the process before and the simplified power relation (Eq. 9) after the necking is localized. The use
of Eq. (12) is recommended to derive the maximal
uniform strain related to the ultimate tensile stress
in order to obtain a better continuity of the logarithmic true stress-strain. This relation defines accurately
the plastic response of metals until maximum load,
in particular for low carbon materials since accounts
for the yield plateau, although this relation should not
be entirely correct after localization occurs. All the
mentioned true stress-strain relationships for a typical low-carbon steel material are plotted in Fig. 1 for
comparison.
Paik (2007) proposed a brand new approach called
the knock-down factor f (e ), which is used to transform the engineering stress-strain curve to the true
stress-strain relationship. However, to the authors
knowledge this material relation has not been implemented for collision simulations of marine structures.
In addition to the mathematical expressions that
define the flow characteristics of metals, accurate true
stress-strain relationships can be obtained using highly
Figure 1. Material relations. (a): (1) Swift hardening law

(2) Simple power curve relation (3) Modified power relation.
(b): (1) Simplified power relation, Rm = 400 MPa (2): Simplified power relation, Rm = 500 MPa (3) Combined material
relation.
accurate measuring systems, recording actual measurements of the cross-sectional area of the tensile
specimens. For example, Ehlers and Varsta (2009)
determined a true material relation until failure, using
optical measurements of the displacements on the
surface of tensile specimens. In that reference, the
strain is calculated from the measured displacements
on the basis of a discrete amount of pixel recordings
which define a strain reference length (see Fig. 2).
To determine the stress, the actual cross-sectional
area is calculated on the basis of the out-of-plane
displacement measurements of the tensile specimen.
Therefore, the stress is determined on the basis of
the minimum cross-sectional area of the specimen,
measured as a function of the strain reference length.
Furthermore, the reference length is a function of the
discrete amount of pixel recordings from the optical
measurements. As a result, the true strain and stress
relation until failure is obtained in a manner that is
dependent on the choice of strain reference length.
This material is referred as strain reference length
based material relation (Ehlers 2010).
A similar procedure to obtain true material relationships is reported in Hogstrm et al. (2009). They examined the stress-strain behaviour of uniaxial tensile tests
and calculated true stress-strain relationships on different length scales across the necking region. The tensile
383
Figure 2. Stochastic pattern of the surface of the specimen with a facet of 15 15 pixels (Ehlers and Varsta 2009).
Symbolic element length-dependent true strain and stress
relation until fracture for numerical collision simulations
(Ehlers 2010).
test are recorded with a optical strain measuring system in order to validate existing failure models used
in finite element simulations, in terms of dependence
on length scale and strain state. Although these experimental methods are very accurate, they lack practical
application in the marine industry since they require
overly expensive instruments; however, further experimental measurements could derive more accurate
mathematical expressions for the flow curve.
The finite element method itself is a strong tool to
derive new flow characteristics of metals as demonstrated Joun et al. (2008). They describe a method to
define material relations at large strains, using tensile
test results coupled with a numerical analysis. A reference true stress-strain curve is modified iteratively
to minimize the difference in the tensile force. The
method is verified by comparing tensile test results
with finite element solutions using the modified true
stress-strain curve. In contrast with the detailed measurements performed in (Ehlers and Varsta 2009,
Hogstrm et al. 2009), this approach can predict the
flow stress at large strains, using only the measured
load-elongation curve of a material and a numerical
tensile test analysis, which yield exact results from an
engineering viewpoint.
FAILURE CRITERIA FOR COLLISION

SIMULATIONS
The material models themselves follow a semi implicit

iteration scheme for an isotropic plane stress material.
Plasticity in these materials is by definition independent of the volumetric component and allows for
modelling of fracture. However, the critical failure
strain at which fracture occurs is affected by the material relation and the mesh size (a coarse mesh requires
a lower value of critical failure strain). As in the previous chapter, here the common failure criteria used in
collision simulations are summarized.
The most popular method of introducing fracture is
done by the effective strain criterion. This is the accumulated plastic strain, which easily can be found from
the normality criterion, which is one of the basic conditions for plasticity in metals. The maximum strains
that can be attained in sheet metal forming prior to
localize necking are called the forming limit strains,
which are represented in forming limit diagrams.
The region above the forming limit curve represents
the strains at which local necking failure occurs, on
the other hand below the forming limit curve the
material can be stretched without localized necking.
Peschmann (2001) obtained forming limit diagrams
experimentally to evaluate the equivalent plastic strain
at the moment of fracture and derived the following
expression:
where g is the uniform strain, = e (xe /t) is a factor depending on the necking strain e and the length
of the neck xe , t is the plate thickness and le is the
individual element length. For plate thicknesses up to
12.0 mm it is assumed that g = 0.1 and = 0.8. It
is commonly recommended that the ratio le /t is not
less than 5 for shell elements. More details of the
Peschmann criterion can be found in (Ehlers et al.
2008, Schttelndreyer et al. 2011).
Another common failure criterion is the thinning
strain criterion (or thru thickness strain criterion). This
criterion applies to shell elements only and is evaluated
on the basis of the incompressibility condition, which
is easily represented for the uniaxial tension. A general
problem with this criterion is that it tends to drop too
fast in the higher order triaxiality ranges. Furthermore,
it requires that the material is actually thinning. If the
stress state is in pure shear, the criterion will never
predict failure and is hence invalid. In finite element
simulations the thru thickness strain criterion erases
an element if the through thickness strain reaches its
critical limit. To evaluate the critical thru thickness
strain at the moment of fracture, an empirical criterion
is presented in Zhang et al. (2004):
where g is the uniform strain, e is the necking strain,

t is the plate thickness, and le is the individual element
length. The recommended values for the uniform and
necking strain for shell elements are g = 0.056 and
e = 0.540, achieved from thickness measurements
related to the calculated stress states. It is commonly
recommended that the ratio le /t is not less than 5 for
shell elements.
384
A criterion that is becoming popular is the RTCL

criterion, since it is recently incorporated in the latest version of the finite element package LS-DYNA
(Hallquist 2010). This criterion is a combination of two
continuum damage models: the Rice-Tracey and the
Crockcroft-Latham. Together they cover the full stress
triaxiality range, defined by the hydrostatic stress over
the effective stress. The RTCL model yields a very
ductile material response in compression and shear
states. The RTCL criterion is presented and verified in
(Trnqvist 2003, Simonsen and Trnqvist 2004),
although the Rice-Tracey and Cockcroft-Latham criteria dates back to the late sixties. The criterion defines
a damage indicator Di as:
where fr is the uniaxial fracture strain, H is the hydrop

static stress, eq is the equivalent stress and deq is the
equivalent plastic strain increment. The critical failure
strain is reached when Di = 1.
4 TENSILE TEST SIMULATIONS
In simulations of small-scale structures subjected to
lateral impact, it is common practice to obtain the
mechanical properties of the material by tension tests
carried out on material cut from the same plates from
which the impact specimens are taken. Therefore, in
order to verify that the impact model gives the correct plastic deformation, and also to make an initial
attempt at failure prediction, simulations of the tensile
tests can be performed. For a purely plastic response
without necking or fracture, the plastic parameters of
the material can be determined from the results of a
tensile test (Eqs. 1 and 2). However, fracture and necking occur over a length which is much smaller than the
side length of the elements considered in numerical
models, and thus these elements cannot capture such
a local phenomenon.
The mesh sensitivity can be approached with an
engineering method, in which the critical failure strain
required to give the actual experimental material fracture strain is found through numerical simulations of
the tensile tests, using different failure strains and
mesh densities. Failure strain denotes the strain value
when fracture occurs and the finite element calculations remove elements when the average normal
Figure 3. Tensile test model (Villavicencio and Guedes

Soares 2012).
strain over the element reaches the critical failure

strain. Villavicencio and Guedes Soares (2012) presented a procedure to simulate tension tests, which
combine different modelling techniques adopted in the
literature (Yu and Jones 1987, Simonsen and Lauridsen 2000, Tabri et al. 2007, Ehlers and Varsta 2009,
Villavicencio et al. 2012). Here, this procedure is
summarized and additional details of the physical
phenomenon seen in the simulations are illustrated.
The tensile specimen is a flat bar (20 6 mm)
with gauge length 100 mm. The specimen is modelled
in eight nodes solid elements with one integration
point, using the constant stress solid element formulation (Hallquist 2010). The disadvantage of this
under integrated solid element is the need to control
the zero energy modes, called hourglassing modes.
These modes tend to have periods shorter than the
ones of the structural response, and they are often
observed to be oscillatory. One way of avoiding hourglassing is adding elements capable of stopping the
formation of the anomalous modes of deformation or
defining viscous damping or elastic stiffness in the element formulations. However, these solutions increase
considerably the processing time.
In the tensile test simulations, only the length of the
tensile specimen between the clamping edges is modelled, see Fig. 3. For initiating necking, the width of the
specimen at the centre is gradually reduced by 0.5%.
The mesh is diagonally orientated to suppress anomalous deformation modes of zero deformation energy,
or in other words to avoid hourglassing. This solution is
cost-efficient since uses the same number of elements
and does not include any additional hourglassing control definition in the element formulation. The material
allows the definition of a true stress-strain curve as
an offset table and failure based on a plastic strain.
The combined material relation is used to define the
flow characteristic of the material (Fig. 4). The translational degrees of freedom are restricted at one end
and a constant displacement of 100 times the experimental speed is prescribed at the other. The critical
failure strains assigned in the true material definition
of the tensile models results from successive numerical
simulations run until they give approximately the engineering plastic strain. The force of the displaced nodes
at the free end is obtained and then plotted versus the
applied displacement, and these values used to give
the engineering stress-strain behaviour. As the critical failure strain depends on the mesh density, various
mesh sizes are simulated.
385
Figure 4. Combined material relation (Villavicencio and

Guedes Soares 2012).
Figure 6. Time steps of a typical tensile test simulation.
Figure 5. Results of tensile test simulations (Villavicencio

and Guedes Soares 2012).
The resulting engineering stress-strain curves for

different mesh sizes are presented in Fig. 5. The
dependence of the failure strain on the element size
is evident showing that this parameter is not a true
material property in numerical simulations. The simulations follow the engineering curve quite precisely
until the point of necking independently of mesh size,
but the post-necking behaviour is highly dependent on
the mesh size. The determined true stress-strain curve
should not be entirely correct after localization occurs,
which causes the engineering stress-strain to deviate
in the necking region from the experimental tensile
tests. However, the good correspondence between the
numerical simulations and the experimental results
until fracture indicates that the finer mesh sizes (1.0
and 0.5 mm) describe the stress and strain state of the
tensile specimen with sufficient accuracy.
The time steps of the simulation are shown in
Fig. 6. It is observed that when the stress exceeds the
yield strength, the specimen undergoes gross plastic
deformation and its cross-sectional area decreases uniformly along the gage length (1). At some time, the
decrease in specimen cross-sectional area is greater
than the increase in axial deformation (2). All further
plastic deformation is concentrated in this region, and
the specimen begins to neck locally (3 and 4) until fracture occurs (5). The beginning of a fracture is initiated
Figure 7. Fracture propagation.
Figure 8. Sketch of end view of rectangular test piece

after fracture showing constraint at corners indicating the
difficulty of determining reduced area.
by a crack in the centre of the specimen (neutral axis)

which propagates through the surface perpendicularly
to the applied tension (Fig. 7).
Reduction of cross-sectional area is customarily
measured only on test pieces with an initial circular
cross section because the shape of the reduced area
remains circular or nearly circular throughout the test
for such test pieces. With rectangular tests pieces, in
contrast, the corners prevent uniform flow from occurring, and consequently, after fracture, the shape of the
reduced area is not rectangular (Fig. 8).
The accurate representation of the tension test
in both shape of deformation and force-elongation
response, resulted from the modelling of two key
parameters: reduction of the central width to induce
386
Figure 9. Tensile test simulations using shell elements

(Villavicencio and Guedes Soares 2011b). GL represents the
simplified power relation. UN + GL represents the combined
material relation.
necking (Yu and Jones 1987) and diagonally orientation of the mesh to avoid hourglassing (Villavicencio
et al. 2012). Although the simulation is conducted
on a solid tensile model, the same procedure can
be adopted for shell tensile models. For example,
Villavicencio and Guedes Soares (2012) conducted
tensile test simulations of a flat bar (20 4 mm) specimen using shell elements (mesh sizes 5.0 and 2.0 mm).
The elements are modelled in four-nodes with fiveintegration points throughout the thickness, defining
the Belytschko-Lin-Tsay shell element formulation.
The computational efficiency of this formulation is
due to the mathematical simplification resulting from
the use of one-point quadrature in the plane of the
element. However, this condition can result in excessive hourglassing deformation. Thus, in addition to
the hourglass control based on the viscosity stresses
(Hallquist 2010) included in this element formulation, the diagonal orientation of the mesh helps to
decrease the deformation modes of zero deformation
energy. The simplified power relation and the combined material relation are defined. The results are
shown in Fig. 9. Since the combined material uses the
exact flow characteristics of the metal up to maximum load, the best approximation is given with this
material relation.
As additional information of tensile test simulations, it is just mentioned that the strain reference
length based material relation was used to evaluate the
numerical response of tensile tests specimens (Ehlers
and Varsta 2009). In the tensile specimens the finite
element length is equal to the strain reference length in
order to achieve a consistent dependency between the
model and the experimental records.An example of the
resulting force-elongation curves is shown in Fig. 10.
The point of fracture is captured accurately by the simulations using the strain based relation. However, the
numerical curves do not comply with the experiments
when the simple power law relation is implemented.
Furthermore, the point of failure is not predicted correctly, as the failure strain and the stress determination
according to the power relation is not finite element
length dependent.
Figure 10. Tensile test simulations: FEM represents the

strain reference length based material. FEM-ASM represents
the modified power relation (Ehlers and Varsta 2009).
COLLISION SIMULATIONS
The mentioned material relations and failure criteria

are commonly used in collision simulations of marine
structures, see for example the proceedings of the 5th
International Conference on Collision and Grounding
of Ships (Ehlers and Romanoff 2010). Although the
literature provides detailed information of the strain
hardening characteristics of the material and its influence on the results, in some cases this information
is omitted. This chapter summarizes representative
investigations on structures subjected to lateral impact
where the focus of the analysis is the validation of the
material relation.
A common characteristic of the selected references
is the use of the element formulations recommended
for the simulation of structural crashworthiness of
marine structures (Paik 2007, Hallquist 2010). Therefore, the solid elements are modelled in eight nodes
with one integration point, using the constant stress
solid element formulation and the shell elements
are modelled in four-node shell elements with fiveintegration points throughout the thickness, defining
the Belytschko-Lin-Tsay shell element formulation,
which is based on a combined co-rotational and
velocity-strain formulation (Hallquist 2010). Considering the restraints of the finite element models, most
of them use fully restrained nodal boundaries. If a
different definition of the restraints is adopted, it is
indicated.
Even though the beams are the simplest structural
elements of ship and offshore structures, they are
widely used to provide design parameters of structures subjected to impact loads. A clear example is
the Cowper-Symonds model to account for the strain
rate sensitivity of the material (Jones 1989, Jones and
Wierzbicki 1983).
Thereby, the applicability of the combined material relation and the critical failure strain predicted
by tensile test simulations is first verified in a prenotched transversely impacted beam (Villavicencio
387
Figure 11. Force-displacement response of a pre-notched

beam impacted at the mid-span (Villavicencio and Guedes
Soares 2012).
and Guedes Soares 2012). The impact specimen is a

flat bar beam (125 20 6 mm) with a notch at the
mid-span and is supported and compressed between
two support plates. The beam is impacted at the
notch location by a hemispherically indenter (diameter 20 mm) travelling with an initial kinetic energy
of 600 J. The restraints are modelled using solid
rigid compressive plates, which provide the experimental clamped condition. This representation of the
experimental supports is one of the models reported
in (Villavicencio and Guedes Soares 2011a).
The material relation and critical failure strain
dependent on the mesh size reproduce accurately the
plastic response and failure of the impact specimen as
seen in Fig. 11.
The selection of the mesh size and critical failure
strain by numerical simulations of the tensile tests is
valid for this particular experiment, because some elements in the original cross-section deform severely
and the most important effect is the development of
membrane forces.
The complete failure of the impact specimen starts
at the notch corner and extends upwards to the neutral
axis of the beam resulting in strength failure given as
a combination of tension and shear (Fig. 12). As an
element in the original cross-section of the specimen
is deformed severely and displaced from the initial
position, the most important effect of this change is
the development of membrane force. As the supports
of the transversely impacted specimen are restrained
axially, the centre line of the specimen is longer in the
deformed configuration. This stretching rises into an
axial membrane strain and an associated membrane
force. Consequently, the bending moment and transverse shear force do not play an important role and
the specimen deforms plastically due to the membrane
forces at the cross-section. Although an interactive
failure criterion between tension and shear could be
adopted, the given reasons justify the selection of a
critical failure strain calibrated by uniaxial tension
tests.
The numerical triaxiality at failure for the tensile
and impact specimens is compared in Fig. 13 in order
Figure 12. Failure mode of a pre-notched beam impacted at

the mid-span (Villavicencio and Guedes Soares 2012).
Figure 13. Numerical triaxiality in the tensile and the

impact specimens (Villavicencio and Guedes Soares 2012).
to verify the calibration of the critical failure strain

by tensile test simulations. The presented triaxiality is
calculated by the ratio between the hydrostatic stress
and the effective stress (Hallquist 2010) and is plotted
versus the normalized axial displacement of the tensile
model, on one hand, and versus the normalized transverse displacement of the impact model, on the other
hand. The chosen critical failure strain is justified due
to the close range of observed triaxiality values at the
point of failure.
Metal-forming simulations have become generalized to calibrate and validate material relations in
order to reflect the special conditions of the material. Unfortunately, the experimental impact response
is examined by penetrating the plates using quasi-static
lateral loads. The quasi-static tests have the advantage
of continuous records of the damage process, obtaining
detailed information from each specimen. However, a
quasi-static test has the disadvantage of removing the
dynamic effects produced during a real collision event.
Simonsen and Lauridsen (2000) performed quasistatic experiments and simulations of lateral indentation on thin plates, using a rigid sphere. The
simulations evaluate the mesh size and critical failure strain by numerical simulations of tensile tests
defining the simple power law relation. However, the
purely plastic response (until maximum load) is not
388
Figure 15. Finite element simulation results for the plate

punching experiments using the element length-dependent
strain and stress relation (Ehlers 2010).
Figure 14. Tensile test simulations (Simonsen and

Lauridsen 2000).
well predicted, and beyond necking is observed distortion of the engineering curves until the point of
fracture, see Fig. 14. The situation before necking
could have occurred for deviations of the log-log plot
of the logarithmic true stress-strain up to the maximum load, resulting in two straight lines with different
slopes. The phenomenon beyond necking could have
occurred for accumulated hourglassing deformation of
the shell elements since they are modelled as perfect
and regular squares.
Similar metal-forming experiments and simulations on thick circular plates are reported in Tabri
et al. (2007). The numerical analysis is focused on the
determination of the onset of fracture using the three
failure criteria summarized in Chapter 3. The parameters for the failure criteria are calibrated using tensile
test simulations of the plate material and compared
with the response of the plates forming process until
fracture. The analysis shows that the onset of fracture
is not always predicted precisely and the deviations are
attributed to the assumption of the simple power curve
relation to define the static flow characteristics of the
plate material.
The strain reference length based material relation
(Ehlers and Varsta 2009) is used to simulate plate
punching experiments in Ehlers (2010b). In the plate
specimens the finite element length is equal to the
strain reference length determined with tensile specimens in order to achieve dependency between the
tensile and plate specimen. The respective radius and
the thickness of the circular plate specimens are 170
and 4.0 mm. The simulations are carried out using
three finite element lengths: 4.4, 2.2 and 0.88 mm.
The restraints of the specimens are represented with
the general contact of LS-DYNA and a one way forming contact between the plate and the support. The
resulting force-displacement curves and shapes of the
deformation are shown Figs. 15 and 16. Prior to fracture the simulations are slightly higher in force level.
However, the fracture and plate thinning are captured
accurately by the simulation.
Figure 16. Fracture and deformation of the simulations

(Ehlers 2010).
The material relation calculated in Hogstrm et al.

(2009) is validated through forming limit tests to study
the multiaxial failure behaviour of the material in terms
of necking and fracture. In the simulations, the evolution of damage is due to initiation and propagation
of a macroscopic crack in the final phase of previous
tensile tests experiments and simulations.
Alsos et al. (2009) used the RTCL damage criterion and the Bressan-Williams-Hill (BWH) instability
criterion (see Alsos et al. 2008) to evaluate the modified power relation in stiffened plates quasi-statically
indented by a spherical nose. Although they found that
both failure criteria are satisfactory, the agreement
with the experiments is obtained introducing additional finite element techniques, such as the modelling
of weld elements in the plate-stiffener intersection and
the use of relatively small shell elements for relatively
complex structures.
In the described metal-forming simulations, the
experimental impact response is examined by penetrating the plates using quasi-static lateral loads,
which does not consider the extremely high impact
forces developed during the indenter-structure contact.
389
Figure 17. Force-displacement response. Diameter of

indenter 30 mm (Liu et al. 2012).
Moreover, it has been found that the engineering fracture strain is significantly dependent on the strain rate
sensitivity of the material since the material becomes
less ductile (more brittle) as the strain rate increases.
Therefore, the critical failure strain used for real collision simulations should take much smaller value than
that obtained from quasi-static forming analysis (Paik
2007).
The later phenomenon is seen in Liu et al. (2012).
They performed experiments and simulations of
fully clamped, laterally impacted rectangular plates
(125 75 1.4 mm). The indenters are hemispherically ended projectiles travelling with an initial kinetic
energy of 392 J. The experimental support is represented as shell rigid compressive plates, which are
based on the model proposed by Villavicencio and
Guedes Soares (2011a). The simulations use the combined material relation and the critical failure strain
is predicted by tension test simulations. The resulting force-displacement curves overestimate the point
of fracture since the dynamic failure strain is predicted by quasi-static tension tests although the plastic
responses have similar tendencies to the corresponding experimental results, see Fig. 17. Nevertheless, the
chosen critical failure strain is justified due to the close
range of observed triaxiality values at the point of failure between the tensile and the impact specimen. It
must be mentioned that in the impact specimen the
crack is extended along the necking circle provoked
by the indenter. In this circle the first failing element
is deformed severely and displaced from the initial
position, thus the most important effect of this failure
is the development of membrane forces.
In the previous references, the impact response is
examined in small-scale specimens, defining a very
fine mesh in order to capture the first failing element.
However, in larger and complex structures this design
concept is impracticable, and thus it is more difficult
to evaluate the influence of the material relation on the
accuracy of the simulations. In spite of that, collision
simulations on complex structures and using initial
velocity focused on the material relation have been
performed.
Figure 18. Comparison of deformed shapes (Choung et al.

2010).
Choung et al. (2010) conducted numerical simulations of stiffened plates struck laterally by a mass
travelling with initial kinetic energy (1635 J). The size
of the specimens is 470 670 mm. They study two
different material constitutive equations, the classical
metal plasticity model based on the von Mises yield
function and the micromechanical porous plasticity
model based on the Gurson yield function. However,
the definition of the true stress-strain relationship
(Swift hardening law) is relatively simple since no
information of strain hardening was available. Choung
et al. putt emphasis on the evaluation of different strain
rates coefficients using the Cowper-Symond model
since the impact on the panels provokes strain rates
larger than 1600 s1 . Although the material relation
is relatively simple, the simulations demonstrates that
the porous plasticity model with the porosity fracture
criterion can quantitatively predict plastic deformation
process and final fracture under impact loading if the
material constants are properly chosen, see Fig. 18.
Ehlers et al. (2008) benchmarked different failure
criteria in collision simulations of ship side structures.
The size of the side panels is 7.9 4.5 m and the plate
thickness is 20 mm. The results are compared with
large-scale ship collision experiments performed in the
Dutch Institute for Applied Physical Research during
the period of 1997 to 1998. The simple power law relation is used and failure is accounted with the three
failure criteria summarized in Chapter 3. Moreover,
three mesh sizes are evaluated: 100, 50 and 25 mm.
The results indicate that for coarser meshes the deletion of elements has a greater influence on the force
than for the fine mesh. Thus, the mesh size sensitivity
might be more important than the failure criterion itself
as is seen in Fig. 19. Intermediate penetration stages
390
role. However, this method is limited to small-scale

specimens where a very small mesh size can be used.
In most of reviewed collision simulations of marine
structures, quasi-static metal-forming analyses are
used to validate the accuracy of the true material relations. However, in real collision scenarios involving
moderate or relative large initial kinetic energies, these
material relations lose importance since the strain rate
sensitivity of the material or simply the mesh size
play the most important roles. Consequently, metalforming investigations must extend their applicability
to all the nonlinear characteristics of the material and
evaluate the response of the specimens with dynamic
impact loads. Furthermore, a detailed sensitivity study
utilising different material models, including a scatter in material properties or failure criteria would be
beneficial in order to further improve the shipbuilding
material relations available.
Figure 19. Force-displacement results (Ehlers et al. 2008)
ACKNOWLEDGEMENTS
The first author has been financed by the Portuguese
Foundation for Science and Technology (FCT), under
the grant SFRH/BD/46369/2008.
REFERENCES
Figure 20. Comparative fracture propagation plots of the

25 mm mesh size using the different failure criteria (Ehlers
et al. 2008).
are shown in Fig. 20 for a visual appreciation of the

failure propagation under the applied failure criteria.
6
CONCLUSIONS
The mathematical expressions and the experimental measurements of true material relations can be
used to perform finite element simulations of metal
structures subjected to dynamic loads involving large
deformations, such as collision simulations of marine
structures. The failure criteria used to simulate ship
collisions until fracture are based on the theory of
plasticity and are calibrated with uniaxial tension and
metal-forming tests. Thus, the applicability of these
criteria is limited to out of plane deformations of structures subjected to lateral impact, instead of the in plane
crushing mechanisms observed during ship collision
events.
The influence of the material relation and the critical failure strain dependent of the mesh is size has
been illustrated through examples of tensile test simulations. The information obtained from the tensile
simulations can be used to predict the plastic response
until fracture of structures subjected to lateral impact
where the membrane forces play the most important
Alsos HS,Amdahl J, Hopperstad O. 2009. On the resistance of

stiffened plates, Part II: Numerical analysis. International
Journal of Impact Engineering; 36: 875887.
Alsos HS, Hopperstad O, Trnqvist R, Amdahl J. 2008. Analytical and numerical analysis of sheet metal instability
using a stress based criterion. International Journal of
Solids and Structures; 45: 20422055.
ASTM (American Society for Testing and Materials), A36.
Standard Specification for Carbon Structural Steel.
ASTM (American Society for Testing and Materials), E8.
Standard methods of tension testing of metallic materials.
Choung J, Cho S-R, Kim KS. 2010. Impact test simulations of stiffened plates using the micromechanical porous
plasticity model. Ocean Engineering; 37: 749756.
Dieter GE. 1986. Mechanical behavior under tensile and
compressive loads. ASM Handbook. 8: 9910.
Ehlers S. 2010(a). The influence of the material relation on
the accuracy of collision simulations. Marine Structures;
23: 462474.
Ehlers S. 2010(b). Strain and stress relation relation until
fracture for finite element simulations of a thin circular
plate. Thin-Walled Structures. 48 (1): 18.
Ehlers S, Romanoff J. 2010. In: Proceedings of the 5th International Conference on Collision and Grounding of Ships.
http://lib.tkk.fi/Reports/2010/isbn9789526032122.pdf
Ehlers S, Varsta P. 2009. Strain and stress relation for nonlinear finite element simulations. Thin-Walled Structures.
47 (11): 12031217.
Ehlers S, Broekhuijsen J, Alsos HS, Biehl F, Tabri K. 2008.
Simulating the collision response of ship structures:A failure criteria benchmark study. International Shipbuilding
Progress; 55: 127144.
Hallquist JO. 2010. LS-DYNA Theory Manual. Livermore
Software Technology Corporation.
Hogstrm P, Ringsberg JW, Johnson E. 2009. An experimental and numerical study of the effects of length scale and
391
strain state on the necking and fracture behaviours in sheet

metals. International Journal of Impact Engineering; 36:
11941203.
Jones N. 1989. Structural Impact. Cambridge University
Press.
Jones N, Wierzbicki T. 1983. Structural Crashworthiness.
Butterworth & Co (Publishers) Ltd.
Joun MS, Eom JG, Lee MC. 2008. A new method for acquiring true stress-strain curves over a large range of strains
using a tensile test and finite element method. Mechanics
of Materials; 40: 586593.
Liu B, Villavicencio R, Guedes Soares C. 2012. Experimental and numerical plastic response and failure of laterally
impacted rectangular plates. In: Proceedings of the 31st
International Conference on Ocean, Offshore and artic
Engineering OMAE2012. Paper: OMAE2012-84015.
Paik JK. 2007. Practical techniques for finite element
modeling to simulate structural crashworthiness in ship
collisions and grounding (Part I: Theory). Ship and
Offshore Structures; 2 (1): 6980.
Peschmann J. 2001. Calculation of energy absorption of ship
steel structures in collision and grounding (in German).
Ph.D. Thesis; Hamburg University of Technology,
Hamburg.
Schttelndreyer M,Tautz I, Kubiczek JM, Fricke W, Lehmann
E. 2011. Influence of bulbous bow structures on their
collision behaviour. In: Guedes Soares, Fricke, editors.
Advances in marine structures. London: Taylor & Francis
Group; pp. 381390.
Simonsen BC, Lauridsen LP. 2000. Energy absorption and
ductile failure in metal sheets under lateral indentation
by a sphere. International Journal of Impact Engineering.
24 (10): 10171039.
Simonsen BC, Trnqvist R. 2004. Experimental and numerical modeling of ductile crack propagation in large-scale
shell structures. Marine Structures; 17: 127.
Tabri K, Alsos H, Broekhuijsen J, Ehlers S. 2007. A benchmark study on ductile failure criteria for shell elements in
multiaxial stress state. Advancements in Marine Structures, Guedes Soares & Das (Eds.). Taylor & Francis
Group: London. 401409.
Trnqvist R. 2003. Design of crashworthy ship structures.
PhD Thesis, Denmark Technical University.
Villavicencio R, Guedes Soares C. 2011a. Numerical
modelling of the boundary conditions on beams stuck
transversely by a mass. International Journal of Impact
Engineering; 38 (5): 384396.
Villavicencio R, Guedes Soares C. 2011b. Numerical prediction of impact loads in rectangular panels. In: Guedes
Soares, Fricke, editors. Advances in Marine Structures.
London: Taylor & Francis Group; pp. 399409.
Villavicencio R, Guedes Soares C. 2012. Numerical plastic response and failure of a pre-notched transversely
impacted beam. Ships and Offshore Structures; doi:
10.1080/17445302.2011.620370.
Villavicencio R, Sutherland LS, Guedes Soares C. 2012.
Numerical simulation of transversely impacted, clamped
circular aluminium plates. Ships and Offshore Structures;
7 (1): 3145.
Yu JL, Jones N. 1997. Numerical simulation of impact loaded
steel beams and the failure criteria. International Journal
of Solids and Structures; 34 (30): 39774004.
Zhang L, Egge ED, Bruhns H. 2004. Approval procedure
concept for alternative arrangements. Proceedings of the
3rd International Conference on Collision and Grounding
of Ships, Izu, Japan; pp. 8796.
392

Response of stiffeners with attached plate subjected to lateral impact

R. Villavicencio, B. Liu & C. Guedes Soares
ABSTRACT: The paper presents drop weight impact tests and finite element simulations examining the
dynamic response of stiffeners with attached plate struck transversely by a mass. The influence of the impact
velocity and the stiffener type is analyzed using the force-displacement responses of the impacted specimens.
The numerical simulations represents the stiffener-plate weld joint using shell and solid elements. Good agreement is found between the finite elements simulations and the experimental results. The plastic response of
the small-scale impact tests is used to predict the energy absorption of a typical full-scale ship bottom panel
subjected to a minor grounding incident as a practical application of this work.
INTRODUCTION
The structural design of ships concerning grounding and collision requires an accurate prediction of
the damage of stiffened plates under impact loading.
Experimental studies on laterally loaded panels have
been conducted in order to derive analytical expressions. For example, Hagiwara et al. (1983) proposed
a method for predicting low-energy ship collision
damage based on combined experiments, which determined the initiation of plate fracture, the effects of
structural details and the deformation of a typical ship
panel. Manolakos and Mamalis (1985) used a rigid
plastic analysis for predicting the structural behaviour
of longitudinally framed shell plating of struck vessel
during a minor oblique collision. Cho and Lee (2009)
developed a simplified method for the prediction of
the extent of damage on stiffened plates due to lateral
collisions.
Finite element analysis is a useful tool to predict
the extent of ship collision and its structural components damage. However, the nonlinear dynamic
analysis should be compared with experimental tests
before being used for a structural design. Unfortunately, experimental tests on full scale ship collision
are rare and very expensive. One approach is to perform scaled collision test on typical ship structural
members to validate the numerical methods for impact
analysis. It is difficult to find results from the comparison between experimental impact tests and numerical
simulations on stiffened plates. The numerical analyses of stiffened plates have been simplified to simulate
quasi-static contact. However, the effect of dynamic
lateral impact must be analyzed considering high initial velocities in the striking mass, thus representing a
realistic collision.
Some experimental-numerical impact analyses
have been performed on stiffened plates which in
general simulate the bottom or side structures in a

ship. For example, Zhu and Faulkner (1994) reported
results on the dynamic response of plates under impact
load in minor ship collision using a simplified model.
Their work gave a better understanding of the collision
process. Caridis et al. (1994) summarized the response
of thin plates subjected to dynamic loads, obtaining
good predictions of the permanent deflections. Wu
et al. (2004) presented results of a scaled double hull
structure representing ship-to-ship collision, obtaining
good results in terms of general structural response.
Ehlers et al. (2008) performed numerical simulations
of the collision response of ship side structures, finding
a strong sensitivity of the failure criteria. Villavicencio
and Guedes Soares (2011) studied numerically the
deflection and failure of small panels subjected to
lateral impact using different stiffener distributions
and impact locations.
The present paper summarizes results from experiments and numerical simulations of stiffeners with
attached plate subjected to lateral loads, thus allowing
to predict the absorption of energy during the impact
event. The sensitivity of the incident velocity and the
stiffener type is reviewed using the force-displacement
response of the tested specimens. The experimentalnumerical results are used to predict the absorbed
energy of a typical ship bottom structure subjected
to impact representing a minor grounding incident,
which is given as a practical application of the present
work.
2
DROP WEIGHT IMPACT TESTS
The experimental tests represent a situation in which

a partially supported stiffener with attached plate is
struck at the mid-span by a mass travelling with an
initial impact velocity. After the impact, the striker
393
is assumed to remain in contact with the specimen

having an initial velocity at the instant of contact and
a common velocity throughout the entire response.
Impact testing was performed using a fully instrumented Rosand IFW5 falling weight machine.A small,
light hemispherically ended cylindrical projectile was
dropped from a known, variable height between guide
rails onto a partially supported stiffener with attached
plate. A much larger, variable mass was attached to
the projectile and a load cell between the two gave
the variation of impact force with time. An optical
gate gave the incident velocity of the impact head,
and hence the velocity, displacement and the energy
it imparted could be calculated from the measured
force-time data by successive numerical integrations,
knowing the impact mass.
The experimental set up can be seen in Fig. 1.
The design of the specimens (denoted by Panel A2
and Panel A3) is shown in Fig. 2. The panel width
and length are 125 and 250 mm, respectively. The
plate thickness is 4.0 mm. The stiffeners are flat bars
25 4 mm and L profiles 50 50 5 mm. The indenter is a hemispherically ended projectile of diameter
30 mm which uses a striking mass of 54.0 kg.
The specimens were partially supported, i.e. the
edges in the length direction were fully clamped
whereas the edges in the width direction were free,
as shown in Figs. 1 and 2. The restrained edges were
supported between two thick rectangular steel plates
and were compressed by two bolts at each support.
The lower support plates were stiffened by two relatively thick plates, and were fixed to a strong structural
base to prevent their movement. Although the structural supports were made of mild steel and they could
experience some deformation during the impact, it was
assumed that they were stiff enough and did not suffer any important deformations. The torque applied to
screw the bolts and compress specimens was measured
providing a known clamping force.
The material of the plate and stiffeners is structural carbon steel and its mechanical properties were
obtained by in-house tensile tests using standard
tensile specimens and procedures (ASTM 1989). The
results of the tensile tests are presented in Table 1.
The impact test results are summarized in Table 2.
The end of the test is determined by zero contact force.
It occurs when the indenter leaves the surface of the
specimen and the specimen acquires its permanent
deformation (Values at End in Table 2).
The resulting force-displacement responses are
shown in Fig. 3.
It is observed that when the impact velocity
increases, larger transverse displacements and impact
forces are developed in both Panel A2 and Panel A3.
The initial reacting forces increase with the impact
velocity. The magnitudes of the maximum force, maximum deflection and permanent deflection are similar
between both panels when impacted at the same velocity. The shapes of the force-displacement responses
of Panel A3 show more oscillations than the ones
of Panel A2, especially at the first moment of the
Figure 1. Experimental set up.
Figure 2. Specimens: stiffeners with attached plate.

Table 1.
Mechanical properties of the material.
Property
FB 4 25
Units PL. 4.0 L 50 50 5
Yield stress
Ultimate tensile strength
Rupture stress
Rupture strain (in 100 mm)
MPa
MPa
MPa
286
426
322
0.21
367
488
384
0.18
impact. The instant change of slope, respectively

transition from plate-stiffener bending to membrane
behavior, is less profound in Panel A2, as shown in
Fig. 4. This statement indicates that the influence
of the stiffener type is important only at the first
394
Table 2.
Results of impact tests.

Values at Peak Force
Values at End
Specimen*
Force
(kN)
Defln
(mm)
Energy
(J)
Defln
(mm)
Energy
(J)
A2V2.0
A2V2.7
A2V3.3
A3V2.0
A3V2.7
A3V3.3
35.7
40.4
46.6
36.3
42.3
46.9
4.51
6.73
8.59
4.43
6.44
8.33
110.9
200.9
299.9
110.9
200.1
299.7
2.34
4.43
6.01
2.75
4.30
6.04
80.7
165.9
258.5
86.1
168.1
260.3
*A2 denotes Panel A2 and A3 denotes Panel A3. V denotes

the impact velocity (m/s).
Figure 4. Force-displacement response. Panel A2 and Panel

A3 impacted at 2.7 m/s.
Figure 5. Details of finite element model.
energy values will be sufficiently accurate, and only

the interpretation of the maximum load data could be
open to discussion.
3
FINITE ELEMENT SIMULATIONS
Previous finite element simulations were conducted in

order to predict the impact loads to use in the actual
experiments (Villavicencio and Guedes Soares 2011).
They aimed at studing the effect of the true stressstrain relationship, the mesh size and the boundary
conditions. The previous model, sketched in Fig. 5,
was computed in LS-DYNA Version 971 (Hallquist
2005). The plate and stiffener were modelled in 4-node
shell elements with 5-integration points through the
thickness. The mesh size was 2.0 mm. The supported
perimeter was constrained in all degrees of freedom.
For the striking mass, only the vertical translation
Figure 3. Force-displacement response.
was free, in which direction the initial impact velocity
was assigned. The contact mass-specimen uses nodal
mass-specimen contact (stiffness and bending response). normal projections resulting in a continuous contact
surface. The selected material provides a definition of
Beyond this point, both panels show similar plastic
the true stress-strain relationship as an offset table. The
responses provoked mainly by membrane forces.
striking mass was modelled as a rigid undeformable
The oscillations seen in the force-displacement
material, and since the falling weight assembly was
responses could be due either to the inertia of the
modelled as a simple hemisphere, an artificially large
specimen and the contact stiffness, or due to vibradensity was used to give the same mass as the one to
tions within the impact machine. Since only contactuse in the experiments.
specimen inertia oscillations are valid material impact
In the actual simulations, the finite element model
responses, it would be pertinent to filter out any
was improved in order to reproduce the experimental
experimental oscillations due to machine vibrations.
plastic response. Three new models were used (Fig. 6):
However, in practice this filtering may well affect the
One was designed in solid elements and the other two
impact response we wish to see, and it is extremely
represented the weld joint using shell and solid eledifficult to ensure that only the machine vibrations are
ments. The fillet weld cross-section takes the shape
filtered out. Since the oscillations are harmonic and
of a triangle and the measured leg length is 4.0 mm.
sufficient cycles occur prior to the end of the test, the
395
Figure 8. Force-displacement response, Panel A2. Experimental results: dashed lines. Numerical results: continuous
lines (Solid Weld model).
Figure 6. Previous and new finite element models.
Figure 9. Transversal view, Panel A2. Shape of deformation

and von mises stress distribution.
Figure 7. Force-displacement response, Specimen A2V2.7.

(E): Experimental. (1): Shell. (2): Solid. (3): Shell Weld.
(4): Solid Weld.
In the Shell Weld model, the welding is accounted

by increasing the plate and stiffener thicknesses at the
intersection. This is necessary when simulating smallscale structural elements, because adding weld gives
an increase in the stiffener resistance and a smoother
cross-section transition between stiffener and plate. An
equivalent thickness of 5.0 and 6.0 mm was considered
for the plate and the stiffener, respectively. In the solid
model the representation of the fillet weld is easier,
which is geometrically represented, see Fig. 6.
Specimen A2V2.7 (Table 2) was selected to compare the force-displacement responses of the new
finite element models. The resulting curves are shown
in Fig. 7. The maximum reacting forces are similar
in the four numerical models. The Shell and Solid
model reproduce almost the same response, overestimating the maximum and the permanent deflections. Although Shell Weld model improves the
results, the maximum and end displacements are
still overestimated. The Solid Weld model has been
favourably validated against maximum force and maximum deflection. However, the permanent deflection
Figure 10. Longitudinal view, Panel A2. Shape of deformation and von mises stress distribution.
is not accurate. Certainly, the real representation of

the weld joint helps to reproduce the experimental
response in this model. As a validation of the Solid
Weld model, Fig. 8 shows a good agreement for the
whole range of impact velocities.
The deformed shape and Von Mises stress distribution are shown in Figs. 9 and 10. The three specimens
suffer mainly global deformation. The observed local
indentation in the plate thickness is very small. The
maximum stresses occur on the lower surface of the
plate opposite the impact point and on the lower edge
of the stiffener. It is noted that the stresses are distributed in the modelled weld even for low incident
energies.
396
Figure 12. Small-scale models. (1): Original model. (2):

Ratio span-space three. (3): Small-scale bottom panel.
Figure 11. Typical ship bottom panel and representation of

a grounding incident.
PRACTICAL APLICATION
The testing of small-scale models is indispensable for

complex structural systems which are difficult to analyze theoretically and numerically or to study experimentally. Dynamic tests are conducted on a small-scale
model in order to obtain the response characteristics
of a geometrically similar full-scale prototype which
is the actual system of interest. In order to satisfy
the requirements of geometrically similar scaling, it is
necessary for the external loads to act at homologous
points.
Results of small-scale drop weight impact tests can
be extrapolated to a full-scale prototype. If the geometric scale factor is and the incident energy is
3 , the deflection is scaled by and the force by
2 . The stresses and strains in a small-scale model
and a geometrically similar full-scale prototype are
identical. The mentioned parameters obey the principles which govern geometrically similar scaling.
However, several phenomena do not scale according to
the same principles, e.g. material strain rate sensitivity
or fracture.
In this Section, the small-scale experiments and
the numerical simulations are used to predict the
force-displacement response of a full-scale ship bottom panel. The bottom panel is indented in order to
represent a minor grounding incident as illustrated
in Fig. 11. It is mentioned that this practical application just serves as a first approximation of the
Figure 13. Force displacement response. (a): Small-scale

models; (1) Original model, (2) Ratio span-space three, (3)
Small-scale bottom panel. (b): Full-scale bottom panel.
plastic response of a double bottom. Consequently, this

procedure is not used for design purposes.
Specimen A2V3.3 was selected in order to extrapolate its plastic response to the full-scale prototype.
In order to verify the boundary conditions, two finite
element models were designed. One represents a stiffener with attached plate (ratio span-space equal to
397
In minor impact events, the stiffener type does not

play an important role in the absorption of energy
and in the global deformation. The main influence of
the stiffener is observed at the very beginning impact
where the specimen reproduces its stiffness and initial
bending.
The numerical simulations of small-scale structural
elements require to include the weld joint in order
to increase the stiffener resistance and represent a
smoother cross-section transition between the stiffener
and the plate.
The practical application of the prediction of the
plastic response in a full-scale prototype has been
demonstrated. The results obey the scaling law in
minor impact events when hydrodynamic effects are
disregarded.
ACKNOWLEDGEMENTS
Foundation for Science and Technology (FCT), under
contract SFRH/BD/46369/2008.
Figure 14. Maximum displacement distribution in the

small-scale models.
three), and the other model represents the small-scale

model of the full-scale prototype (see Fig. 12). It must
be noted that the scale factor is 5. The new models gave similar force-displacement responses and
deformations at the impact point compared to the
original model, see Figs. 13a and 14. It is mentioned
that the very local indentation helps to reproduce
similar responses in the new models. Afterwards, the
minor grounding incident was simulated using fullscale numerical model (Fig. 13b). It is observed that
the displacements and the reacting forces obey the
scaling laws, i.e. the displacements are times longer
and the forces are 2 times longer. The shapes of the
plastic response are similar in both small and fullscale models. This example just shows an application
of small-scale impact tests in the prediction of energy
absorption in realistic grounding or collision scenarios. Further work should extend the applicability of
small scale dynamic tests in complex structures.
5
CONCLUSIONS
Detailed information of the impact response of stiffeners with attached plate has been obtained through drop
weight impact tests and nonlinear explicit dynamic
simulations. The discrepancies between numerical and
experimental results were due to overestimation of the
permanent deformation, whereas the maximum force
and maximum deflection were generally very well
predicted.
REFERENCES
ASTM (American Society for Testing and Materials). 1989.
Section 3, Metal tests methods and analytical procedures.
Caridis PA, Samuelides E, Frieze PA. 1994. On the dynamic
response of ship plating under lateral impact. International
Journal of Impact Engineering. 15 (2): 149164.
Cho S-R, Lee H-S. 2009. Experimental and analytical investigations on the response of stiffened plates subjected
lo lateral collisions. Marine Structures. 22: 8495.
Ehlers S, Broekhuijsen J, Alsos HS, Biehl F, Tabri K. 2008.
Simulating the collision response of ship side structures: A failure criteria benchmark study. International
Shipbuilding Progress. 55: 127144.
Hagiwara K, Takanabe H, Kawano H. 1983. A proposed
method of predicting ship collision damage. International
Journal of Impact Engineering. 1 (3): 257279.
Hallquist JO. 2005. LS-DYNA Theory Manual. Livermore
Software Technology Corporation.
Manolakos DE, Mamalis AG. 1985. On ship collisions: The
plastic collapse of longitudinally framed shell plating
subjected to oblique loading. International Journal of
Impact Engineering. 3 (1): 4155.
Villavicencio R, Guedes Soares C. 2011. Numerical prediction of impact loads in rectangular panels. Advances in
Marine Structures; C. Guedes Soares & W. Fricke (Eds).
Taylor & Francis Group: London. pp: 399409.
Wu F, Spong R, Wang G. 2004. Using Numerical Simulation
to Analyze Ship Collision. Proceedings of the 3rd International Conference on Collision and Grounding of Ships,
Izu, Japan; pp. 2733.
Zhu L, Faulkner D. 1994. Dynamic inelastic behaviour of
plates in minor ship collisions. International Journal of
Impact Engineering. 15 (2): 165178.
398

Explicit FE simulation of slamming load on rigid wedges

with various deadrise angles during water entry
Shan Wang, Hanbing Luo & C. Guedes Soares
ABSTRACT: The two-dimensional water entry of rigid wedges is studied in this work by using an explicit
finite element method based on a multi-material arbitrary Lagrangian-Eulerian formulation and penalty coupling
method. The predicted vertical force and pressure distribution on the surface of a wedge with a deadrise angle of
30 are compared with available experimental results to validate the current FEM model. The effect of deadrise
angle on the slamming load is studied by comparing the predicted results of wedge with varying deadrise angles
in terms of maximum pressure coefficient, pressure distribution and maximum slamming force. Comparisons
with analytical calculations are included as well.
INTRODUCTION
The impact between bodies and water often causes an

extremely large load and high local pressure on the
bodies. This slamming load often leads to damage on
the hull of marine structures, and affects the structural
responses. Due to its importance in the safety of marine
structures, many theoretical and experimental studies
on the slamming impact have been done in the past
years with various methodologies.
The studies were focused on three aspects: the probability of slamming for a given loading condition and
heading angle, the local slamming load and pressure
distribution on specific bodies and the motions of ship
in response to a slamming load.
For the probability of occurrence of slamming, Ochi
et al. (1964) showed figures with the typical probability distributions. Guedes Soares (1990) combined this
with the changes of course in severe weather which
would lead to increasing the probability of slamming.
The pioneering works on the estimation of slamming loads were carried out by von Krmn (1929)
who developed an asymptotic theory for wedges,
which was invalid when the deadrise angle was close
to zero. Wagner (1932) developed a 2D theory which
took the free surface elevation of the water into consideration. Ochi et al. (1973) predicted the slamming
characteristics and hull response, and evaluated the
non-dimensional factor k for the maximum slamming
pressure by a three parameters mapping of a section
shape into a circle. Stavovy & Chuang (1976) obtained
the k value according to experiments results. Ramos &
Guedes Soares (1998) proposed a method to evaluate
the vertical transient load by combining the methods
of Ochi et al. (1973) and Stavovy & Chuang (1976),

and assessed the ship structural response to the vertical
slamming induced forces.
Zhao & Faltinsen (1993) generalized the work of
many researches and proposed a fully nonlinear solution comparing it with analytical results. The method
was extended to general asymmetric bodies. Motivated
by the work of Zhao & Faltinsen (1996), Mei et al.
(1999) developed an analytical solution for the general
impact problem by adopting the conformal mapping
technique. Yettou (2007) presented an analytical solution to symmetrical water impact problems of a twodimensional wedge by taking into account the effect
of velocity reduction of the solid body upon impact.
Recently, explicit finite element methods have been
used in analysing slamming load, such as Bereznitzki
(2001), Stenius et al. (2006), Alexandru, et al. (2007)
and Luo et al. (2011).
This paper follows that line of work and the slamming loads on the surfaces of symmetric wedges with
different deadrise angles are studied by using the
explicit commercial software LS-DYNA. The results
of the vertical slamming force and pressure distribution from the experiment by Zhao & Faltinsen (1993)
are adopted in the case of the rigid wedge with a deadrise of 30 . The Wagner (1932) theory and Stavovy &
Chuangs (1976) formulations for maximum pressure
coefficient are used to compare with the predictions
of wedges with various deadrise angles from 10 to
80 . Furthermore, the pressure distributions on the
wetted surface and free surface elevation at different
time instances for wedges with deadrise angles 30 and
60 are presented to discuss the variation of pressure
during water entry.
399
2
2.1
METHODOLOGY
Impact force
As a pioneering work, von Krmn (1929) proposed

the first theoretical solution. When a rigid wedge with
deadrise angle of and unit mass of M , drop into
calm water at the speed of V0 , ignoring the gravity,
buoyancy and water resistance, the impact force may
be obtained by
Figure 1. Symmetric wedge with deadrise angle of .
2.2 Pressure distribution

where V is the impact velocity of the wedge.
According to momentum theory, V can be
expressed as:
Wagner (1932) developed a flat plate theory based

on the theory of von Krmn (1929). The solution
was derived by considering the impacting wedge as
an expanding flat plate, and solving a boundary value
problem for the plate. The pressure distribution on the
impact surface of a wedge subjected to water entry
was obtained from Laplace equation and Bernoulli
equation. For a constant impact velocity, the pressure
distribution is given by:
where Ma is the unit added mass of the wedge.

Thus,
where x is the transverse coordinate which is shown in
Figure 1 and
where Ma = 12 L2 , and L is the horizontal size of the
wedge immersed in water, so the total impact force is:
According to the idea of Stenius et al. (2006),
the pressure distribution from t = 0 to t = T when
the value comes up to maximum can be obtained by
equation (9) and the maximum value of pressure.
2.3 Maximum value of the pressure coefficients
when the deadrise angle is close to zero, the impact
force will be infinite. Therefore, the formulation above
is not applicable when the deadrise angle is small.
Mei et al. (1999) developed an analytical solution for the general impact problem by adopting the
assumptions of Zhao & Faltinsen (1996) and the
conformal mapping technique.
Based on the momentum theory and ignoring gravity, the velocity of the wedge during impact is given
by equation (3). But the added mass of the wedge is
calculated by:
where depends on the deadrise angle , and Y is

given by:
The definition of the maximum value of the pressure

coefficient is:
Assuming that the wedge enters into the water

with a constant velocity, the peak value of pressure
coefficient is obtained by Wagner (1932):
Chuang (1976), based on a large amount of experiments proposes an empirical formulation for pressure
coefficient k as:
where k1 is obtained using a series of polynomials that

fit experimental results.
400
2.4
LS-DYNA approach
The explicit finite element method LS-DYNA (Version 971 2007) is used in this work to predict the
impact force and pressure distribution on the surface
of a wedge with different deadrise angles during water
entry.
The FEM is based on a multi-material Arbitrary
Lagrangian-Eulerian (ALE) formulation and penalty
coupling method. The fluid, water and air, are modelled with Solid 164 element which is an 8-nodes
brick element, and they are defined as void materials (*MAT_NULL) which allows equations of state to
be considered without computing deviatory stresses.
The wedge is modelled with Shell 163 element which
is a 4-nodes element and can only be used in explicit
dynamic analysis, and rigid body material.
Two-dimensionality is applied by fixing all nodes
in the z-direction and assuring that the model has an
only one element in the z-direction. The boundaries of
fluids are defined as non-reflecting in order to decrease
the effect of the reflection of acoustic wave.
In order to obtain precise results from the modelling calculations, mesh density is a prominent factor.
Stenius et al. (2006) indicated that a careful selection of mesh density in relation to contact parameters
was required. Furthermore, a convergence study for
the modelling of wedge subjected to water entry was
carried by Luo et al. (2011).
The aim of this paper is not focused on the modelling technique by LS-DYNA, which is discussed in
detail in Luo et al. (2011), but on the effect of deadrise
angle of wedge exert on the slamming load.
3 VERIFICATION OF THE EXPLICIT FINITE
ELEMENT METHOD
In this section, the predicted total slamming force and
pressure distribution on a wedge with a deadrise angle
of 30 are compared with the experimental results published by Zhao & Faltinsen (1996). Corresponding to
the test conditions, the breadth of section is 0.5 m,
the total mass of wedge is 241 kg/m, and the initial
impact velocity of the section is 6.15 m/s in vertical
y-direction. Only half of the section is modelled, and
the mesh size of the fluid and wedge are both 2.5 mm.
The instant when the vertex of 2D wedge touches the
element on the water surface is set as 0.0 s. The gravity
is neglected.
As seen in Figure 2, the predicted slamming forces
have very good agreement with the experimental ones
in the initial stage, while the predictions are about 5%
larger than the measured values in the middle stage
during the water entry, which might be due to three
dimensional effects. For the results of Wagner (1932)
and Mei et al. (1999), the flow separation was not taken
into account, so the slamming forces in the later stage
are not included in the figure. In the initial stage, the
predicted slamming forces have great similarities with
the ones of Mei et al. (1999), while Wagners solution
significantly overestimates the impact force.
Figure 2. The slamming force on the symmetric wedge with

a deadrise of 30 .
Figure 3. Pressure distributions on the symmetric wedge

with a deadrise angle of 30 , for LS-DYNA, t = 0.0158 s.
The predicted non-dimensional pressure distributions on the wetted surface of the wedge are presented
in Figure 3, the theory values obtained from equation
(9) and the pressure values on P1P5 of Zhaos test are
also included. Y
is the vertical coordinate on the wedge
surface, Yd = V (t)dt is the draft of the wedge, and
V (t) is drop velocity of the wedge. However, if equation (9) is used to predict the pressure distribution, V
is assumed constant.
The predicted pressure distribution agrees well
with the experimental results. Generally speaking, the
measured values are smaller than the predictions, especially in the middle stage of the impact, while the
Wagners solution significantly overestimates them.
However, the pressure variations trend on the body
surface is exactly the same, and the locations that the
pressure coefficient comes up to maximum value are
almost the same.
SLAMMING LAODS ON WEDGES WITH

DIFFERENT DEADRISE ANGLES
4.1 Maximum value of pressure coefficient

The pressure coefficient is a non-dimensional factor
that depends on the section geometry. Obviously, the
maximum value of pressure coefficients is of much
401
Figure 4. Comparison for maximum values of pressure

coefficients between predicted results and analytical values.
significance, and it can be generally obtained from the

equation (11). When the wedge enters into the water
with a constant velocity, the method for determination
of the maximum value was proposed as equation (12)
(Wagner 1932) and equation (13) (Stavovy & Chuang
1976).
In this section, the maximum slamming pressure
coefficients are computed for the wedge with deadrise
angle varying from 10 to 45 by using the solutions
of Wagner, Chuang and LS-DYNA. In the case of LSDYNA, the maximum values on different elements are
somehow distinct. So, the average value of maximum
pressure coefficients of all elements on the surface of
wedge is considered as the global maximum value.
The results are shown as Figure 4.
When the deadrise angle is small, the predicted values are much smaller than the results from Wagners
solution, while they are closer to the results from
Chuangs solution. When the deadrise is larger, the
predictions agree well with the results from both analytical methods, however, the values are a little smaller
than those from Chuangs formulation. When the deadrise angle is close to 0 , the maximum pressure from
Wagners solution is close to infinite.
As shown in Figure 4, the smaller the deadrise
angle, the larger the pressure coefficient. This was
noticed previously by several researchers such as
Dobrovolskaya (1969), Zhao and Faltinsen (1993),
Mei (1999) and others.
4.2
4.2.1
Pressure distribution and free surface elevation
Comparisons between analytical and

numerical solution
Figure 5 compares predicted pressure distributions by
LS-DYNA and the Wagners solution. Wedges with
deadrise angles of 10 , 15 , 20 , 25 , 30 , 35 , 40 ,
45 , 60 , 70 , 80 are analysed. The impact velocity
is assumed constant in the analytical solution, so the
velocity of wedge in the FEM model is defined as
6.15 m/s in the vertical y-direction from the beginning to the end of the water entry. The wedge mass is
0.30125 kg.
The following curves are plotted in terms of the nondimensional pressure coefficient CP and the relative
location on the wetted part of the wedge surface.
The analytical predictions of pressure distributions
on the surface of wedge are calculated according
to the equation (9). The variation of the pressure
coefficient is the same at any time when the deadrise angle is defined. However, for the predictions of
LS-DYNA, the pressure distribution is varying from
one time instance to another. Corresponding to the
analytical results, the pressure distributions at the
time when the pressures come up to peak values
are presented in the figures.
For small deadrise angles, there is good agreement
between the FEM and the analytical formulation in
the initial stage of the water entry; however, the agreement is not good in the region near which the pressure
is maximum value. Figure 3 and Figure 5(e) show that
the predicted pressure is in much better agreement
with the analytical results when the impact velocity
is constant. For larger deadrise angles, the agreement is less satisfactory. Furthermore, in the case of
Wagners solution, the pressure drops fast to zero after
the pressure comes up to peak value. However, for
the predictions of pressure from LS-DYNA, the pressures drop to zero gradually along the surface of the
wedge.
Generally speaking, when the deadrise is small, the
maximum pressure is located near the spray root of the
jet before separation occurs, and the pressure distribution is approximately uniform along the surface of the
wedge when the deadrise angle is close to 45 .
When the deadrise angle is larger than 60 , the
maximum pressure is located at the keel of the
wedge, at the time instance when the wedge touches
with water.
4.2.2 Pressure distributions and free surface
elevation during water entry
Different from the analytical solution, the FEM can be
used to predict the variation of pressure distributions
on the surface of wedges at different time instances.
In this section, the wedges with deadrise angle 30
and 60 are calculated to study the differences on variation of pressure distributions. The vertical velocities
of wedges are constant.
The pressure distributions for different time instants
are presented in Figure 6, where x is the horizontal
coordinate on the wedge surface and B is the width of
the wedge. Only half of the section is modelled.
Figure 6(a) shows the pressure distributions on the
wedge with deadrise 30 at different time instances,
which indicate that the maximum pressures are located
near the spray root of the jet before flow separation occurs. After the flow separation, the maximum
pressures move to the keel of the wedge. As shown
in Figure 6(b) and (c), the maximum pressures are
located in the keel of the wedge during the entire
impact when the deadrise angle is 60 , where Figure
6(b) presents the pressure distributions before flow
separation and Figure 6(c) presents those after flow
402
Figure 5. Pressure distributions on symmetric wedges with

constant vertical velocity during water entry. CP is the pressure coefficient,
Y is the vertical coordinate on the wedge

surface, Yd = V (t)dt is the draft of the wedge, and V (t) is
drop velocity of the wedge.
.
403
Figure 7. Predicted water jet flows and pressure contours

in water by LS-DYNA for the wedge with deadrise angle
30 . (a) t = 0.00608 s, (b) t = 0.0165 s, (c) t = 0.0190 s. t = 0
corresponds to that the keel touches the water surface.
Figure 6. The predicted pressure distribution on wedges
with deadrise angle 30 and 60 during the water entry by
LS-DYNA. t = 0 corresponds to that the keel touches the
water surface. (a) Pressure distributions on the wedge with
deadrise angle 30 . (b) Pressure distributions on the wedge
with deadrise angle 60 before flow separation. (c) Pressure
distributions on the wedge with deadrise angle 60 after flow
separation.
separation. Furthermore, the maximum pressure has

small differences before flow separation, after that,
the peak value decreases gradually.
The same conclusions can also be obtained from
Figure 7 and Figure 8 which show the free surface
elevations and the pressure contours at different time
instances for wedges with deadrise angle 30 and 60 .
Before flow separation, as shown in Figure 7(a) and
Figure 7(b), the maximum pressures are located in the
inner domain, or the up-rise which was described in
Wagner (1932). The pressure in the jet flow is much
smaller than the one near the spray root of the jet. That
is the reason why the pressure coefficient drop fast
after the peak value as seen in Figure 5(e). Figure 7(c)
displays the pressure distribution along the wedge
surface after flow separation, and it shows that the

pressure reduces quickly and the peak value move to
the keel of the wedge.
As seen in Figure 8, corresponding to Figure 6(b)
and Figure 6(c) the maximum pressures are located on
the keel of the wedge during the water entry process
when the deadrise angle is 60 .
4.3 Vertical slamming force
Generally speaking, the total slamming force can be
obtained by directly integrating pressure distribution
with respect to the wetted surface of the wedge, or by
applying equation (1).
To analyse the slamming force of the wedges with
various deadrise angles, wedges with deadrise angle
from 15 to 80 are calculated by LS-DYNA in this
section. The initial impact velocity is 6.15 m/s in the
y-direction. The total masses of wedges are 241 kg/m,
and the mesh sizes are 2.5 mm which means that the
mass of the models are 0.30125 kg. According to equation (1), the vertical slamming forces can be obtained
by multiply the accelerations of wedges with the mass
of measuring section (Zhao & Faltinsen (1996).
404
Figure 8. Predicted water jet flows and pressure contours in

water by LS-DYNA for the wedge with deadrise angle 60 .
(a) t = 0.00608 s, (b) t = 0.01907 s, (c) t = 0.025967 s. t = 0
corresponds to that the keel touches the water surface.
Figure 9. Maximum values of slamming force for symmetric wedges with different deadrise angles.
CONCLUSIONS
Numerical simulation of slamming loads on symmetric wedges with various deadrise angles are carried
out by the explicit finite element code LS-DYNA to
analyse the relationship between deadrise angle and
slamming loads.
In order to validate the finite element method, the
predicted vertical slamming force and pressure distribution on the wetted surface of a wedge with deadrise
angle 30 are compared with experimental results and
analytical values.
For the time history of slamming forces, predictions
of LS-DYNA are in good agreement with experimental
results by Zhao (1996), and they have great similarities with the ones of Mei (1999) in the initial stage
of water entry, while Wagners solution significantly
underestimates the impact forces.
For the pressure distribution at one time instant,
the measured values are smaller than the predictions of LS-DYNA, especially in the middle stage of
the impact, while the Wagners solution significantly
overestimates them.
When the impact velocity is assumed constant, the
maximum slamming pressure coefficients are computed for the wedge with deadrise angle varying from
10 to 45 by using the solutions of Wagner, Chuang
and LS-DYNA. The results show that the smaller is
the deadrise angle, the larger is the pressure coefficient. Pressure distributions for wedges with the
deadrise angle varying from 10 to 80 are presented as
well, from which can be concluded that the maximum
pressure occurs near the spray root for the small
deadrise angle, while it moves to the keel of the wedge
for larger deadrise angles.
Taking the deadrise angle 30 and 60 for example, the pressure distributions on the wetted surface of
wedges at different time instances are presented, and
the free surface elevations are included. The predictions of LS-DYNA can correctly describe the variation
of pressure along the surfaces of wedges from the
initial stage, flow separation, to the end. For a small
deadrise angle, the maximum pressures are located
near the spray root of the jet before flow separation occurs, after the flow separation, the maximum
pressures move to the keel of the wedge, while the maximum pressures are located on the keel of the wedge
during the entire impact when the deadrise angle is
large.
Finally, the maximum slamming forces for symmetric wedges with different deadrise angles show that
the larger is the deadrise angle, the smaller is the peak
value of slamming force.
ACKNOWLEDGEMENTS
Figure 9 shows the maximum values of slamming

force for symmetric wedges with different deadrise
angles. As expected, the larger the deadrise angle, the
smaller the peak value of slamming force.

405
REFERENCES
Alexandru, I., Brizzolara, S., Viviani, M., Couty, N., Donner,
R., Hermundstad, O., Kukkanen, T., Malenica, S. &
Termarel, P. 2007. Comparison of experimental and
numerical impact loads on ship-like sections. Advancements in Marine Structures, Guedes Soares, C, and Das,
P.K., (Eds), Taylor & Francis, UK, 339349.
Aquelet, N., Souli, M. & Olovsson, L. 2006. EulerLagrange
coupling with damping effects: Application to slamming
problems. Computer Methods in Applied Mechanics and
Engineering. 195, 110132.
Bereznitski, A. 2001. Slamming: the Role of Hydroelasticity.
International Shipbuilding Progress. 48, 333351.
Dobrovolskaya ZN. 1969. On some problems of similarity
flow of fluids with a free surface. Journal of Fluid Mech.
36, 805829.
E1-M. Yettou, A. Desrochers & Y. Champoux. 2007. A new
analytical model for pressure estimation of symmetrical water impact of a rigid wedge at variable velocities.
Journal of Fluid and Structures, 23, 501522.
Guedes Soares, C., 1990. Effect of the heavy weather maneuvering on the wave induced vertical bending moments in
ship responses, Journal of Ship Research, 34, 1, 6068.
LS-DYNA, Keyword Users Manual, Livermore Software
Technology Corporation, Version 971, May 2007.
Luo, H., Wang, S. & Guedes Soares, C. 2011. Numerical
prediction of slamming loads on a rigid wedge subjected
to water entry using an explicit finite element method.
Advances in Marine Structures, Guedes Soares, C, and
Fricke, W., (Eds), Taylor & Francis, UK, pp. 4147.
Mei, X., Liu, Y. & Yue, D.K.P., 1999. On the water impact
of general two-dimensional sections. Applied Ocean
Research, 21, 115.
Ochi, M.K. & Motter, L.E., 1973, Prediction of slamming characteristics and hull response for ship design,
Transactions SNAME, 81, 144190.
Ramos, J. & Guedes Soares, C. 1998. Vibratory response of
ship hulls to wave impact loads. International Shipbuilding Progress. 45(441), 7187.
Stavovy, A.B & Chuang, S.L. 1976, Analytical determination
of slamming pressures for high speed vessels in waves,
Journal of Ship Research, 20, 190198
Stenius, I., Rosn, A. & Kuttenkeuler, J. 2006. Explicit
FE-modeling of fluid-structure interaction in hull-water
impacts. International Shipbuilding Progress. 53, 1031
121.
Von Krmn, T. 1929. The impact on seaplane floats during landing. National Advisory Committee for Aeronatics.
Technical note No. 321, 309313.
Wagner, H. 1932. Uber Stoss- und Gleitvergange an der
Oberflache von Flussigkeiten. Zeitschrift fuerAngewandte
Mathematik und Mechanik, 12, 193215.
Zhao, R. & Faltinsen, O.M. 1993. Water Entry of TwoDimensional Bodies. Journal of Fluid Mechanics, 246,
593612.
Zhao, R., Faltinsen, O.M. & Aarsnes, J.V. 1996. Water
entry of arbitray two-dimensional sections with and without flow separation. Proc. 21st Symposium on Naval
Hydro-dynamics. 408423.
406

Comparison of numerical results with experiments on ultimate strength of

stocky stiffened panels
Mingcai Xu & C. Guedes Soares
ABSTRACT: The ultimate strength of several stocky stiffened panels under axial compressive load is calculated
by finite element (FE) analyses, and then are compared with experimental results. Three type stiffeners with
different steels are considered, including flat, L and U stiffeners. They are made of mild or high tensile steel
for bar stiffeners and mild steel for L and U stiffeners. To analyze the behavior of the middle bays, avoiding
the effect of the boundary conditions, three bays stiffened panel were used in the tests and in the FE analyses.
The influences of the stiffeners geometry on the ultimate strength are investigated.
INTRODUCTION
Finite element (FE) codes have been used to analyze

the strength of very complicated structures as it can
predict in detail the complex collapse behaviour of
the stiffened and unstiffened panels (IACS, 2006).
The experimental results can be used as validation,
but the calculations would give more information to
investigate the essence of collapse behaviour.
Several ultimate strength tests have been conducted
in the past on simple stiffened panels under compression (Faulkner 1977; Horne 1976; Mathewson and
Vinner 1962; Smith 1979). Tanaka and Endo (1988)
carried out a series of experimental and numerical
investigations on the ultimate compressive strength
characteristics of longitudinally stiffened panels having three flat bar stiffeners by three bays, which
were intended to fail by local buckling or tripping of longitudinal stiffeners. Ghavami and Khedmati (2006) conducted FE studies comparing the
results with the two previously mentioned experimental programs. They studied the influences of
stiffener cross-section with rectangular (R), L and
T shapes. Moreover, the spacing of the stiffeners
and the presence of rigid transverse stiffeners on the
buckling behaviour up to collapse have also been
investigated.
Gordo and Guedes Soares (2008) have conducted
an experimental study of stiffened panels under compression, which are calculated in this paper with FE
analyses. The base geometry is the one used on the
box girders tests of Gordo and Guedes Soares (2007).
In this regard, the results can be compared with those
of similar stiffened plates belonging to much larger
structures.
DESCRIPTION OF THE SPECIMEN
Figure 1 and Table 1 show the geometry and material of the different panels adopted in the experiment
and in the present study. The A and B mean different
number of stiffeners. Four series of experiments and
FE analysis were carried out using two different types
of steel as follows:
Fully S690 structure: S690, on plating and bar
stiffeners.
Hybrid bar structure: S690, on plating and mild
steel on bars.
Figure 1. Geometry of stiffened panels.
407
Table 1.
Geometry and material of stiffened panels.

Dim (mm)
Plate
FS2-A3
FS2-B3
BS2-A3
BS2-B3
LS2-A3
LS2-B3
US2-A3
US2-B3
300 600 4
600 600 4
300 600 4
600 600 4
300 600 4
600 600 4
300 600 4
600 600 4
690
690
690
690
690
690
690
690
Stiffener
FS2-A3
FS2-B3
BS2-A3
BS2-B3
LS2-A3
LS2-B3
US2-A3
US2-B3
I 20 4
I 20 4
I 30 8
I 30 8
L38 19 4
L38 19 4
U (40 150 40) 2
U (40 150 40) 2
690
690
343
343
296
296
200
200
Figure 2. Setup of the test of stiffened plates (left). Details

of the lateral support of the frames in vertical guides (right).
collapse, due to the bi-dimensional geometry of the

cross section of the panels. The hydraulic flow was
controlled manually due to limitations of the control
device which means that the shortening rate was not
constant during the tests.
4
Hybrid L structure: S690, on plating and mild steel

on L stiffeners.
Hybrid U structure: S690, on plating and mild steel
on U stiffeners.
The S690 steel was supplied by Dillinger
Hutterwerke in the form of sheet of 4 mm thickness,
and the mild steel was supplied by Lisnave Shipyard.
The stiffened plates were manufactured at Lisnave
Shipyard according to the standard techniques of the
shipyard (Gordo and Guedes Soares 2008).
3 EXPERIMENTAL WORK
The experiment work that is used here as reference
for the numerical calculations have been conducted by
Gordo and Guedes Soares (2008) and a brief account
is given here for completeness.
A 300 ton hydraulic press was used to perform
the tests of the panels under uniaxial compression.
Figure 2 shows the general arrangement of the tests
(left) and a detailed view of the support for the framing
systems which intends to reproduce simply supported
boundary conditions. The lateral edges of the panels
are totally free to move out-of-plane and to rotate.
This means that large panels (B series) should be less
affected by the lack of effectiveness at the lateral plating edges during buckling. In fact, the percentage of
the total cross-section area with reduced effectiveness
due to unsupported lateral edges is lower on the wide
panels than in the narrow ones and thus, the expected
ultimate load is higher for the wide panels.
The transverse framing system is simply supported
on a U bar in each side, allowing longitudinal displacement and rotation but avoiding out-of-plane displacement from the initial plane of load. The loaded
top edges have full contact with the steel beds corresponding to nearly clamped conditions, at least until
NONLINEAR FINITE ELEMENT ANALYSIS
4.1 Finite element model

The geometric and material nonlinearities are both
taken into account, including elastic-plastic large
deflection. The material property assumed use the
characteristic values of yield strength and Youngs
Modulus, where appropriate, a bi-linear isotropic
elastic-plastic material model excluding strain rate
effects is to be used. A plastic tangent modulus of
1000 MPa is acceptable for normal and higher strength
steel (ABS 2006). The following are the other material
properties: Youngs modulus, E = 206 GPa; Tangent
modulus, ET = 1000 MPa; Poisons ratio, = 0.3.
The FE code used for simulation is ANSYS. This
is a widely used finite element code for nonlinear
structural analyses. The shell 181 is adopted which
is suitable for analyzing thin to moderately-thick shell
structures. It is a 4-node element with six degrees of
freedom at each node: translations in the x, y, and z
directions, and rotations about the x, y, and z-axes.
The shell element mesh should be fine enough to properly describe the model shape (also after deformation).
Therefore, a balance between required accuracy and
efforts is needed. It is considered that the element
size to thickness ratio (usually at least 5) (ISSC 2009).
Hence, there are 40 elements in one bay in the longitudinal direction, 16 elements in the transverse direction,
5 elements in the web and 3 elements in the flange.
Figure 3 shows FE model of stiffened panels.
4.2 The boundary condition and loading
Figure 2 shows that the loaded top edges have full
contact with the steel beds corresponding to nearly
clamped conditions, at least until collapse, due to the
bi-dimensional geometry of the cross section of the
panels. The coordinate system is shown in Figure 4.
408
Figure 4. The coordinate and model of stiffened panel.
The importance of their research is that the buckling

mode component can be isolated. Three types of initial
out-of-straightness should be accounted for plate and
stiffener imperfections, as following (Paik 2009):
Hungry horse mode initial deflection of local plate
panels
Column-type initial deflection of stiffeners

Figure 3. The FE model of stiffened panel.
The boundary conditions of the stiffened panels are as

following:
Side-ways initial deflection of stiffeners due to

angular rotation about panel-stiffener intersection
line
A-A1 at the stiffener and plate: UX , UY , UZ , RX ,

RY and RZ .
B-B1 at the stiffener and plate: UY , UZ , RX , RY , RZ
and equal x-displacement by coupling UX degree.
C-C1 and D-D1 at t transverse frame and plate
intersection: UZ .
The pressure in the z direction is applied on the edge
of the plate and stiffeners. The pressure value of the
plate is Pp , and then the pressure value on stiffener is
Ps , Ps = Pp tw /tp , where tw and tp are the thickness
of web and plate.
4.3 Geometrical initial imperfections
The imperfections are caused during a complex fabrication process and are subject to significant uncertainty related to the magnitude and spatial variation.
These initial imperfections affect significantly the ultimate strength of the stiffened panels and should be
accounted for. The buckling model component of the
deflected shape has the most significant weakening
effect (Guedes Soares and Soreide 1983).
Kmiecik (1971) considered the initial deflection as
the superimposition of the Fourier components for
the first time. The behavior of plates subjected to
buckling loads depends to a considerable degree on
the shape of their initial deflection (Kmiecik 1995).
where a and b are the length and width of the plate.

To get the initial imperfection in FE analyses, the
shapes of initial imperfections are divided into plate
initial deflection, column-type initial distortions of
stiffeners and sideways initial distortions of stiffeners. Firstly, linear buckling analysis is performed for
the target stiffened panel and find out the related buckling modes of plating and stiffener. Then the geometry
properties, for example the thickness of plates and
stiffener, are changed to decouple those deformations
of interest from lower eigenmodes and get desired
shapes for plate out-of-plane and stiffener out-of-plane
deformations. The three types of initial distortions are
superimposed altogether in FE model.
COMPARISION BETWEEN EXPERIMENTS

AND FEM ANALYSIS
The strength of the panel was obtained by summing

the reaction force on each node (Ri ), on the opposite
409
boundary were the load is applied and divided by the

sectional area of the stiffened panel (At ):
When the test starts and the edge starts being displaced,
large displacements occur under a small level of load.
This very long initial shortening at low stresses shown
in the tests is due to the rearrangement of the test setup
until every part of the panel, support and hydraulic
machine is in full contact. This was partially removed
from the other graphics due to lack of interest. So the
initial shortening at low stresses should be removed
from the test data. The Su and Syp are averages stress
of stiffened panels and yield stress of plate. The Ea and
Eyp are averages strain of stiffened panels and yield
strain of plate.
5.1
FS series results
Figure 5 and Figure 6 show the average strain stress

curves of the experiments and FE analyses for FS
series, which show a linear behavior until the ultimate
compressive stress was achieved, and then the ultimate
strength drop very quickly in FE analyses. However,
this phenomenon is very different with the experiment.
The ultimate strength of the FE analysis is bigger
than the test. Furthermore, the average strain stress
curve at unload phase are still different between them.
Because the support devices of the transverse frames
on the experiments showed to be not totally effective
at high loads allowing some out-of-plane movement of
the transverse frames. Those movements correspond,
in practice, to have a higher effective length of the
column which very much reduces the Euler stress
(buckling elastic stress). The ultimate stress obtained
in the test (249 MPa) corresponds to an effective length
of the column of 344 mm which is 57% of the total
length of the panel and 1.72 times the spacing between
frames.
The stiffeners experience lateral buckling and then
induce panel failure from Figure 7 and 8. It can
be seen that the plate and stiffener collapses as a
unit. This phenomenon explains that the collapse was
sudden resulting from the yielding of the stiffeners.
The stiffeners are not strong enough.
5.2
Figure 5. Average s strain stress curves of FEM for FS2A3.
Figure 6. Average strain stress curves of experiment and

FEM for FS2B3.
BS series results
Figure 9 and 10 show the average strain stress curves

of the tests and the FE analyses for the BSseries specimens. The stiffeners are thick bars made of steel of
343 MPa yield stress and the associated plate is made
of 690MPa nominal yield stress steel. There was a
very great discharge of load after the ultimate load
was achieved during the development of large out of
plane deformations of the panel between frames. The
ultimate strengths of BS2A3 are close between the test
Figure 7. Von Mises stress distributions at the ultimate limit

state for FS2A3.
and the FE analysis. However, their stiffness are still

different.
In the test of the BS2B3, the ultimate average
stress achieved was very low compared to the expected
410
Figure 11. Premature failure of the loading support leading

to premature collapse on the adjacent span of the panel.
Figure 8. Von Mises stress distributions at the ultimate limit
state for FS2B3.
Figure 12. Von Mises stress distributions at the ultimate

limit state for BS2B3.
FEM for BS2A3.

FEM for LS2-A3.
FEM for BS2B3.
result. The ultimate stress expected to be higher than

that obtained on the BS2A3 experiment because the
panel was wider than the A panel and thus, the free
end edges effects were not so important. However,
in this test, the geometry of the support used on this
experiment was not appropriate and there was a premature collapse on the middle part of one of the external
spans of the panel due to a non-uniform distribution of
load, especially near the contact with the supports in
Figure 11. The ultimate average stress of the panel at
collapse in the test was 366 MPa, which much smaller
than the calculation 489 MPa. The deformation of
side bays is bigger than middle bay at the ultimate limit

state in Figure 12 with this boundary condition and
model configurations. Hence, the experiment should
use a much stronger supporting condition.
5.3 LS series results
of the experiments and the FE analyses for LS specimens. The slope of the average strain stress curve
is also different between FE analysis and the test.
The stiffness of the FE analysis is bigger than the test.
The main reason may be the residual stresses in stiffened panel, and this aspect is not considered in FE
analysis.
411

FEM for LS2B3.

FEM for US2A3.

limit state for LS2-A3.
FEM for US2B3.
5.4 US series results

limit state for LS2-B3.
The buckling of plate occurs in the middle span of

the panels, and then induces stiffener failure in the LS
series specimens from Figure 15 and 16. The series
of panels reinforced with L stiffeners allowed obtaining results as expected that the L series have good
capability to avoid buckling.

of the test and the FE analysis. The slopes of the
average strain stress curves are also different between
FE analysis and the test.
The panel of US2A3 had a premature global collapse in one of the external spans and near the supports
induced by the first failure of the flange plating in test
as shown in Figure 19. The ultimate load achieved was
323MPa and after some shedding, the panel regained
some ability to sustain more load. Hence, the ultimate
strength of the US2A3 in test is much smaller than in
FE analysis.
The US series has U stiffeners of 2 mm thickness.
Thus, the slenderness of the flange plating is higher
than the slenderness of the associated plating. The
flange plate occurs buckling in Figure 19 and 20, and
then the overall collapse is very sudden in US series
specimens.
There are three panels failure in the tests, which
are FS2A3, BS2B3 and US2A3. Table 2 shows the
412
Table 2. Ultimate strength comparison of successful

experiments (MPa).
Panel
Test
FS2A3
249
FS2B3
341
BS2A3
440
BS2B3
366
LS2A3
500
LS2B3
542
US2A3
323
US2B3
462
Mean value
FEM
Type of collapse in test
479
501
472
489
468
489
417
444
1.47
1.07
0.94
0.90
0.96
1.07
Lateral support failure

Stiffener
Plate
Premature failure
Plate
Plate
Premature failure
Plate
Note: Sut- ultimate average stress by tests; Suf- ultimate

average stress by FE analysis; R- Suf-/ Sut.
Figure 19. Final collapse shape of US2B3 panel on the
plating in test.
ultimate strength comparison between the test and the

FE analyses, in which only the successful experiments
are compared. The mean value of R is 1.07, which
indicates that the ultimate strengths are closed between
the tests and FE analyses. However, the stiffness of
panels are different in all specimens. The main reason
may be the residual stresses in stiffened panel, and this
aspect is not considered in FE analyses.

limit state for US2B3.
CONCLUSIONS
The ultimate strengths of three bays stiffened panels

by FE analyses are compared with the experimental results. Column-induced collapse modes occur in
FS and BS series. In the design, the stiffener should
be stronger to avoid column-induce collapse and the
aspect ratio should be considered.
The stiffener flange plate collapse first in US
series. The collapse is relation with not only the aspect
ratio but also the type of stiffener. The best results
in terms of ultimate strength were obtained for LS
panels, which have good capability to avoid buckling.
The ultimate strengths are closed between the tests
and FE analyses. However, the stiffness of panels are
different in all specimens. The main reason may be the
residual stresses in stiffened panel, and this aspect is
not considered in FE analyses.
The ultimate strength stress distribution of the U
series show that buckling in the flange occurs very
early.
ACKNOWLEDGEMENTS

limit state for US2B3.

para a Cincia e Tecnologia), under contract
SFRH/BD/65120/2009.
This work contributes to the activities
of MARSTRUCT VIRTUAL INSTITUTE, (www.
mar-struct-vi.com) in particular its Technical
Sub-committeees 2.3 on Ultimate Strength and 3.3
Experiments on Ultimate Strength.
413
REFERENCES
ABS, 2006. Rules for building and classing, steel vessels.
Faulkner, D., 1977. Compression tests on welded eccentrically stiffened plate panels. In: Dowling PJ, et al., editors.
Steel Plated Structures. London: Crosby Lockwood
Staples; pp. 1309.
Ghavami, K., Khedmati M.R., Numerical and experimental investigations on the compression behaviour of
stiffened plates, Journal of Constructional Steel Research
62 (2006), 10871100.
Gordo, J.M., Guedes Soares, C., 2007. Experimental evaluation of the behavior of a mild steel box girder under
bending moment. In: Guedes Soares C, Das, P.K., (Eds).
Advancements in Marine Structures. Taylor and Francis;
pp. 37783.
Gordo, J.M., Guedes Soares, C., 2008. Compressive tests on
short continuous panels, Marine Structures 21, 113137.
Guedes Soares, C., Soreide, T.H., 1983. Behaviour and
Design of Stiffened Plates under Predominantly Compressive Loads, International Ship building Progress,
Vol.300-January 1983 No. 341.
Horne, M.R., Narayanan R., 1976. Ultimate capacity of stiffened plates used in girders. Proc Inst Civil Eng 1976;
61:25380.
IACS, 2006. Common structural rules for double hull
oil tankers, International Association of Classification
Societies, London.
Paik, J., Branner, K., Choo, J., Czujko, J., Fujikubo, M.,
Gordo, J.M., Parmentier, G., Iaccarino, R., ONeil, S.,
Pasqualino, I., Wang, D., Wang, X. and Zhang, S.,
Committee III.1 Ultimate Strength, 2009, In: 17th
International Ship and Offshore Structures Congress
(ISSC2009), Vol. 1. C. Jang and S. Hong, editors. Seul,
South Korea: University of Seoul, pp. 375475.
Paik, J.K., 2009. Nonlinear finite element method models for ultimate strength analysis of steel stiffened-plate
structures under combined biaxial compression and lateral
pressure actionsPart II: Stiffened panels, Thin-Walled
Structures 47(2009), 9981007.
Kmiecik, M., 1971. Behaviour of axially loaded simply supported long rectangular plates having initial deformations,
Report No. R84, Ship Research Institute, Trondheim,
1971.
Kmiecik, M., Jastrzebski, T., Kuzniar, J., Statistics of Ship
Plating Distortions, Marine Structures 8(1995), 119132.
Mathewson, J., Vinner, A., The strength and stiffener of
plating stiffened by flat bars. Part 1: axial compressive
loading tests. Report 392. UK: BSRA; 1962.
Smith, C., Compressive strength of welded steel ship grillages. Trans RINA 1979;117:32559.
Tanaka, Y., Endo, H., Ultimate strength of stiffened plates
with their stiffeners locally buckled in compression. J Soc
Naval Architects Japan, 1988; 164:45667.
414

Numerical study of the effect of geometry and boundary conditions

on the collapse behaviour of long stiffened panels
Mingcai Xu & C. Guedes Soares
ABSTRACT: A finite element (FE) analysis is conducted of 120 configurations of stiffened panels with
different boundary conditions to analyze the influences of the stiffeners geometry and boundary conditions
on the ultimate strength of long stiffened panels under compression. The long stiffened panels with four type
stiffeners are made of mild or high tensile steel and have bar, L and U stiffeners. From the results of the FE
analyses, it is concluded that two of the four models studied can produce adequate boundary conditions at the
loaded edges longitudinally. The stiffened panels with different combinations of mechanical material properties
and geometric configurations are considered. The initial geometric imperfections affect the collapse behavior of
stiffened panel and are analyzed in FE analyses.
INTRODUCTION
The finite element (FE) codes have been used to analyze the stress distributions and deformations of very
complicated structures with the accuracy demanded
in engineering applications under all kind of loading
conditions. They are also suitable tools for assessing
the ultimate strength of ship structures. The advanced
buckling analysis method bases on nonlinear analysis
techniques or equivalent, which predict the complex behaviour of stiffened and unstiffened panels
(IACS-CSR 2006). The extent of the model used in the
buckling assessment is to be sufficient to account for
the structure that is surrounding the panel of interest,
and to reduce the effect introduced through the boundary conditions. In general, the model is to include more
than one stiffener span in the stiffener direction and the
portion between two support members in the direction
normal to the stiffeners. However, it is not explained
what extension of the model is suitable.
Zhang & Khan (2009) and Fujikubo et al. (2005)
analyzed the ultimate strength of plates using nonlinear FE software by one full bay plus two half
bays (1/2+1+1/2 bays) model. Tanaka and Endo
(1988) carried out a series of experimental and
numerical investigations on the ultimate compressive strength characteristics of longitudinally stiffened
panels having three flat bar stiffeners with three bays,
which were intended to fail by local buckling or
tripping of longitudinal stiffeners.
A series of nonlinear finite element method computations were carried out in Paik et al. (2009) in two full
bays (1 + 1 bays) model with various parameters of
influence to investigate the ultimate strength characteristics of stiffened panels representative of ship hulls.
Moreover, the boundary of stiffened panels is supported by members such as longitudinal girders and
transverse frames. The degree of rotational restraints
at the panel boundary is equivalent to neither free nor
fixed.To circumvent these problems, the tests of Gordo
and Guedes Soares (2008) used specimens with three
bays panels longitudinally. The use of three-bay panels
instead of one single-bay panels allows to have more
realistic results by avoiding boundary conditions problems for the central plates related to eccentricity of load
and to include the interference between adjacent panels, which was found to be significant by Lus et al.
(2008a, b).
Xu and Guedes Soares (2011) conducted a series
of calculations to compare with the tests results of
Gordo and Guedes Soares (2008) and discrepancies were evident, suggesting that there might have
been difficulties in the experiments in imposing the
boundary conditions on the panels.
The purpose of this paper is to investigate the influence of the type of boundary condition and model
geometry of stiffened panel on their ultimate strength,
in order to determine whether other configurations
of stiffened panels could be less sensitive to the
details of the boundary conditions used in experiments. The ultimate strength of stiffened panels under
axial compression are calculated for 120 specimens
with different boundary conditions and geometry configurations. These stiffened panels include one bay,
two full bays (1 + 1 bays), two half bays plus one full
bay (1/2 + 1 + 1/2 bays) in the longitudinal direction
and three bays. The plate is always made of very high
strength steel (S690), but the stiffeners are made of
mild or high tensile steel for bar stiffeners and mild
steel for L and U stiffeners. The base geometry is
415
Table 1.
Geometry and material of stiffened panels.

Plate
Sample
Dim (mm)
FS4-I3
FS4-I21
FS4-I22
FS4-I1
BS4-I3
BS4-I21
BS4-I22
BS4-I1
LS4-I3
LS4-I21
LS4-I22
LS4-I1
US4-I3
US4-I21
US4-I22
US4-I1
(300 i)(400 3) 4
(300 i) (200+400+200) 4
(300 i) (400+400) 4
(300 i) 400 4
(300 i) (400 3) 4
(300 i) (200+400+200) 4
(300 i) (400+400) 4
(300 i) 400 4
(300 i) (400 3) 4
(300 i) (200+400+200) 4
(300 i) (400+400) 4
(300 i) 400 4
(300 i) (400 3) 4
(300 i) (200+400+200) 4
(300 i) (400+400) 4
(300 i) 400 4
690
690
690
690
690
690
690
690
690
690
690
690
690
690
690
690
Figure 1. Geometry of 1/2 + 1 + 1/2 and 3 bays.
Stiffener
Dim (mm)
FS4-I3
FS4-I21
FS4-I22
FS4-I1
BS4-I3
BS4-I21
BS4-I22
BS4-I1
LS4-I3
LS4-I21
LS4-I22
LS4-I1
US4-I3
US4-I21
US4-I22
US4-I1
I 20 4
I 20 4
I 20 4
I 20 4
I 30 8
I 30 8
I 30 8
I 30 8
L38 19 4
L38 19 4
L38 19 4
L38 19 4
U (40 150 40)
U (40 150 40)
U (40 150 40)
U (40 150 40)
690
690
690
690
343
343
343
343
296
296
296
296
200
200
200
200
2
2
2
2
Figure 2. Geometry of 1 + 1 bays.
the one used on the box girders tests of Gordo and

Guedes Soares (2007) (2008).
DESCRIPTION OF THE MODELS FOR THE

ANALYSIS
Table 1 shows the geometry and material of the

stiffened panels.
The A-E means different number of stiffeners. The
value of I and i: i = 1 when I = A, i = 2 when I = B,
i = 3 when I = C, i = 4 when I = D, i = 5 when I = E;
Frame dimensionL50 20 6 (mm);
Figures 1, 2 and 3 show the geometry of different
panels. Four series of FEM analyses were carried out
using two different types of steel as follows:
Fully S690 structure: S690, on plating and bar
stiffeners.
Hybrid bar structure: S690, on plating and mild
steel on bars.
Figure 3. Geometry of 1 bay.
Hybrid L structure: S690, on plating and mild steel

on L stiffeners.
Hybrid U structure: S690, on plating and mild steel
on U stiffeners.
3
3.1
NONLINEAR FINITE ELEMENT ANALYSES

Finite element model
To investigate the influence of different models and

boundary conditions, 3 bays, 1/2 + 1 + 1/2 bays, 1 + 1
416
Table 2. The boundary condition of FE model.
Boundary
condition
A-A1 :UX , UY , UZ , RX , RY and RZ
A-A1: UX , UY , UZ ,
B-B1 : UY , UZ , RX , RY , RZ and equal Ux
B-B1 : UY , UZ , equal Ux
C, C1 , D, and D1 on frame: UZ
C, C1 on frame: UZ
The intersection between frame and plate: UZ
AB, A1 B1 edge: UY, RZ and RX
3 bays
C1
C2
C3
2 bays
(1 + 1)
2 bays
C4
C5
C6
C7
1 bay
C8
C9
Note: Different models and boundary conditions correspond to different location of stiffener.
The other material properties are: Youngs modulus, E = 206 GPa; Tangent modulus, ET = 10 GPa;
Poisons ratio, v = 0.3. The FE code used for simulation is ANSYS. The shell 181 element is adopted
which is suitable for analyzing thin to moderatelythick shell structures. It is a 4-node element with six
degrees of freedom at each node: translations in the x,
y, and z directions, and rotations about the x, y, and
z-axes. The shell element mesh should be fine enough
to properly describe the model shape (also after
deformation). Therefore, a balance between required
accuracy and efforts is needed. It is considered that
the element size to thickness ratio (usually at least
5) (ISSC 2009). Hence, there are 40 elements in the
longitudinal direction, 16 elements in the transverse
direction, 5 elements in the web and 3 elements in the
flange. Figure 4 only shows the FE model for three
bays stiffened panels.
3.2 The boundary conditions and loading
Figure 4. The FE model of continuous stiffened panel.
bays and 1 bay stiffened panels are modeled for FE

analyses.
The geometric and material nonlinearities are taken
into account, including elasticplastic large deflection.
The assumed material properties use the characteristic values of yield strength and Youngs Modulus.
Where appropriate, a bi-linear isotropic elastic-plastic
material model excluding strain rate effects is used.
The second tangent modulus, representing the hardening in the bilinear isotropic elastic-plastic material
model is assumed as 1000 MPa, which is acceptable
for normal and higher strength steel (ABS 2006).
The stiffened panel models, including 3 bays,

1/2 + 1 + 1/2 bays, 1 + 1 bays and 1 bay, are simulated
with different boundary conditions. To investigate the
effect of the model geometry and boundary conditions on the collapse behavior of stiffened panel, nine
boundary configurations are investigated.
The pressure in the z direction is applied on the
edge of the plate and stiffener. The pressure value of
the plate is Pp , and then the pressure value on stiffener
is Ps , Ps = Pp tw /tp , where tw and tp are the thickness
of web and plate. Table 2 shows all configurations
with different boundary conditions and models. The
coordinate systems is shown in Figure 5.
3.3
Geometrical initial imperfections
The initial imperfections decrease the stiffness and

ultimate strength of plates. The initial imperfections
significantly affect the ultimate strength of stiffened
panel and should be accounted for. Imperfections
are caused during a complex fabrication process and
are subject to significant uncertainty related to the
magnitude and spatial variation. To model them the
417
Figure 5. The FE model.

Figure 6. Ultimate strength of three bays with C1.
most accurate method is to use the data of realistic

measurements, which is not always available.
Kmiecik (1971) considered the initial deflections
as the superimposition of the Fourier components for
the first time. The behavior of plates subjected to
buckling loads depends on the shape of their initial
deflections (Kmiecik 1995). So the equivalent initial
imperfections are used, which are considered as the
superimposition of the Fourier components. In most of
the initial theoretical studies initial deflection assumed
to have the same shape as the buckling mode. The following three types of initial deflection are accounted
for (Paik 2009):
4 THE RESULTS OF THE FEM ANALYSIS

The total force at the boundary/compressed edge is
determined by the following equation:
where N is the number of nodes at boundary; Ri is the

reaction force on each node.
The stress at the boundary is that total reaction force
(R) is divided by the sectional area of the stiffened
panel (At ):
Hungry horse mode initial deflection of local plate

panels
Column-type initial deflection of stiffeners
Side-ways initial deflection of stiffeners due to

angular rotation about panel-stiffener intersection
line
where a and b are length and width of the plate.

To get the initial imperfection in FE analysis, the
shapes of initial imperfections are divided into plate
initial deflection, column-type initial distortion of
stiffeners and sideways initial distortion of stiffeners.
Firstly, linear buckling analysis is performed for the
target stiffened panel and find out the related buckling modes of plating and stiffener. The geometry
properties, for example the thickness of plates and
stiffener, are changed to decouple those deformations
of interest from lower eigenmodes and get desired
shapes for plate out-of-plane and stiffener out-of-plane
deformations. The three types of initial distortions are
superimposed altogether FE model.
These calculations must be performed for each step

of the non-linear analyses.
Figure 6 shows the ultimate strength of the three
bays models with C1 boundary condition. From the
results of 3 bays-C1 shown, increasing the width of
panel, namely number of stiffeners or a length of the
frame, decreases the ultimate strength of the stiffened
panels. Because the stiffeners are supported on the
frames, not like are fixed in the lateral displacement.
The frame is flexible which would cause the displacement in the z direction of the frame increases as the
number of stiffeners, namely the length of the frame,
is increasing, as shown in Figure 7 and 8. So when
designing the panel specimen, the dimension of the
frame should be considered carefully to make sure that
the stiffness of the frame can support the stiffener of
panel. In the other words, the frame should be included
in the ultimate strength simulation of panel instead of
boundary condition when the width of panel is very
large or the stiffness of the frame is not big enough.
The 3 bay-C2 configuration is designed to understand the effect of the panel width on the ultimate
strength of the stiffened panel. To ignore the influence
of frame stiffness and to focus on the effectiveness at
the lateral plating edges, the z direction displacement
is constrained at the intersection nodes of the plate and
the frame for 3 bay-C2 configuration.
The ultimate strength increases slowly with increasing width of the panel in the 3 bay-C2, shown in
Figure 9, because the lateral edges of the panels are
418
Figure 7. Displacement at the ultimate limit state for C1.
Figure 9. Ultimate strength of three bays with C2.
Figure 8. Von Mises stress at the ultimate limit state for C1.
Figure 10. C2 divided by C1 for three bays.
totally free to move out-of-plane and to rotate. This

means that larger panels should be less affected by
the lack of effectiveness at the lateral plating edges
during buckling. In fact, the percentage of the total
cross-section area with reduced effectiveness due to
unsupported lateral edges is lower on the wide panels than in the narrow ones and thus, the expected
ultimate load is higher for the wider panels. But in
this circumstance, the effectiveness affects slightly the
ultimate strength. The ultimate strength of 3 bay-C2 is
commonly bigger than 3 bay-C1.
Figure 10 shows the ratio of ultimate strength
between C2 and C1. It is observed that the mean value
is 1.14 and standard deviation is 0.23. So it would be
too optimistic if the intersection nodes between frame
and plate is supported for z direction displacement.
The more the width of the panel increases, the bigger the ratio become between C2 and C1. When the
width of panel is 600 mm, the biggest ratio is 1.28.
When the width of the panel increases to 750 mm, the
biggest ratio is 1.91. Because the frame is flexible in
the realistic structure, there is displacement in the z
direction when the stiffened panel collapses. The stiffness of the frame would affect the ultimate strength.
This means that the frame is also important in simulation and should be included in FE analysis when the
width of panel is very large or the frame is not strong
enough.
To remove the influence by the lack of effectiveness at the lateral plating edges during buckling and
investigate the difference between the clamped and
simply supported boundary condition, the symmetric
boundary condition is applied on the lateral edges of
the panels for C3 and C4 configuration with clamped

and simply supported boundary condition at the end
of loading edge. This can avoid the lateral edges to
move out-of-plane and to rotate. The mean value of
the ratio between 3 bays C4 and C3 is 0.98. This illustrates that the boundary condition at the end of loading
edge affects slightly on the ultimate strength when the
symmetric boundary conditions are applied on the lateral side edges. Because the middle bays collapse in
the both boundary conditions, and then the boundary
condition at the end of loading edge only affect slightly
on the middle bays in this circumstance.
Due to the symmetric boundary condition is applied
on the lateral edges of the panels which eliminate
the influence of the effectiveness, the magnitude of the
ultimate strength is almost the same with width of the
panel increasing for C3 and C4, as shown in Figure 11,
Figure 12 and Table 3. That means that the width of
stiffened panel wouldnt affect the ultimate strength
of the stiffened panel in this circumstance. So the A
series stiffened panel with C3 and C4 for three bays can
get reasonable result when the symmetric boundary
condition is applied on the lateral edges of the panels.
The following calculations will focus on influence
of the geometry of the model on ultimate strength,
and thus the A and B series are further studied.
The results are almost the same between the A and
B series, when the symmetric boundary condition is
applied on the AB and A1 B1 edges. The mean value of
ratio between three and two bays for C5/C3 and C6/C4
are 1.09 and 1.11 respectively, namely the result of
1/2 + 1 + 1/2 bays bigger than three bays with same
419
Figure 11. Ultimate strength for three bays with C3.
Figure 13. The ultimate strength of FEM analysis with

different configurations.
Figure 12. Ultimate strength for three bays with C4.
Figure 14. Von Mises stress distributions of the stiffened

panel at the ultimate limit state for 3 bay-C1.
Table 3. The ultimate strength of FE analyses for three bays

(MPa)
Case
C1
C2
C3
C4
FS4-A
FS4-B
FS4-C
FS4-D
FS4-E
BS4-A
BS4-B
BS4-C
BS4-D
BS4-E
LS4-A
LS4-B
LS4-C
LS4-D
LS4-E
US4-A
US4-B
US4-C
US4-D
US4-E
217
214
206
187
148
318
313
294
263
314
426
451
464
423
358
318
358
291
283
195
219
221
223
223
222
321
324
326
327
328
429
446
453
455
453
330
331
334
363
373
230
232
232
231
189
363
359
354
361
360
449
453
453
455
454
358
356
373
376
370
224
224
224
224
224
337
337
356
356
357
422
424
425
425
425
368
375
369
370
372
boundary condition. The biggest ratio is 1.25 and 1.37

for clamped and simply supported boundary conditions. This difference between the between three and
1/2 + 1 + 1/2 bays models is caused by the difference
stiffness of the plate and the stiffeners in the side bay.
The length of middle bay for the 1/2 + 1 + 1/2 bays
model is shorter than three bays model. This means the

middle bay have bigger stiffness and stronger boundary condition by side bay for the 1/2 + 1 + 1/2 bays
model than three bays model.
Figure 13 presents a comparison of the ultimate strength with different configurations, including
clamped and simply supported boundary conditions at
the end of loading edge, and different models. The ultimate strength of the 1 bay-C9 is the biggest and the
3 bay-BC1 is the smallest in all configurations. The
mean values of the ratio of C4/C3, C6/C5 and C8/C7
are 0.98, 0.99 and 0.98 respectively. This indicates that
the clamped or simply supported boundary conditions
on longitudinal edges have only a minor influence
on panel strength, but when the edges are clamped
condition the panel strength increases slightly. The
ultimate strength of the LS series is the biggest in
all the panels. The results of ultimate strength are very
different which depends not only on the geometric
model but also on the boundary conditions. It is necessary to investigate what kind of configuration is
more reasonable to calculate the collapse behaviour
of stiffened panels.
The collapse modes of these stiffened panels are
also different between the different geometric models and boundary conditions as shown in Figure 14
to Figure 22. Only the collapse modes of LS series
are shown in this paper. The collapse is similar independently of whether or not the intersection nodes
between plate and frame are supported for z direction
displacement in present setting, shown in Figure 14
420



and Figure 15. It shows that the symmetric boundary condition at the lateral edges of the panels affects
not only the ultimate strengths but also the collapse
mode indicated in Figure 15 and Figure 16, because
the constraint at the lateral edges changes the stress
distribution of the edge plate. The collapse modes in
the middle bays of three and 1/2+1+1/2 bays are also
different due to the different length of the side bay,
namely different stiffness for middle bay, shown in
Figure 1619.
The asymmetric collapse modes occur for 1+1 bays
with clamped and simply supported boundary condition in Figures 20 and 21, because the clamped or
simply supported boundary condition would cause the
asymmetric stress distribution for the both bays which
increases the uncertainty of the results. If the 1+1 bays
model is used, the periodic or symmetric boundary
condition should be adopted to avoid the uncertainty.
The 3 and 1/2+1+1/2 bays stiffened panels are more

panel at the ultimate limit state for 2 bay1-C5.



robust than 1+1 bays in this circumstance. The collapse mode of 1 bay model is different from the other
models, and the 1 bay model can not consider the
interaction between adjacent panels and is not recommended in ultimate strength simulation of stiffened
panels.
421
This work contributes to the activities of

MARSTRUCT VIRTUAL INSTITUTE, (www.
marstruct-vi.com) in particular its Technical Subcommittee 2.3 on Ultimate Strength.
REFERENCES

CONCLUSIONS
The ANSYS is used to analyze the stress distributions

and deformations of nine geometric configurations of
stiffened panels, including 3 bays, two (1/2+1+1/2)
bays, 1+1 bays and 3 bay with different boundary
conditions.
Because the frame is flexible in the realistic structure, there is displacement in the z direction when the
stiffened panel collapses. The stiffness of the frame
will affect the ultimate strength. Hence, the frame
should be included in the FE analysis, specially when
the width of panel is very large or the frame is not
strong enough according to the ratio of C2/C1.
The ultimate strength with clamped boundary condition are slightly bigger than with simply supported
boundary condition. The loading rate of average stressshortening for 1bay-C9 is the biggest in all cases.
The one bay model can not account for the interference between adjacent panels and is not recommended
in ultimate strength analysis of stiffened panel. The
asymmetric collapse modes occur in 1+1 bays. This
phenomenon will increase the uncertainty of the result
and the 1+1 bays is not recommended with clamped
or simply boundary condition. The 1+1 bays model
with the periodic boundary condition could avoid
the asymmetrical collapse. But the periodic boundary
condition is difficult to apply in the tests.
Hence, the 3 and 1/2+1+1/2 bays stiffened panels
are better than 1+1 bays for tests, which allow to have
more realistic results by avoiding boundary conditions
problems for the central plates related to eccentricity of
load and to include the interference between adjacent
panels.
ACKNOWLEDGEMENTS
The first author has been financed by the Portuguese Foundation for Science and Technology
(Fundao para a Cincia eTecnologia), under contract
SFRH/BD/65120/2009.
ABS, 2006. Rules for building and classing, steel vessels.

Fujikubo, M., Yao T., Khedmati M., Harada M., Yanagihara
D., 2005. Estimation of ultimate strength of continuous stiffened panel under combined transverse thrust
and lateral pressure Part 2: Continuous stiffened panel,
Marine Structures (18), pp. 411427.
Gordo, J.M., Guedes Soares C., 2007. Experimental evaluation of the behavior of a mild steel box girder under
bending moment. In: Guedes Soares C, Das, P.K., (Eds).
Advancements in Marine Structures. Taylor and Francis,
pp. 377383.
Gordo, J.M., Guedes Soares, C., 2008. Compressive tests
on short continuous panels, Marine Structures (21),
pp. 113137.
IACS CSR (2006), Common structural rules for oil tankers.
Kmiecik, M., 1971. Behaviour of axially loaded simply supported long rectangular plates having initial deformations,
Report No. R84, Ship Research Institute, Trondheim.
Kmiecik M., Jastrzebski T., Kuzniar J., 1995. Statistics
of Ship Plating Distortions, Marine Structures (8),
pp. 119132.
Lus, R. M., Guedes Soares C., Nikolov P. I., 2008a. Collapse
Strength of Longitudinal Plate Assemblies with Dimple Imperfections. Ships and Offshore Structures. 3(4),
pp. 359370.
Lus, R. M., Witkowska M., Guedes Soares C., 2008b.
Ultimate Strength of Transverse Plate Assemblies Under
Uniaxial Loads. Journal of Offshore Mechanics and
Arctic Engineering 130(2): 021011-1021011-7.
Paik, J.K., 2009. Nonlinear finite element method models for ultimate strength analysis of steel stiffened-plate
structures under combined biaxial compression and lateral pressure actions-Part II: Stiffened panels,Thin-Walled
Structures (47), pp. 9981007.
Paik, J., Branner K., Choo J., Czujko J., Fujikubo M., Gordo
J.M., Parmentier G., Iaccarino R., ONeil S., Pasqualino
I., Wang D., Wang X. and Zhang S., Committee III.1
Ultimate Strength, 2009, In: 17th International Ship and
Offshore Structures Congress (ISSC2009), Vol. 1. C. Jang
and S. Hong, editors. Seul, South Korea: University of
Seoul, pp. 375475.
Smith, C., 1979. Compressive strength of welded steel ship
grillages. Trans RINA (117), pp. 325359.
Tanaka, Y., Endo H., 1988. Ultimate strength of stiffened
plates with their stiffeners locally buckled in compression.
J Soc Naval Archit Jpn (164), pp. 456467.
Xu, M.C., Guedes Soares C., 2011. Comparison of numerical
results with experiments on ultimate strength of stocky
stiffened panels, Maritime Engineering and Technology,
Guedes Soares, C. et al. (Eds.), Taylor & Francis Group,
in this volume.
Zhang, S.M., Khan I., 2009. Buckling and ultimate capability
of plates and stiffened panels in axial compression, Marine
Structures (22), pp. 791808.
422
Risk and reliability

Risk assessment to the approach and berth of LNG vessels

at the FLNG systems
P. Anto & C. Guedes Soares
ABSTRACT: In the present study a risk assessment is performed for the approach manoeuvres of an LNG
(Liquefied Natural Gas) vessel to berth and moor at a FLNG (Floating Liquefied Natural Gas) platform. In
order to fulfil that objective, an identification of the procedures of berthing and mooring in good and bad
weather is performed. Based on this analysis several hazards are identified. These hazards are mainly a result
of human errors or equipment failures and are then quantified based on expert opinion in terms of frequency of
occurrence and potential consequences. These procedures correspond to the first two steps of the IMO Formal
Safety Assessment (FSA). Based on the obtained results it is possible to identify situations of higher risk based
on a predefined risk acceptance criterion.
INTRODUCTION
The current world fleet of LNG carriers is comparably

small in relation to other ship types, but it has been
increasing steadily in recent years. The present LNG
fleet contains more than 183 ships with a substantial
number of ships under construction or ordered. Most
of these ships are of large size with an average size of
almost 120,000 cubic meters while the ordered ships
present even higher average (150,000 cubic meters).
With the increase on LNG demand several practical
solutions have been proposed in order to optimize the
logistic chain flow and, at the same time, lower the
costs and reduce the risks associated with the transport, process and storage of LNG. In this view the
development of the FLNG platforms are considered
today a practical and viable solution to some of the
above challenges, since these structures are more than
one ship, as they extract, process and store natural gas,
i.e., acts as a floating refinery.
When performing a risk analysis and assessment,
several main risks can be considered, i.e., risk to
the people and crews, to the environment, and to
the economy and property. The primary risks are
mainly related to operational risks as it is the cases
of the berthing, loading/unloading (keeping the position of both LNG and FLNG) and unberthing phases.
In the present study, the focus is given just to the
first and third mentioned phases, i.e., to berthing and
unberthing procedures. The environmental issues were
left out of the study because LNG doesnt represent
any notable hazard to the marine environment. However, risks for people (crews of both ships performing
mooring/unmooring procedures) and to the property
(both ships) are considered in the analysis. Although

the potential of LNG carriers for large consequence
accidents is high, in case of evaporation or pool fires
events, the historical data shows that LNG ships are
one of the safest in maritime transportation.
Understanding the hazards associated to these particular phases allows the development of criteria for the
design of an FLNG system (FPSO, shuttle tanker, tug,
offloading mooring) and statistics for the estimation of
the safety and economics of its operation (production
and offloading). The various stages in the offloading
process involve high risks and therefore an overall risk
assessment is required to ensure that the overall risk
posed by the activity is within tolerable levels.
2
HISTORIC LNG ACCIDENT BACKGROUND
In recent studies IMOs FSA analysis (IMO, 2002) was

applied to LNG vessels (Vanem et al., 2007, 2008).
In these studies a literature survey on the accident
record of LNG shipping reveals 182 incidents with or
without LNG spillage involving LNG carriers of more
than 6,000 GRT. These accidents range between the
period of 1964-2005, i.e., to 41 years of LNG maritime
transportation. Very few fatalities have been reported
as a result of LNG carriers in operation during this
period. In fact only two incidents resulted in a fatality,
one where a terminal worker died and one incident
where a fatality among the crew on a bulk carrier
that collided with a LNG carrier occurred. According to one study, this fatality frequency would result
in a potential loss of lives (PLL) of 7.05 104 per
shipyear (Vanem et al., 2008).
425
seen that it is dominated by the collision, grounding

and contact scenarios. However, fire and explosion are
dominating the low consequence risk contributions in
the order of one fatality.
Table 1. Distribution of historic LNG accidents on categories (Vanem et al., 2007).
Accident category
Collision
Grounding
Contact
Fire and explosion
Equipment and machinery
failure
Heavy weather
Events while loading/
unloading cargo
Failure of cargo
containment system
Total
Accidents
(#)
Frequency
(per shipyear)
19
8
8
10
55
6.7 103
2.8 103
2.8 103
3.5 103
1.9 102
9
22
3.2 103
7.8 103
27
9.5 103
158
5.6 102
BERTHING PROCEDURES TO THE FLNG
The berthing procedures of a LNG into a FLNG were

derived from an extract developed by an Energy major
following a series of full mission simulations.
In it a set of parameters or requirements for safe
berthing were identified. These requirements range
from weather conditions, FLNG and LNG equipment or tug(s). The classification of these parameters
allowed the identification of potential failure modes
associated to each of the procedures involved in
the berthing/unberthing of the LNG. Therefore, the
variables and requirements present and necessary for
safe manoeuvres can be described as the followed:
LNG characteristics
Displacement; Main dimensions (Lbp, B, etc.);

Engine Power;Type of propulsion system;Thrusters;
Existence of berthing aid systems (ex: DGPS, etc.);
Number of mooring lines.
LNG design enhancements
Figure 1. FN-curve for crew, broken down on accident

types (Vanem et al., 2008).
However, according to consultation with the LNG

industry representative society, SIGTTO (the Society
of International Gas Tanker and Terminal Operators),
comprising membership of 181 LNG Companies, has
recorded only 85 LNG related incidents from 1965 to
the present day. Therefore, the potential loss of lives
frequency per shipyear presented above can be considered the upper limit and used to obtain conservative
risk measures.
Table 1 presents a distribution of accidents for the
whole history of LNG shipping per period of years
(IMO, 2007) considering the accumulated shipyears.
In it one may see that loading/unloading and collision events present higher frequencies, as expected.
All of the accident events relevant for the present
study present frequencies are within ALARP considering the individual risk acceptance criteria for LNG
crew (Vanem et al., 2008). In general, the accident
frequencies are found to be somewhat lower for LNG
carriers than for these other types of vessels, but this
was expected considering the high focus on safety
on these ships and the generally high competence of
LNG crews (Figure 1).
The FN-curve shows the contribution from each of
the main accidents to the overall risk, and it can be
45 degree operating or other high efficiency

rudder; Bow thrusters of at least 35 tons force;
Strong points and fairleads for connection of tugs;
HMPE (High Modulus PolyEthylene) mooring lines;
Dedicated messenger retrieval winches (aft and
forward); Monitoring systems for weather data and
mooring line loads monitoring; Pneumatic line
throwing device.
FLNG requirements
Mooring positions on the starboard side; 18 mooring leads and QR (quick release) hooks; Stern
thrusters; Automatic heading control; Availability of a berthing master; Pneumatic gun lines; a
mooring crew of at least 3 persons forward and
three persons aft to handle mooring lines, with
an additional Supervisor; Visible berthing markers;
Leading lights.
Mooring equipment requirements
426
6 pneumatic fender pieces of 4.5 m 9 m with

50 kPa at rated 60% compression; LNG mooring
lines with 42 mm diameter and minimum breaking load of 125 tonnes; Wires/HMPE lines fitted
with 22 m long nylon or polyester pennants (92 mm
diameter with minimum breaking load greater than
125% of the mooring line).
Figure 2. Berthing with tugs alongside FLNG benign

conditions only.
Figure 3. Berthing with tugs on long line in a bad weather

condition.
Weather variables
Wind force and direction; Sea: wave height and

period; Swell: wave height and period; Currents;
Visibility.
Weather requirements
LNG: Weather head on

Wind 15 m/s
Current 0.8 m/s
Sea: wave height 2.5 m; period 6.5 s
Swell: wave height 0.8 m; period 12 s
LNG: Weather 90
Wind 7.5 m/s
Current 0.8 m/s
Sea: wave height 2.8 m; period 6.5 s
Swell: wave height 0.8 m; period 12 s
Tug: Long line towing
wave height <2.5 m and period 6.5 s
wave height <2.8 m and period 9 s
Tug: Push-Pull alongside
wave height <1 m
Visibility
>0.5 miles
3.1 Berthing procedures with Good weather

All the following consider procedures are based on
an extract from berthing procedures based upon simulations conducted by an energy major for an FLNG.
Pilot transfer and start of berthing approach
Tug variables and requirements
Tugs manoeuvrability; Stern azimuth tugs; Auto

tensioning, heave compensated winch; Effective tug
bollard pull of 90t; Bowthruster of suitable power.
Situation variables
Experience of the berthing master; LNG and

FLNG hydrodynamic interaction; Experience of the
bridge and deck personnel; Existence of Portable
Pilot devices and/or Doppler docking aid; Depth
at the berthing location.
After identifying the variables and the berthing/

unberthing requirements one may proceed with
description of the different phases leading to the safe
berthing and unberthing of the LNG into the FLNG
in two given conditions: good weather and bad
weather. Good weather is defined here as an environment in which tugs can push the LNG towards the
FLNG, i.e., wave height <1 m whereas a Bad weather
is defined here as an environment in which tugs must
operate on long tow line, i.e., wave height >1 m.
These differences can better observe in Figures 2 and
3 where the berthing procedures and the active role of
the tugs are presented.
Identification of weather conditions (berthing procedures should start when weather requirements
mentioned above are met);
Identification of the FLNG heading;
Agreement of the location for initiate berthing
(depending on weather conditions window);
LNG arrival at the designated location approximately 2 miles astern from the FLNG location
and 10 of the FLNG centreline (location may vary
from situation to situation depending on the first
two points);
Transfer of the berthing master (pilot) from the
FLNG to the LNG after arrival of latest to the agreed
location;
LNG provides lee to permit the pilot to board safely;
Bow and stern tugs are connected as the LNG
commences the approach, prior to the distance to
FLNG of 1 mile;
FLNG adopts natural heading relative to the prevailing wind, current and wave forces;
Berthing Master may request a change of heading
of the FLNG to optimize the relative environmental
forces to assist with berthing the LNG;
Relative wind should be within 30 degrees on the
bow of the LNG;
FLNG heading is maintained by the automatic
heading control system linked to the thrusters.
Approach
427
For Good weather conditions berthing with tugs

pushing alongside may be undertaken (wave/swell
height of less than approximately 1 meter);
Manipulation of the heading of the FLNG relative to the environment, to achieve a sheltered area
alongside the LNG for the tugs;
LNG approached the FLNG from a direction on the
starboard quarter;
LNG stops at a lateral distance from the FLNG of
around 100150 m.
Berthing
Heading of the FLNG stabilised with thrusters and

heading control;
Tugs start to push the LNG alongside;
Tugs control approach rate;
The Berthing master and the LNG captain decide
what is the best distance in order to send the mooring
lines.
Mooring
LNG mooring lines may be passed to the FLNG

when the distance between the vessels has closed
sufficiently for a pneumatic line throwing device to
be used (<100 m);
If tugs are able to push alongside, and hold the LNG
against the fenders once fender contact is made,
it may be more efficient to send lines at a lesser
distance;
Tugs remain connected on tow lines until the mooring operation is completed.
3.2
Berthing
Berthing procedures with bad weather
Pilot transfer and start of berthing approach
Identification of weather conditions (berthing procedures should start when weather requirements
mentioned above are met);
Identification of the FLNG heading;
Agreement of the location for initiate berthing
(depending on weather conditions window);
LNG arrival at the designated location approximately 2 miles astern from the FLNG location and
10 of the FLNG centerline (location may vary from
situation to situation depending on the first two
points);
Transfer of the berthing master (pilot) from the
FLNG to the LNG after arrival of latest to the agreed
location;
LNG provides lee to permit the pilot to board safely;
Bow and stern tugs are connected as the LNG
commences the approach, prior to the distance to
FLNG of 1 mile;
FLNG adopts natural heading relative to the prevailing wind, current and wave forces;
Berthing Master may request a change of heading
of the FLNG to optimise the relative environmental
forces to assist with berthing the LNG;
Relative wind should be within 30 degrees on the
bow of the LNG;
FLNG heading is maintained by the automatic
heading control system linked to the thrusters
Tugs maintained away from the LNG hull on long
line;
Tugs are secured through centre leads bow and stern
for optimum operation.
Approach
Approach is made with an angle of around 10

degrees to the centerline of the FLNG;
Pilot must guaranty that the approach is made outside of a line extended from the starboard side of
the FLNG stern;
If line crossed than pilot must suspend the approach;
Thrusters, in auto heading control, maintain the
FLNG heading;
LNG maintains engine ahead, with rudder to cause
a lateral movement towards the FLNG;
The stern tug works as a hold back tug;
Bow tug tows towards FLNG;
Tugs control approach rate;
Tugs maintain sufficient reserve power if necessary;
Final approach is at a speed with which the LNG
heading can be controlled by a combination of tug
assist and main engines and rudder.
LNG ground speed shouldnt be above 1.5 knots

when passing the stern of the FLNG;
LNG should be stopped under control at manifold
alignment;
LNG arrive at a position approximately 100200
meters of the FLNG;
LNG is approximated parallel to the FLNG with
manifolds and loading arms aligned;
Longitudinal ground velocity of the LNG should
be 0;
LNG is then brought slowly alongside using tugs
and propeller and rudder thrust.
Mooring
LNG mooring lines may be passed to the FLNG at

an adequate close distance;
Heaving on the stern mooring lines should be
avoided;
FLNG heading control will be required to prevent
the sterns coming together in excess of the design
berthing speed;
Maximum berthing impact should not exceed
0.25 m/s;
Final positioning when alongside the fenders may
be achieved using the LNG mooring lines and tugs.
There are a set of procedures to be met independently of the good/bad weather conditions which is the
case of unmooring procedures since they are similar
in both situations.
3.3
Identification of potential failure modes and

their quantification
Based on the description of the procedures mentioned

above one may identify potential failures modes.
These failure modes can be of equipment failure nature
or simply human errors while performing a given task.
For the latest common errors as done too late, too
soon, too fast can have serious implications in
certain situations in the present analysis.
428
Table 2.
Definition of probability index (stvik et al., 2005).
PI
Probability
Definition
P (per ship year)
8
7
6
5
Very frequent
Frequent
Probable
Reasonably probable
100
10
1
0.1
Little probable
Remote
2
1
Very remote
Extremely remote
Likely to happen once or twice a week on one slip

Likely to occur once per month on one slip
Likely to occur once per year on one slip
Likely to occur once per year in a fleet of 10 ships, i.e. likely to occur
a few times during a ships life
in the total of a ships life
in the total of several similar ship
Likely to occur once per year in a fleet of 10000 ships
Likely to occur once in the lifetime (20 years) of a world fleet of 5000 ships
Table 3.
0.01
0.001
0.0001
0.00001
Definition of consequence index (stvik et al., 2005).
Human
safety
Cargo/
Monetary
losses
CI
Consequence
Minor
Single or
minor
injuries
Negligible release
negligible pollution
no acute environmental
or public health impact
30.000
US$
Significant
Multiple or
severe
injuries
300.000
US$
Severe
Catastrophic
Single
fatality or
multiple
severe
injuries
Multiple
fatalities
Minor release minimal

acute environmental or
public health impact
small, but defectable
environmental
consequences
Major release effects
on recipients short
term disruption
of the ecosystem
30 mil.
US$
Disastrous
Severe pollution
medium-term effect
on recipients mediumterm disruption
of the ecosystem
Uncontrolled pollution
long-term effect on
recipients long-term
disruption of the
ecosystem
Large
number of
fatalities
Environment related
For the quantification of the identified failure

modes a quantitative approach must be used as no
available statistics exists. Therefore the quantification
had to be based on expert opinion of masters or LNG
specialize personnel. For the quantification a risk criteria had also to be applied. For the effect a similar criterion used in a recent FSA study for LNG tankers was
used (stvik et al., 2005). Therefore for the probability
a criterion as presented in Table 2 can be used.
The risk assessment and consequence matrix below
is completed on the basis that the FLNG will receive
between 50 and 100 ships per year.
3 mil.
US$
300 mil.
US$
3rd party
assets
Equivalent
fatalities
Local equipment
damage (repair on
board possible,
downtime
negligible)
Non-severe ship
damage (port
stay required,
downtime
1 day)
Minor
damage
0.01
Significant
damage
0.1
Severe damage
(yard repair
required,
downtime
<1 week)
Total loss (of,
e.g., a medium
size merchant
ship)
Severe
damage
vicinity of
ship
Extensive
damage
10
Total loss (of,

e.g., a large
merchant
ship)
Major public
interest
100
Effect on ship
In the following tables an extract of potential modes

of failure are presented. These are just strictly related
to events that can occur during approach phases. It is
not considered other events as it is the case of occupational accidents (ex: fall of the pilot while boarding
the LNG, etc.).
The mitigating controls are listed and considered to
be implemented by a reputable operator who complies
with all required regulatory controls, industry guidelines and operating under an implemented HSSE
(Health, Safety, Security, and Environment) management system. Failures are mitigated with multi-level
429
Table 4. List of the potential failure modes in bad weather conditions.
Phase
Potential Failure Modes
Mitigating Controls
Frequency HS E
CML ES
3rd P
Pilot Transfer and

Start of Berthing
Approach bad
weather conditions
Failure of the systems (ex: pilot

berthing aids) to provide accurate
weather data to the crew of the LNG
Multi-source weather information from LNG as well as FLNG and shore

based systems. Berthing decision based on forecast and actual weather
The pilot fails to board the LNG delay

>24 hours
Tugs fail to connect to the LNG delay
>24 hours
Severe contact of the tugs to the LNG
Procedural controls, weather limitations, tug design, LNG provides lee
Procedural controls, weather limits on tug push mode, long line towing in
seas >1 m, tug fendering, strengthened tug push points on LNG hull
Tug design and power, tug crew training, Pilot training, adjust FLNG heading,
abort berthing, delay berthing until weather moderates
Tug design and power, tug crew training, Pilot training, abort berthing, delay
berthing until weather moderates
Procedural controls, berthing aids, LNG Master and FLNG PIC intervention,
abort berthing, assess weather conditions before making 2nd attempt, comply
with weather limitations
Multiple thrusters for redundancy, dual control systems, procedural controls
for berthing without thrusters Sapring Philosophy
Pilot training, berthing aids (speed of approach) LNG Master intervention,
procedural controls, stop engines, abort berthing, tug assistance
Pilot training, LNG crew training, procedural controls, weather limits for
berthing, excess capacity of tug power, abort/emergency procedures, FLNG
thrusters ability to move stern away from LNG, fenders on FLNG
berthing, weather forecasting and observation, excess capacity of tug power,
abort/emergency procedures, FLNG thrusters ability to move stern away
from LNG, fenders on FLNG
Tug design and power, tug crew training, Pilot training, abort berthing,
assess weather conditions before making 2nd attempt, comply with weather
limitations. Abort Berthing
Multiple line throwing devices available, maintenance and inspection regime,

operator training
Speed of approach information for Pilot, Pilot training, berthing procedures,
abort points, tug design and power, tug crew training, weather limitations for
berthing, abort berthing
LNG berthing pilot and captain cant

maintain the wind within 30 of the bow
Approach
430
Consequence
Bow tug does not manage to

pull adequately the LNG
Pilot fails to suspend the approach as the
LNG crosses inside of approved approach
line of FLNG
Failure of the thrusters of the FLNG
LNG pilot applies too much ahead power
Berthing
Collision of the LNG to the FLNG due

to uncontrolled approach
Hard contact of the tugs to the LNG while
pulling it
weather changes (wind strength, etc.)
Tugs cant control approach rate
Mooring
Failure of the pneumatic line throwing

device
pulling it into position
Energy of contact of the LNG above
limitations
Table 5. List of the potential failure modes in good weather conditions.
Phase
Potential Failure Modes
Mitigating Controls
Frequency HS E
CML ES
3rd P
Pilot Transfer and

Start of Berthing
Approach good
weather conditions
Failure of the systems (ex: pilot berthing

aids) to provide accurate weather data to
the crew of the LNG
The pilot fails to board the LNG delay
>24 hours
Tugs fail to connect to the LNG delay
>24 hours
Failure of the automatic control system of
the FLNG of maintaining heading
Severe contact of the tugs to the LNG
Multi-source weather information from LNG as well as FLNG and shore

based systems. Berthing decision is based on forecast and actual weather
Dual auto-heading control, manual heading control function. Sparing

Philosophy n + 1
Tugs dont manage to push adequately the

LNG
FLNG doesnt manage to change heading
to achieve sheltered area/correct aspect to
weather
LNG fails to approach the FLNG from
starboard on correct approach line
Tugs dont manage to stop the LNG at a
lateral distance of 100 m of the FLNG
Tug design and power, tug crew training, Pilot training, abort berthing, change
to long line towing, delay berthing until weather moderates
Change heading with tug(s) and determine if thrusters can maintain. Delay
berthing until weather moderates
Adjust approach track, abort berthing, assess weather conditions before

making 2nd attempt, and comply with weather limitations
Tug design and power, tug crew training, Pilot training, abort berthing,
limitations
Multiple thrusters for redundancy, dual control systems, procedural controls

for berthing without thrusters. Berthing aborted
Tug design and power, tug crew training , Pilot training, abort berthing,
limitations.
berthing, excess capacity of tug power, abort/emergency procedures, FLNG
thrusters ability to move stern away from LNG, fenders on FLNG
berthing, weather forecasting and observation, excess capacity of tug power,
abort/emergency procedures, FLNG thrusters ability to move stern away
from LNG, fenders on FLNG
Multiple line throwing devices available, maintenance and inspection regime,
operator training
Speed of approach information for Pilot, Pilot training, berthing procedurals,
abort points, tug design and power, tug crew training, weather limitations
for berthing, abort berthing, fender capacities
Approach
431
Consequence
Berthing
Failure of the thrusters of the FLNG

pushing it
Tugs cant control approach rate
Collision of the LNG to the FLNG due to
uncontrolled approach
weather changes (wind strength, etc.)
Mooring
Failure of the pneumatic line throwing

device
pushing it against the fenders
Energy of contact of the LNG above
limitations
Figure 4. Results of the QRA of the berthing of the LNG to the FLNG.
controls, i.e., multiple barriers per risk, such that

the failure of one control does not mean the hazard is released. Consequence is assessed considering
activation of controls.
From the results of the quantification by expert
opinion of several LNG operators for each of failure modes identified in the tables presented above
it was possible to draw some conclusions. Most
of the hazards identified are within the acceptable
region or in ALARP, for the defined risk acceptance
criteria. The hazards which present higher risk levels are related to a specific delay of the operation
(due to pilot fails to board or tugs fail to connect
to the LNG) which have an high economic impact
associated.
CONCLUSIONS
In the present study a high level risk assessment of

the procedures of approaching and berthing of a LNG
vessel to a FLNG platform is performed. A set of 69
main hazards related to these two procedures are identified based on the descriptions of the operational tasks
involved. In order to quantify the identified hazards
expert opinion was used with a criterion similar to the
one used in other LNG studies. The overall results of
the risk assessment lead to the conclusion that most of
the hazards identified are within the acceptable region
or in ALARP and the ones which present higher risk
levels are directly related to delays in the operations
with direct economic implications.
ACKNOWLEDGEMENTS
The work presented was performed within the project
SAFEOFFLOAD, funded partially by the European
Commission under the contract number TST4-CT2005-012560.
The first author acknowledges the financial support of the Portuguese Foundation for Science and
Technology under the contract BD/31272/2006. The
authors would like to thank to Mr. Phil Jameson, from
Shell Shipping Technology, for his support on the tasks
descriptions related to vessels approach and berthing.
REFERENCES
IMO, 2002, Guidelines for Formal Safety Assessment
(FSA) for Use in the IMO Rule-making Process.
IMO, 2007. Formal Safety Assessment Liquefied Natural
Gas (LNG) Carriers, Details of the Formal Safety Assessment. MSC 83/INF.
Vanem, E., Anto, P., Castillo de Comas, F. and Skjong,
R., (2007), Formal Safety Assessment of LNG tankers,
10th International Symposium on Practical Design of
Ships and Other Floating Structures, Basu, R, Belenky, V.,
Wang, G. & Yu, Q. (Eds.), American Bureau of Shipping,
USA, pp. 8290.
Vanem, E., Anto P., stvik, I., and Comas, F., (2008),
Analyzing the risk of LNG carrier operations, Reliability Engineering and System Safety, Vol. 93, Issue 9,
pp. 13281344.
stvik, I., Francisco Del Castillo de Comas and Vanem,
E. (2005), HAZID for LNG Tankers, Task 4.3.1.
SAFEDOR Design, Operation and Regulation for
Safety.
432

Modeling the risk of product spills in LNG tankers

F.B. Natacci, P. Anto & C. Guedes Soares
M.R. Martins
Department of Naval and Ocean Engineering, Universidade de So Paulo, Brazil
ABSTRACT: The growing Liquefied Natural Gas (LNG) maritime transport requires the corresponding construction of new terminals where some basic operations are performed, such as: loading, unloading, storage,
liquefaction and regasification of LNG. The construction of the terminals shall follow strict regulations according to the country of construction which, in most cases, require a licensing procedure that is mainly focused
on safety analysis aspects. This is more relevant when terminals are located near dense populated areas. The
purpose of the present study is to develop the risk model associated with LNG spills during the whole shipping
process, identifying their causes as well as the corresponding operations when spills are detected. Some of
these operations take place in the LNG terminals. The spillage frequencies of occurrence are also quantified.
These activities are inherent part of the safety analysis procedure, employing the fault tree technique. Both, the
cause identification and the frequency quantification are based on data collected earlier (IMO, 2007).
INTRODUCTION
With the growing industrialization in countries such as

China and India, the worldwide oil production was not
sufficient to supply the corresponding energy demand
(Woodward et al., 2010). In this scenery, LNG appears
as one of the alternatives to this growing market. As a
consequence, LNG shipment shall follow this growth,
also requiring new projects of LNG terminals. Figure 1
shows the historical and projected LNG demand by
region (American Gas Foundation, 2008). It can be
observed the increasing growth expected to occur in
the next years.
Safety analysis reports shall be required during
the licensing procedure of these new installations
Figure 1. LNG global demand by region by 2016 (American

Gas Foundation, 2008: pp: 24).
as well as of existing installations undergoing through

renewing projects. The classification societies are
also following the International Maritime Organization directives for the development of formal safety
assessment during ship design (IMO, 2002).
The procedure to elaborate the safety analysis of
any installation requires the development of many
activities related to risk and reliability engineering,
using well known tools. In each phase of the analysis, there are specific tools to be employed. For the
hazard identification phase, there are many tools available such as HAZID (hazard identification), HAZOP
(hazard and operability analysis) and the What If
technique (CCPS, 2008).
Other techniques, such as fault tree analysis and
event tree analysis, are more appropriate for the development of accident scenarios, as well as for the quantification of their associated probability or frequency
of occurrence (Kumamoto & Henley, 2000). For the
quantification of consequences, there are many specific references about consequence analysis that deal
with spill dispersion and evaporation as well as with
possible cascading events such as fires, explosions,
cryogenic effects, asphyxiation, rapid phase transition
and rollover. These references present specific tools
that must be employed in order to calculate the magnitude of possible consequences (CCPS, 1994, 1995a,
1995b, 1996, 1999, 2000).
The present study focuses on the development of a
risk model where the considered undesired event is the
occurrence of LNG spills in tankers during the maritime transport cycle, from the first operation, which is
433
Table 1.
List of the Intermediate and Basic Events of the Risk Model for Gas Spills.
IE Intermediate Events
BE Basic Events
IE1
BE2
BE3
BE4
BE5
BE6
BE7
BE8
BE9
IE10
BE11
BE12
BE13
BE14
BE15
BE16
BE17
IE18
BE19
BE20
BE21
BE22
BE23
BE24
BE25
BE26
IE27
BE28
BE29
BE30
BE31
BE32
BE33
Description
High movements in LNG transfer system
Personnel failing in adjusting the moorings during loading/unloading
External forces from wind, wave and tide
Faults in ballast system or fault operation of ballast system
Lack of crew competence and training
Being struck by passing vessel
Rollover
Fault in sail away procedure
Collision with other ships or facilities
Collision
Steering and propulsion failure
Rudder failure
Severe weather causing vessel to ground/collide
Faults in navigation equipment
Lack of pilot competence
Lack of interaction between vessel and pilot
Collision
Steering and propulsion failure
Rudder failure
Severe weather causing vessel to ground/collide
Faults in navigation equipment
Loss of power in steam turbine/boiler system
Severe sea/weather
Collision with other ships or facilities
High movements in LNG transfer system
Personnel failing in mooring the vessel during loading/unloading
External forces from wind, wave and tide
Faults in ballast system or fault operation of ballast system
Lack of crew competence
Being struck by passing vessel
Rollover
the loading of the tanker at the LNG supply terminal,

until the unloading procedure is completely finished
at the LNG demand terminal. Basically, the fault tree
technique is the tool employed for this development,
mainly because one of the aims was to be able to quantify the different levels of LNG spills. First, a complete
risk model will be presented, considering all possible
causes for the occurrence of LNG spills. Then, a historical analysis of accidents with LNG tankers that
resulted in spillages will be presented. Finally, a new
risk model will be constructed, considering the historical spillage data. Based on these data, the frequencies
of occurrence of different spill sizes will be calculated.
2
RISK MODEL FOR LNG SPILLS IN TANKER

ACCIDENTS
The first step undertaken in the present study is to

identify potential hazards that could eventually lead to
a spillage scenario. Based on the HAZID developed
under the scope of the IMO (2007), all possible causes
for LNG spills in tankers are identified. These hazards
are then aggregated into different operational phases
as their likelihood can differ substantially even if the
same basic event is present in two different operational

phases.
From the Table 1 one may see that although LNG by
itself is already a hazard, the majority of the hazards
identified are generic to all kinds of ships and not LNG
exclusive. In fact only 3 of the identified hazards are
LNG specifically related while the remaining ones are
mainly human error originated.
Based on the identified hazards it is possible to
develop a general risk model for LNG spills in tankers
though fault trees. At this specific point the likelihood of each of the basic events identified is not
yet evaluated. The general fault tree model rather
presents the conditions that can instigate the LNG spill
outcome. The complete risk model is shown in Figure 2, where the causes are related to each different
phase of the LNG transport cycle. The events shown
in the figure are described in Table 1, where IE are
the intermediate events and BE are basic events of
the fault tree. From Figure 2 one can see that there
are 7 branches, which can result in the LNG spills and
with 3 of them being a single, independent basic event,
namely, collision with other ships or facilities (while
maneuvering or arriving in port) and fault in sail away
procedure.
434
Figure 2. Risk model for LNG spills in tankers.
2.1
Historical analysis of tanker accidents

resulting in LNG spills
Despite the fact that LNG carriers have more than 40

years of operational experience, the number of accidents with this type of vessel is rather small when
compared with other ship types. In Vanem et al. (2007)
a survey of historical LNG accidents was performed
which identifies 182 LNG carriersaccidents. It is then

not surprising that the number of LNG spills is small
given the accident sample.
The compilation of the accidents occurred with
LNG tankers that resulted in spills is shown in Table 2.
These specific data were extracted from IMO (2007)
report, which presents a list of known past LNG
435
Table 2.
Historical Analysis of Tanker Accidents with LNG Spills.
Injuries or
Fatalities
1
2
3
4
None
None
None
None
None
None
50 injured
1 dead
None
10
None
11
12
None
None
13
Causes
Year
Ship Name
Activity
Spill Size
C1: Overfilling
C2: Valve leakage
C3: Cargo leakage (no details)
C4: Violent sloshing of LNG caused
electric cabe tray to break loose,
perforating the primary barrier
C5: Faulty connection between tank
dome and membrane wall
C6: Rollover. Boil off from LNG held
on ship for month before discharge
produced warmer denser gas
C7: Short circuit ignited LNG vapour
1965
1965
1966
1969
Cinderella
Methane Princess
Methane Progress
Polar Alaska
Loading
Unloading
Unkown (1)
Transit
Small
Small
Small
Medium
1971
Descartes
Unkown (1)
Small
1971
Esso Brega
Unloading
Medium
1976
Unloading
Large
1977
Loading
Medium
1977
LNG Aquarius
Loading
Small
1979
Mostefa Ben
Boulaid
Pollenger
LNG Taurus
Unloading
Small
Unloading
Transit (2)
Small
Small
None
C8: Wrong aluminum alloy used on

replacement valve caused it to fail on
contact with cryogenic temperatures
C9: Venting of LNG believed to be
due to leaking relief valves on cargo
tanks. Lightning struck two vent
risers simultaneously during discharge
C10: Check valve leakage during
discharge
C11: Valve leakage during discharge
C12: Fractures in main deck and
internals
C13: Flange leakage
Unkown
(Guayaquil
Equador)
LNG Delta
14
None
C14: Faulty arm disconnection
15
None
16
None
17
18
None
None
19
20
None
27 died
21
None
22
None
C15: Fractured upper deck and

internals
C16: Several cracked suction manifolds
and membrane leakage
C17: Cargo space leakage
C18: Overflow due to cargo valve
failure during discharge
C19: Pressurized cargo tank
C20: Heavy weather caused the vessel
to break free and sink
C21: Overpressure in tank leading
to leakage through vent
C22: Human error isolating alarm
leading to overfilling and spillage
1979
1981
? Prior to
1982
? Prior to
1982
1982
El Paso
Consolidated
Larbi Ben MHidi
Unkown (1)
Small
Unloading
Small
LNG Aquarius
Transit (2)
Small
1984
Ramdane Abane
Transit (2)
Small
1984
1985
LNG Gemini
Isabella
Transit
Unloading
Small
Medium
1985
1989
Anabella
Tellier
Transit (2)
Loading
Small
Small
2001
Khannur
Unloading
Small
2002
Mostefa Ben
Boulaid
Unloading
Medium
Remarks:
(1) The ship operation has been considered as unknown for incidents where there was no available information of the ship
during the spillage.
(2) Even though not clearly mentioned, the ship has been considered in transit for incidents where it can be denoted that they
happened under this condition.
incidents. Although this list is not extensive, it is useful

to confirm that there has never happened any collision
or grounding which resulted in loss of containment of
cargo systems, as clearly mentioned by IMO (2007).
Although in Figure 2 the main causes were attributed to
errors and operational mistakes, the historical analysis
of LNG spills shows that the events are rather a consequence of equipment failures without direct human
intervention. However, in some cases these equipment

failures could eventually be traced to human and
organizational failures as a result of inadequate maintenance procedures, wrong equipment selection or
budget restrictions. Nevertheless, this low human error
frequency in LNG spills highlights, once again, that
the good records that LNG vessels have are also due to
high standard procedures, training and crew selection.
436
Figure 3. Risk model for LNG spills in tankers considering historical spillage data.
SPILLAGE FREQUENCIES AND RISK

MODEL BASED ON HISTORICAL DATA
Based on 41 years of operational experience, from

1964 to 2005, there were only 22 occurrences of accidents which resulted in LNG spills, as it is stated in
Table 2. From these, only three caused injuries or fatalities and, in the last one, occurred in 1989, no trust is
put on this information, as it is contradictory and there
is no way to make it clear; moreover, the deaths might
be due to sinking of the ship. Furthermore, it has to
be emphasized that most accidents occurred prior to
1986. This can lead to the conclusion that although
LNG vessels already presented a very good record,
particularly when compared to other product tankers,
with a frequency of 19 spills in 21 years (0.90
spills/year) much was done in the last decades in order
to reduce this frequency to nearly 0.15 spills/year, as a

result of 3 spills in 20 years. These figures assume
more relevance by the fact that the LNG fleet has
increased considerably in the last decades, and therefore the number of ship exposed to risk is substantially
higher.
This fact could be related to the introduction of
more reliable technologies as well as more strict safety
procedures implemented along the years and it shows
that the LNG tankers have an increasing safety level,
reflecting in less frequent occurrences, particularly
associated to the loading/unloading operations, despite
the fact that in the last decades there was a substantial
increase of LNG vessels navigating. Also, the introduction of more monitoring systems allowed to an
earlier detection, which in turn minimized the size of
potential spills.
437
Figure 4. Bow-Tie approach Applied to the Risk Study of LNG Spillages.
From the 22 reported spillages, only one was of a

great magnitude; five were considered to be as medium
spillages; and the other 16 were of small magnitudes.
It can also be stated that the unloading operation is
critical to the LNG maritime transport cycle since most
of the spillages occurred under this condition. From
the 22 observed events, 9 (40.91%) occurred when the
ships were unloading LNG to the demand terminal,
including the only large observed spill. 6 (27.27%)
occurred while the ship was in transit, and
4 (18.18%) during loading. There were no details on
3 (13.64%) occurrences. It is important to mention that
the events which occurred during loading and unloading of cargo correspond to 13 (59.09%) from the total;
these conditions have to be carefully studied since most
terminals are located near populated areas.
Considering the total accumulated exposure of 2838
shipyears, as presented in Vanem et al. (2007), the
corresponding frequencies of occurrence of small,
medium and large spills are 5.64 1003 per shipyear,
1.76 1003 per shipyear and 3.52 1004 per
shipyear, respectively.
The risk model considering the historical occurrences of LNG spills is presented in Figure 3. The
basic events, C1 to C22, are described in the column
Causes on Table 2 above.
From the obtained results it is clear that, based on
historical data, although the triggering events can be of
different natures their materialization in an occurrence
is rather rare, particularly for the ones which present
larger consequences. It is worth noticing the role
that electric problem have in the occurrence of large
and medium size spills, and ultimately, in potential
explosion events.
For further work, consequence analyses of LNG
spillages using PHAST RISK software (DNV, 2009)
shall be performed in order to calculate their impacts
under different contexts as presented by Natacci et al.
(2010). The idea is to verify the impact of possible
fires and explosions as a function of spill size and LNG
composition. It is known that for simplification, LNG
is supposed to be 100% methane, which in fact is not
true. It consists of 95% methane, with smaller amounts
of ethane (3%), propane (1%), nitrogen (0.5%) and

heavier hydrocarbons (0.2%) (Van Horn et al., 1977).
These analyses may verify the difference in the magnitude of the consequences due to the variation of the
LNG chemical composition, as well as to the size of
the spillages.
Figure 4 illustrates the basic idea for evaluation
of the whole risk study of undesired LNG spillages,
where the bow-tie approach will be applied. This
present paper is related to the left part of the scheme,
concerned with the causes and frequencies of spills,
whilst the right part is related to further work, where
the consequences of spills shall be quantified in order
to calculate the risks involved.
4
CONCLUSIONS
In the present study an analysis of all historical

LNG tanker spillage accidents is performed. Based
on the HAZID developed for the IMOs formal safety
assessment of LNG carriers, a first risk model was
built, where all possible spill causes were identified. The fault tree technique is used for the development of this model. Then, based on 41 years
of operational experience, concerning the historical accident data presented in the risk analysis,
the model is rebuilt. This new model comprised
only the evidences observed during these 41 years.
The spills were categorized by size, namely: small,
medium and large spills, and the frequencies of occurrence of each category were estimated based on the
historical data.
Only one large spill was observed during the total
exposure of 2838 shipyears which shows the high
safety index associated to the LNG marine transport cycle. The total frequency obtained for LNG
spillage was 7.75 1003 per shipyear, where the
major contributors were small and medium spills,
corresponding to 72.73% and 22.73%, respectively,
of the total frequency.
Based on the quantification of the different spill
levels it will be possible to estimate the consequence
438
of a given spillage in a given location with less

uncertainties associated.
ACKNOWLEDGEMENTS
The first author has been financed by the EM
ECW Erasmus Mundus External Cooperation
Window, EU-Brazil Startup. The second author
acknowledges the financial support of the Portuguese
Foundation for Science and Technology under the
contract BD/31272/2006.
REFERENCES
American Gas Foundation, 2008. Ability of the United
States to Compete in the Global LNG Marketplace An
Assessment of Challenges and Opportunities, Benjamin
Schlesinger and Associates, Inc. USA.
Center for Chemical Process Safety of the American Institute of Chemical Engineers (CCPS), 1994, Guidelines for
Evaluating the Characteristics of Vapor Cloud Explosions,
Flash Fires and BLEVEs.
Center for Chemical Process Safety of the American Institute
of Chemical Engineers (CCPS), 1995a, Understanding
Atmospheric Dispersion of Accidental Releases.
Center for Chemical Process Safety of the American Institute
of Chemical Engineers (CCPS), 1995b, Understanding
Explosions.
Center for Chemical Process Safety of the American Institute of Chemical Engineers (CCPS), 1996, Guidelines for
Use of Vapor Cloud Dispersion Models.
Center for Chemical Process Safety of the American Institute of Chemical Engineers (CCPS), 1999, Guidelines
for Consequence Analysis of Chemical Releases.
Center for Chemical Process Safety of the American Institute of Chemical Engineers (CCPS), 2000, Guidelines
for Chemical Process Quantitative Risk Assessment,
second ed.
Center for Chemical Process Safety of the American
Institute of Chemical Engineers (CCPS), 2008,
Guidelines for Hazard Evaluation Procedures.
Det Norske Veritas (DNV), 2009, PHAST RISK Software,
version 6.53.1.
IMO, 2002, Guidelines for Formal Safety Assessment
(FSA) for Use in the IMO Rule-making Process.
IMO, 2007. Formal Safety Assessment Liquefied
Natural Gas (LNG) Carriers, Details of the Formal Safety
Assessment. MSC 83/INF.
Kumamoto, Hiromitsu & Henley, Ernest J., 2000, Probabilistic Risk Assessment for Engineers and Scientists.
Wiley-IEEE Press, 2nd Edition.
Natacci, F. B. , Ikeda, N. H., Martins, M. R. (2010). Consequence analysis of a liquefied natural gas leakage.
Proceedings of the 29th International Conference
on Ocean, Offshore and Arctic Engineering, Paper
OMAE2010-20689.
Van Horn, Andrew J. and Wilson, Richard, 1977, The Potential Risks of Liquefied Natural Gas, Energy, Vol. 2,
pp. 375389.
Vanem, E., Anto P., stvik, I. & Comas, F., (2008),
Analyzing the risk of LNG carrier operations,
Reliability Engineering and System Safety, Vol. 93,
Issue 9, pp. 13281344
Woodward, John L. & Pitblado, Robin M., 2010, LNG Risk
Based Safety Modeling and Consequence Analysis.
John Wiley & Sons, Inc.
439

Production regularity assessment using stochastic Petri nets with predicates

F.P. Santos, A.P. Teixeira & C. Guedes Soares
ABSTRACT: Generalized stochastic Petri nets (GSPNs) coupled with Monte Carlo simulation (MCS) are
presented in this paper as a flexible method for assessing the system production regularity quantified by its
throughput capacity distribution (PDC). The PDC describes the probability distribution of the various levels of
production of the system, which can be obtained by simulation or by analytical methods based on Markov modelling. Although more computational effort is required by the simulation methods, this approach is less restricted
in modelling complex systems and, consequently, in the assessment of the performance and regularity of real
production systems, as will be demonstrated in the present paper. The basic concepts of analytical and simulations approaches to assess system production regularity in terms of PDC are presented. One case study with
two different straightforward maintenance considerations is used to show the applicability of both approaches.
A short comparison of the results obtained is drawn and the merits of using the simulation approach are
enhanced.
INTRODUCTION
According to IEC (2005) the Life Cycle Cost (LCC)

is the cost of acquisition and ownership of a product
or system over a specified period of its life cycle and
can take into account the cost of development, acquisition, operation, maintenance, disposal of the product
or system. Since ownership costs may often become
higher than the acquisition ones, it is crucial to predict
and minimize it at an early phase of the system life
cycle.
Knowing that the output of a production system
can vary significantly during its life cycle, an availability evaluation may not capture this variability as it
only considers two states: functioning state and failed
state. Therefore, availability is not a fully adequate
measure of the systems performance (Rausand &
Hyland 2004). Among several proposed alternative
measures, NORSOK Z-016 standard introduces the
concept of production regularity as a measure of the
operational performance of oil and gas production systems, although its concepts can be applied in other
industries too. NORSOK (1998) defines regularity as
the capability of a production system to meet demands
for deliveries or performance.
An important input parameter to LCC evaluations
is regularity predictions which are essential to support
decision making concerning maintenance operations,
for example, or to reduce economic or technological risks. In turn, the production regularity is the
most relevant ownership cost parameter. However,
production regularity is defined as a broad concept
that is not directly applicable to quantify the systems
performance.
Kawauchi & Rausand (2002) have used the probability distribution of throughput capacity (PDC) to
measure the production regularity quantitatively. This
measure describes the probability distribution of the
various levels of production (e.g. numbers of barrels
of oil per unit time) of the system or the percentage
of time the system production is in each level. Despite
recognizing the merits of the simulation methods in
accurately predicting a systems performance, they
consider that its analysis are time and cost consuming. Whilst analytical methods may provide rougher
predictions than the simulation ones they require less
effort and as such Kawauchi & Rausand (2002) have
suggested an approach to assess the production regularity using an analytical method based on Markov
modelling to be used in LCC analysis in the oil and
chemical process industries.
A major challenge in the analysis of production
systems is to model correctly the complexity that
real systems usually have, namely, the dependencies
and dynamic interactions between the several components of a system (Siu 1994). In such cases, the
method proposed by Kawauchi & Rausand (2002)
to estimate the PDC can become quite laborious
analytically, since Markovian modelling can lead to
state-explosion. Therefore, their method is limited to
small scale systems.
Since Petri nets were first introduced by Carl
Adam Petri in 1962, several variants have been proposed for specific applications of production systems
and for the fields of computer science, communications, and automation, among others. Recently, several
techniques based GSPNs (e.g. Malhotra & Trivedi
1995, Dutuit et al. 1997) coupled with MCS were
441
developed for modelling and analysing systems in

terms of reliability, availability, and especially the productive efficiency (production availability) of complex
industrial systems (e.g. Dutuit et al. 1997, Boiteau
et al. 2006, Dutuit et al. 2008, Teixeira & Guedes
Soares 2009).
More recently, GSPNs with predicates were proposed. These have formulae/variables of Boolean,
integer, or float type that may be associated with some
functions (min, max, etc.) which can be used to validate transitions and be updated when transitions are
fired (Dutuit et al 2008). This paper presents a simulation method based on GSPNs with predicates, coupled
with MCS as a tool for assessing the production regularity of complex industrial systems through the PDC.
In order to show the applicability of and simulation
methods the analytical case study of Kawauchi &
Rausand (2002) is used in the following two maintenance situations: the original system (having no
dependency between failures and repairs of the components of the system since there is one repair team
for each of them); same system but with higher failures and repair rates, and only one repair team for
all the components (dependency between failures and
repairs). In these maintenance situations the aim is
to show that the simulation method based on GSPNs
with predicates is more efficient than the analytical one
proposed by Kawauchi & Rausand (2002), especially
when there are dependability issues.
The basic concepts of the analytical method with
Markov modelling are summarised and the simulation
approach with GSPNs is presented. A short comparison of the obtained results is drawn and the merits of
using the simulation approach are enhanced.
The case study in this paper is solved using the
Petri net analysis module of the software GRIF 4.08.1
(Folleau & Clav 2007), developed for the French oil
company, TOTAL, as an efficient tool for analyzing
the production efficiency of complex systems (Boiteau
et al. 2006).
2
Figure 1. Basic elements of a marked Petri net.
be assembled to model and simulate a system: places

(conditions) and transitions (events) as nodes; directed
arcs to relate the nodes. Usually, places, transitions
and arcs are depicted by circles, bars or boxes, and
directed arrows, respectively (Peterson 1981, Murata
1989, Schneeweiss 2001). Generally, places can be
seen as the systems and/or components states as
also as deposits of resources whereas transitions are
a representation of the events (e.g. failures, tests,
maintenance, etc) that handle those resources and
are responsible for triggering the changes of states.
Graphically, resources are symbolized by small
bold-face dots, named tokens, located inside places.
A state of the system and its resources are defined
in a Petri net by the position of a nonnegative integer
quantity of tokens in each place, i.e. by its marking.
In these conditions a Petri net is said to be marked
(Murata 1989). The change in the marking or states
is a function of the transitions and it is accomplished
by removing and/or creating tokens in places according to the direction defined by the arcs that connect
places to transitions (Peterson 1981, Murata 1989).
This characteristic of a Petri net allows simulating the
dynamic behaviour of a system (Teixeira & Guedes
Soares 2009). Figure 1 shows a graphic representation
of the basic elements of a marked Petri net.
PETRI NETS
2.2 Transitions: enabling and firing
Petri nets are a graphical and mathematical modelling

tool named after Carl Adam Petri who first introduced
the concept in 1962 on his Ph.D. dissertation about the
study of communications between automata (Peterson
1981, Murata 1989). The potential of its modelling
capabilities on discrete event systems was a breakthrough at the time leading afterwards to numerous
research studies on the subject. The upcoming years
also brought a considerable development of a variety
of Petri nets with enhanced modelling and simulation capabilities which nowadays are applied to a wide
range of scientific fields and engineering applications.
2.1
Basic elements
The dynamic behaviour of a real system can be defined

by a discrete changing of a set of conditions and events
in time. Petri nets have three basic elements that can
Arcs link places to transitions and are classified

according to their direction relative to a transition.
Hence, an arc connected to a transition is called an
input or output arc if it points to or from a transition, respectively. Places are named after the type of
arc they are linked to, so, there are input and output
places (Murata 1989).
From the Petri net theory the rule for enabling and
firing transitions is the most important (Murata 1989).
A transition is said to be enabled when all input places
contain tokens in a number equal or greater than the
weight of each arc. After a deterministic or random
delay the transition fires adding and subtracting tokens
to output places and from input places, respectively,
according to each arc weight.
Figure 2(a) shows an example of a transition (t1 )
not enabled (inhibited) since place (p1 ) has no tokens.
442
Figure 3. Two examples of transitions with inhibitor arcs.

Figure 2. Two examples of transitions of ordinary Petri nets.
Table 1. Meaning of the 5-tuple of a GPN (Peterson 1977,

Murata 1989, Folleau & Clav 2007).
Tuple
Meaning
P = {p1 , p2 , . . . , pm }
T = {t1 , t2 , . . . , tn }
P T = and P T ],
=
A: (P T) (T P)
Set of m places
Set of n transitions
Sets P and T are disjoint
and nonempty
Set of input and output arcs,
respectively
Weights of the arcs
W: A Z or A
{2, 1, 0, 1, 2, }
M0 : P N0 or P
{0, 1, 2, }
Initial marking, i.e before the firing

of the first transition; a marking M
is an m (total no. of places) size
vector and M (p) the number of
tokens in place p
On the other hand, Figure 2(b) represents an enabled

transition (t2 ) which after firing removes one token
from each input place (p1 and p2 ) and creates a new
one in the output place (p3 ).
A Petri net in which all arcs have a weight of
one is called ordinary (Peterson 1981, Murata 1989).
When weights larger than one are used, the net is
named a Generalized Petri Net (GPN) and has the
same modelling power as ordinary Petri nets (Peterson,
1977). GPNs are formally defined by a 5-tuple, R =
(P, T , A, W , M0 ) presented in Table 1.
Another particular aspect of weighted arcs worth
mentioning is that an output arc can only comprise a
positive weight, whereas an input one can have negative or zero, besides positive weights. A negative
weighted arc is called an inhibitor arc which graphically is represented by an arc with a small white
dot at the end. It inhibits the firing of a transition
while the input place has a number of tokens greater
than or equal to the absolute weight value (Folleau &
Clav 2007). A Petri net with inhibitor arcs is called
an extended Petri net (Murata 1989). Figure 3(a) and
Figure 3(b) show two small examples of a Petri net
with an inhibitor arc of weight 1 for which the transition is inhibited and enabled, respectively. The pairs
(p4 t 3 ) and (p4 t 4 ) are inhibitor arcs of weight 1.
The pairs (t3 p6 ) and (p5 t 3 ) are arcs of weight one
and two, respectively.
There are many other issues about Petri nets,

such as methods of analysis, behaviour and structural
properties, etc, which are not discussed in this article
because are beyond the scope. For those who wish a
more in depth reading it is advised Peterson (1981)
and Murata (1989).
Several classes of Petri nets were proposed over
time since the first introduction on the theme. The Petri
nets variants were developed from ordinary ones in
order to increase modelling capabilities according to
different and new applications needs.
The Petri nets covered in this article are the GSPNs
with predicates, which have modelling characteristics of both High-Level Petri Nets (HLPNs) and
Timed Petri Nets (TPNs), such as predicates (e.g.
variables) and time, respectively. A GSPN can have
in the same model transitions with random firing
time (e.g. exponentially distributed firing delays) and
firing times that are deterministically equal to zero
(Wang 1998).
CASE STUDY
3.1 Description
The case study considered is the one proposed by
Kawauchi & Rausand (2002). The production facility of Figure 4 comprises eight units. Basically, the
gas is separated from the well fluid and is treated
through high pressure separators (HPS) and dehydrators (DEH), from which follows to the compressors
(CMP-A and B). The purpose of the make-up compressor (MUP) is to enable CMP-A and CMP-B to
discharge gas with full flow rate even if some of the
gas treatment units (HPS or DEH) are failed. It is
assumed that the MUP has its own gas treatment units.
The maximum throughput capacity (TC) for each unit
is shown in Figure 4. These capacity values mean
that a unit is capable of dealing with the throughput
volume.
Each of the three basic subsystems (BS), 1, 2,
and 3 has their components in active redundancy. For
simplification purposes, all units are stochastically
independent with constant failure and repair rates.
Further, the following assumptions are considered:
units are independently repaired; each BS has identical
components with the same failure and repair rates, and
TCs. Table 2 contains the systems data.
443
Figure 6. Markov model of a parallel system (Kawauchi &

Rausand 2002).
Figure 4. System configuration of the case study

(Kawauchi & Rausand 2002).
Figure 5. Analytical method (Kawauchi & Rausand 2002).
3.2
Brief summary of the analytical approach
To assess the production regularity of a production

system by its PDC, Kawauchi & Rausand (2002)
proposed an analytical method depicted in Figure 5
by a three level approach.
First, the components (CMP), BSs and merged
subsystems (MSs) are identified. Second, a Markov
analysis is performed to determine the PDC for each
BS. If prior knowledge concerning the PDC of a BS is
available the Markov modelling is not needed and the
PDC data can be directly applied to define the PDC
of the BS. Finally, the merging rules are applied to the
BSs or MSs until the PDC of the whole system is
obtained.
The development of a Markov model for each BS
is important to establish the possible TC states (statespace) and transitions between states, thus allowing
us to obtain the TC distribution formulas. For this, it
is required to define the components most relevant
states, such as, functioning, failed and repaired and
to know their failure () and repair () rates. Markov
methods make possible to derive the time varying and
steady state PDC of BSs.
Figure 6 presents the Markov model of a BS

of two identical components (redundant) in parallel.
It is assumed that each component of a BS has two
possible TCs, c or 0.
Figure 6 may well represent the Markov model
of BS4 (see Figure 4) with three TCs states of 0%,
52% and 104%. The reasoning is the same for a BS
made of three parallel components like BS2, leading
to states: 0%, 55%, 110% and 165%. The probability
of occurrence of each state is a function of and
and the PDC of a BS may be obtained, for example,
by calculating the steady state probabilities for all its
states (see Table 3). This can be applied to any other
BS of the system in which components are in parallel. Notice that BSs have a unique PDC and are an
aggregation of components.
The PDC of a MS is estimated by applying two
merging rules according to the way BSs (or MSs)
are connected to each other: series or parallel. These
rules allow replacing BSs by a merged one.
Thus, a MS aggregates BSs or other merged ones
and has its own unique PDC.
To explain how the merging rules are applied
consider the following: two independent BSs, A
and B, and their TC values given by the discrete random variables, X (x) and Y (y), respectively; a MS, C,
resulting from merging A and B, with a discretely
distributed random variable Z(z) of TC values. If A
and B are in series the TC, Z, of C is equal to the
minimum of X and Y , and has the value Z = z, only
if at least one of the following conditions are met:
The TC of A is z, and the one of B is equal to, or

greater than z;
The TC of B is z, and the one of A is greater than z.
Accordingly, the probability of C having a TC value
z is given by,
or in other words, (1) is the PDC of C.

In case A and B are connected in parallel, the TC,
Z, of the resulting MS, C, has the value z if the sum
of the TCs of A and B is equal to z, that is, X +Y = z.
Therefore, the PDC of C is analytically given by,
444
Table 2.
Systems data (Kawauchi & Rausand 2002).

Components
Unit
Unit ID
Basic Subsys.
Capacity (%)
Repair rate
(per hour)
Failure rate
(per hour)
Subsystem
Prior PDC
HPS-A
HPS-B
HPS-C
DEH-A
DEH-B
CMP-A
CMP-B
MUP
CMP1
CMP2
CMP3
CMP4
CMP5
CMP6
CMP7
N/A
BS2
BS2
BS2
BS3
BS3
BS4
BS4
BS1
55
55
55
65
65
52
52
N/A
8.91 105
8.91 105
8.91 105
3.11 105
3.11 105
3.50 105
3.50 105
N/A
2.54 103
2.54 103
2.54 103
3.95 103
3.95 103
5.14 103
5.14 103
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
f(100) = 99.9%; f(0) = 0.1%
This paper is only focused on how to estimate the

PDC for discretely distributed TCs. For further and
more detailed information, such as, on how to calculate the PDC for continuous distributions and the
merging rules for resource (ownership and acquisition)
calculations, a reading of the article of Kawauchi &
Rausand (2002) is advised.
3.3 GSPNs with predicates as a support for
Monte Carlo simulation
The modelling capabilities of the GSPNs with predicates coupled with MCS are demonstrated by obtaining the PDC of a system through the modelling of
a case study with two different and straightforward
maintenance considerations. A comparison between
this approach and the analytical method proposed by
Kawauchi & Rausand (2002) is drawn. The advantages
of the simulation method are discussed.
3.3.1 Case study considering one repair team for
each unit
Each of the seven components, CMP#, of the system
of Figure 4 is modelled like the CMP1 of Figure 7.
Due to space limitations only the GSPN model of
another unit, MUP, is represented in Figure 7, since
there are slightly differences to the CMP# models.
Each of the CMP# has three states (conditions): functioning (e.g. CMP1_Funct), failed (e.g.
CMP1_Failed) and in repair (e.g. CMP1_inRepair).
In turn, the following three operation events are
modelled by as much transitions between states:
component fails (e.g. CMP1_Fail), start repairing (e.g. CMP1_Repair) and completion of maintenance (e.g. CMP1_Repaired). The failure times
(e.g. event CMP1_Fail) and repair times (e.g. event
CMP1_Repaired) of the components are exponentially distributed. However, the starting of the repair
of a failed component (e.g. event CMP1_Repair) is
represented by a Dirac function (drc) with a parameter
of zero (hours). This means that the repair task begins
immediately after the transition is enabled to be fired
(e.g. CMP1_Failed with jets = 1), i.e. the repair team
starts repairing the component immediately (with no
delays) after its failure.
Figure 7. GSPNs model of two systems units (GRIF).
Since it is assumed that all units (CMP#s and MUP)

are independently repaired, there are no dependencies
between their failures, meaning that each failed unit is
repaired immediately regardless of how many others
are being repaired at the time. That is why the Dirac
function (constant delay) with parameter zero is used
in all units. In other words, there is one repair team for
each unit.
The initial condition of the system is considered to
be full production, i.e. all places that represent the
functioning state of units are marked with one token
( jets = 1).
The MUP basic subsystem (BS1) is broadly modelled in the same way as the other units. There is a
difference though, it is the only BS, for which a prior
knowledge about its PDC is available. In fact, Table 2
shows that the probability of MUP having a TC level
of 100% or 0% in its life time is respectively 0.999
( f (100)) or 0.001 ( f (0)). So, in 1001 hours of operation, for example, the MUP works for 1000 hours
and it is failed for 1 hour. Consequently, the fail event
(MUP_Fail) and repaired event (MUP_Repaired) are
445
Table 3.
PDC of BS4 for the analytical and simulation methods.

Throughput capacity distribution
GSPNs with predicates and MCS
(n = 10, T = 108 h) GRIF
Analytical approach
Basic subsys.
BS4
Throughput capacity
level (%)
Formula
Value
Variable
Value
0
52
104
2 / ( + )2
2 /( + )2
2 /( + )2
4.57 105
1.34 102
9.87 101
Prod_BS4_0
Prod_BS4_52
Prod_BS4_104
4.35 105
1.34 102
9.87 101
represented by two deterministic transitions modelled with Dirac functions of parameters 1000 and 1,
respectively.
As the states of the model evolve over n missions of
duration T , the MCS of the system generates random
time values for all the transitions with valid conditions to be fired, according to their probabilistic laws.
This way, as soon as a transition becomes enabled a
time delay is computed and stored by the software.
All the stored time delays of these transitions are then
sorted and the one with the lowest delay is fired first.
After a transition has been fired it becomes inhibited
and its time delay is removed from storage. This process repeats itself until the simulation comes to an
end. The GSPNs with predicates as a support for
MCS allows drawing several statistics from places,
transitions, variables/formulae, etc.
Each place in the model of Figure 7 is identified
by a sequential number, and has either one token or
none, meaning that its value is one ( jets = 1) or zero
( jets = 0), respectively. The distributions of BS4s TC
levels rely on the combinations of states of places
CMP6_Funct and CMP7_Funct, i.e. its set of states
is given by jets = {0,0; 0,1; 1,0; 1,1}. Hence, BS4
has three different TC levels resulting from the sum
of the components capacity levels for each state:
0% (1 state), 52% (2nd and 3rd states) and 104%
(4th state).
To use the jets values of a place as a parameter of a
variable in GRIF software, one applies the semantics,
#, followed by the number of the place of interest.
Accordingly, the TC of BS4 in a simulation, is given
by Prod_BS4 = #16 52 + #19 52.
To know the PDC of BS4, or in other words, the
probability or percentage of time BS4 as a production
of 0%, 52% or 104% one has to define three variables
with an if then else (ite) function:
Prod_BS4_0=ite(Prod_BS4 == 0,true,false);
Prod_BS4_52=ite(Prod_BS4 == 52,true,false);
Prod_BS4_104=ite(Prod_BS4 == 104,true,false).
By the end of the simulation one has the mean
cumulated time for which these variables were true, i.e.
the total time as true divided by n. Dividing each one of
these times by T one gets the PDC of BS4. In fact, the
Prod_BS4_# variables have the same meaning as (2).
Table 3 shows the PDC of BS4 obtained by

both, the analytical approach (Kawauchi & Rausand
2002) and simulation method using GSPNs with
predicates.
The PDC estimation of BS2 and BS3 is achieved as
for BS4 since both comprise components in parallel.
The merged subsystem MS1 is based on two basic
subsystems in series, BS2 and BS3. Hence, the TC
of MS1 is the minimum (min) value between the
TC values of BS2 and BS3 at each randomly generated time. In GRIF this is accomplished by defining
the variable, Prod_MS1 = min(Prod_BS2,Prod_BS3)
where Prod_BS2 = #1 55 + #4 55+#755 and
Prod_BS3=#10 65+#1365. The PDC of MS1 is
estimated as was done for BS4, but for the following
TC levels: 0%, 55%, 65%, 110% and 130%. This way
one gets what is achieved analytically by (1).
The procedures explained until now are the
basis for the PDC estimation of any system. That
allowed the PDC calculation of MS2 and the system (MS3), taking into account their parallel and
series configurations. The systems TC is given by
Prod_Sist = min(Prod_BS4,Prod_MS2).
Table 4 presents the PDC values of the system for
the analytical and simulation methods.
The simulation results of Table 4 were obtained
in just 34 seconds using a regular computer and are
equal to the analytical ones. Although this is a short
computation time it can become higher with bigger
and complex systems but so can the effort to solve them
by the analytical method, which may even become
impossible for large systems.
Besides being a small system having BSs with
equal components, one shall not forget that it has
no dependencies, which make it simpler to estimate
the systems PDC. For simple systems both methods
require similar times and efforts to solve, but when
systems get more complex, such when dependability
is part of the equation, it may be cumbersome to use
the analytical approach, however, the advantages of
the simulation method are clear, as shall be seen in the
second maintenance case study.
To estimate the systems PDC one must capture the
time-independent PDC values which occur after some
initial transient effects. That is why the mission duration is very high compared with the MTTFs of the
446
Table 4. Systems PDC for the analytical and simulation

methods.
System
Probability (%)
TC level
(%)
Analytical
approach
GSPNs with predicates and MCS

(n = 10, T = 108 h) GRIF
0
52
55
65
100
104
4.58 103
1.34 100
3.28 104
1.52 103
9.85 103
9.86 101
4.36 103
1.34 100
3.30 104
1.51 103
9.36 103
9.86 101
components: to guarantee that the PDC steady state

behaviour is achieved.
A convergence analysis performed indicates that
only 10 simulations are required, since the variation
on the results obtained is less than 0.1%.
For a simple system like the one considered it is
quite straightforward to know its TC levels. However,
it can be quite laborious for a complex system. Using
GSPNs with predicates and MCS that may be more
easily achieved for any kind of system complexity.This
is accomplished by looking for the different generated values of the TC (e.g. Prod_Sist) for one history
only, so that those values are not averaged. One must
notice that the history duration must be sufficiently
large so that all possible TC levels of the system may
be captured in the simulation.
Another aspect worth mentioning is that when a
BS has different components, i.e. different failure and
repair rates and throughput capacities, the Markov
model gets more complex. For example, with the same
number of components but different from each other
the model of Figure 6 would have one more state, thus,
an additional analytic expression. This can even be
more difficult just by adding another different component to the BS, resulting in eight states instead
of four if the components were to be equal. With
GSPNs with predicates the model of the BS would
not change but for only its life time parameters and
a slight change in the predicates. So, in this context,
modelling more complex systems through GSPNs
with predicates requires much less effort than with the
analytical method.
3.3.2 Case study considering one repair team for
the whole system
All units considered in the previous case study have
very low failure rates which mean that the MTTFs
are particularly high. Hence, failures occur with a
very low frequency. Usually, failures rates of this kind
are associated to rare or major events (faults) in
critical systems that for safety reasons have many
redundancies. Under these circumstances, it is quite
unlikely that two failures may occur at the same time,
meaning that there is almost no time interdependency
between failures of system components. So, as an
approximation the interdependency can be neglected

by modelling one repair team for each unit.
Approximations are suitable for simplification purposes and can be considered for analytical methods
as Markov analysis, reliability block diagrams (RBD)
and fault tree analysis (FTA). In particular, RBD and
FTA are limited to Boolean algebra and are much less
appropriate to model time dependencies.
However, this is not the case for production regularity and production availability assessments in operating conditions. Generally, a production system has
much higher failure rates than the ones of Table 2.
Therefore, time interdependencies between failures of
components are no longer negligible. Furthermore,
resources are often limited and thereby must be used
efficiently. As part of the ownerships costs, there must
be given a special attention to maintenance costs,
and its tasks should be properly managed. Hence, the
maintenance teams must be sized according to the
available and necessary resources to meet the systems operation, production and safety requirements.
Instead of using a repair team for each unit as was
previously done, it will be considered the same system
but with different failure and repair rates; with only
one repair team; three distinct corrective maintenance
(CM) strategies.
The gas production facility is considered to be in
an offshore installation. The repair rates of the BS2
and BS3 units are the same as in Table 2 whereas the
failure rates are ten times higher. The components
failure and repair rates of BS4 are one hundred and
ten times greater than the ones of Table 2, respectively.
The prior PDC of BS1 is f (100) = 0.7 and f (0) = 0.3.
Each unit of the system is modelled in the same
way as in the previous case study consideration (see
Figure 7) but for the three maintenance strategies presented hereafter: team mobilization equal to 0, 48 and
120 hours.
With just one repair team for the whole system only
one repair task can be performed at a time. Hence, if
there is more than one unit failure at a given moment
in time the team jumps from one unit to another as
soon as it is done repairing, keeping meanwhile some
units on hold for repair until all are repaired. The order
in which repairs are done is the failure order of units.
The repair team is a maintenance human resource
which can only start repairing units when on site, i.e.
on the offshore installation. As such, it is represented
in a Petri net by a place, Repair_Team_OnSite, with
just one token ( jets = 1) since there is only one team
available, as can been seen in Figure 8. This figure
only shows the model of one unit since all others are
all modelled in the same way. It depicts an example of
how to model a system with one CM team which does
not need time for mobilization, i.e. it is always on site.
This is the first CM strategy considered. The
place number 27 of Figure 8 is used as another
condition to validate the start repairing (e.g.
CMP1_Repair) and completion of maintenance
(e.g. CMP1_Repaired) events. Thus, the repairing
of a unit starts (e.g. CMP1_Repair fires) when it
447
Figure 8. CM model for 0 hours mobilization (GRIF).
Figure 10. CMP1 model with repair tasks included (GRIF).
Figure 9. CM model for 48 hours mobilization (GRIF).
is failed (e.g. CMP1_Failed with jets = 1) waiting

for repair and the repair team is available (e.g.
Repair_Team_OnSite with jets = 1). Therefore, the firing of transition CMP1_Repair removes a token from
CMP1_Failed and Repair_Team_OnSite creating a
new one in place CMP1_inRepair. So, while CMP1
is being repaired there are no other repair teams available (i.e. Repair_Team_OnSite with jets = 0) for repair
in the event of a failure of another unit.
After the maintenance task is completed (e.g. firing
of CMP1_Repaired) the unit is set to function again
(e.g. CMP1_Funct with jets = 1) and a token is created
in place Repair_Team_OnSite, meaning that the repair
team is released from repairing CMP1 and is available for maintenance tasks involving other units that
might have failed meanwhile or that may fail during
the systems life time.
The starting point for the models of the second
and third maintenance approaches is the model of the
first strategy. However, it must be considered that the
repair team is not initially on site, i.e. it is demobilized. Thus, the team needs to be called or mobilized
from land to the offshore installation whenever there
are CM tasks to perform on the system. The repair
team takes 48 and 120 hours from the time it is mobilized to the moment of arrival on site, for the 2nd
and 3rd maintenance strategies, respectively. The team
only leaves the installation when all the maintenance
tasks are completed.
Both strategies are modelled in the same way by
the small Petri net and the auxiliary place of Figure 9.
These are associated to the Petri nets of the units as
it is shown by the example of Figure 10. Both figures
show how just the introduction of a positive mobilization time in the CM can change the GSPN model in
comparison to the 1st maintenance strategy, i.e. other
places and transitions are added to complement the
Repair_Team_OnSite place.
Initially, all units are functioning with no failures
(No_Units_Failed with jets = 0) and the repair team
is on land prepared to be summoned at short notice
(Repair_Team_Demobilized with jets = 1).
Place number 28, works as a counter of the units
that are failed. A token is created in this place each
time a unit fails, so there are as many tokens as failed
units. On the other hand, a token is removed whenever
448
Table 5.
MCS results of the system considering three maintenance strategies

GSPNs with predicates and MCS (n = 10, T = 106 h)
CM strategies
PDC (%)
System
TC level
(%)
Probability (%)
0h
48 h
120 h
480 h
1200 h
0
52
55
65
100
104
1.07 100
1.19 101
3.90 100
2.95 100
1.29 100
7.89 101
1.20 100
1.49 101
4.18 100
3.09 100
1.36 100
7.52 101
1.55 100
1.83 101
4.69 100
3.41 100
1.52 100
7.05 101
11.93
25.74
7.27
4.87
5.81
4.67
44.15
54.06
20.30
15.63
18.32
10.62
a unit is repaired. As an example, when the CMP1

component of Figure 10 fails the transition CMP1_Fail
fires and adds a token to No_Units_Failed. When it
is repaired the transition CMP1_Repaired fires and a
token is removed from the counter.
The team is mobilized from the moment there is at
least a unit failure. Accordingly, the logic condition
#28 >= 1 of the transition Repair_Team_Mobilization, that was initially false, takes the value true and
it fires according to a Dirac function of parameter
48 (hours). Once on site (Repair_Team_OnSite with
jets = 1) the team starts immediately working, meaning that the token of place number 27 is removed by the
firing of transition CMP1_Repair. Afterwards, when
the task ends, a token is created in the same place by
the firing of transition CMP1_Repaired.
The team stays performing maintenance tasks as
long as #28 >= 1. By the time all units are repaired
(i.e. No_Units_Failed with jets = 0) place number 27
as one token and the logic condition #28 = = 0 of
the transition Repair_Team_Demobilization becomes
true enabling it to be fired following a Dirac function of parameter 0 (hours). This way the team is
immediately demobilized.
The Monte Carlo simulation results of the systems
PDC considering the three maintenance strategies
are in Table 5. The computation time is just 2.54,
4.29 and 3.46 seconds for the mobilization times of 0,
48 and 120 hours, respectively.
One must notice that the variables used in PDC
estimation for the three maintenance strategies are
precisely the same of the one repair team for each
unit consideration.
From Table 5, one concludes that the PDC of
the 104% level decreases with the increasing in the
mobilization time. With a cost/benefit analysis one can
choose the best CM strategy that reduces ownership
costs while complying with production needs.
The maintenance strategies introduced in the one
repair team consideration were modelled almost
without effort since were needed small modifications
of the model. This is more difficult to model with
Markov analysis as there are dependabilities between
components, i.e. they have to share the same repair
team when more than one unit fails, which turns the
Markov model of Figure 6 more complex and cumbersome to solve analytically. Besides, introducing more
complexity in the system will lead to state-explosion
in a Markov analysis.
CONCLUSIONS
By using a case study with two maintenance conditions, this paper shows that the simulation method
(MCS) based on GSPNs with predicates is more flexible to calculate the PDC of a system than the analytical
method proposed by Kawauchi & Rausand (2002),
especially when dealing with dependability problems.
The simulation approach can be as accurate as the analytical one, depending on the number of simulations
executed.
Generally, the downturn of simulations methods is
that the computation time increases with the number
of simulations. However, a great number of simulations may not be needed to achieve a good accuracy
with GSPNs as was shown in the convergence analysis (n = 10) for the first maintenance consideration.
In this case the computation time was only 34 seconds.
When there are dependencies between failures and
repairs of components, as in the second maintenance
case, the Markov models become more complex and
can lead to state explosion. In this situation the simulation method proves to be more efficient than the
analytical method.
GSPNs with predicates and MCS are more flexible than analytical methods (Markov analysis, RBD
and FTA) since their structure (elements) and the use
of predicates can reduce the model size. Furthermore,
as the system complexity increases (e.g. number of
components) the Markov modelling leads to stateexplosion, thus being applicable only for small and
simple systems.
The GSPNs with predicates as a support for MCS
are a powerful tool for modelling and simulating
complex production systems, their dynamic behaviour,
performance, dependability (reliability, availability
and maintainability) and LCCs. Adopting simulation
449
methods and in particular the use of the GRIF software

has the following advantages:
smaller models when dealing with time interdependencies;

can handle large and complex systems since is less
prone to state-explosion as the systems complexity
increases;
can model stochastic and determinist events within
the same MCS;
the accuracy of the results obtained can be easily
checked and increased by increasing the number of
simulations;
plan and manage maintenance strategies more
effortlessly and efficiently;
TC levels calculations can be more easily accomplished for any kind of system complexity, whereas
it gets quite laborious analytically as the systems
complexity increases;
the complexity of the PDC estimation do not
change from a system containing BSs with equal
components in parallel to an equal system but with
different units in parallel. This gets much more
complicated for Markov analysis;
non-exponential laws can be used which is almost
impossible to adequately model analytically;
the comprehension of most of the systems dynamic
behaviour and performance is more intuitive due to
its graphic interface.
ACKNOWLEDGEMENTS
The first author acknowledges the financial support of the Portuguese Foundation for Science and
Technology.
REFERENCES
Boiteau, M., Dutuit, Y., Rauzy, A. & Signoret, J.-P. 2006.
The AltaRica data-flow language in use: modelling of
production availability of a multi-state system. Reliability
Engineering and System Safety 91: 747755
Dutuit, Y., Chtelet, E., Signoret, J.-P. & Thomas, P.,

1997. Dependability modelling and evaluation by using
stochastic Petri nets: application to two test cases.
Reliability Engineering System Safety 55 (2): 117124.
Dutuit, Y., Innal, F., Rauzy, A. & Signoret, J.-P. 2008. Probabilistic assessments in relationship with safety integrity
levels by using Fault Trees. Reliability Engineering and
System Safety 93: 18671876.
IEC 2005. IEC 60300-3-3 Dependability management,
Part 3: Application guide-Section 3: Life cycle costing,
International Electrotechnical Commission.
Kawauchi, Y., Rausand, M. 2002. A new approach to production regularity assessment in the oil and chemical
industries. Reliability Engineering and System Safety 75:
379388.
Malhotra, M. & Trivedi, K.S. 1995. Dependability modelling
using Petri nets. IEEE Transactions on Reliability 44(3):
428440.
Marsan, M.A., Balbo, G., Conte, G., Donatelli, S. &
Franceschinis, G., 1995. Modelling with generalized
stochastic Petri nets. John Wiley & Sons inc.
Murata, T. 1989. Petri Nets: properties, analysis and applications. Proceedings of the IEEE 77(4): 541580.
NORSOK 1998. Regularity management & reliability
technology. NORSOK Standard Z-016, Rev.1, Norwegian
Technology Standards Institution
Peterson, J. L. 1977. Petri nets. Computing Surveys 9(3):
223252.
Peterson, J. L. 1981. Petri net theory and the modelling of
systems. Englewood Cliffs: Prentice-Hall, Inc.
Rausand, M. & Hyland, a. 2004. System reliability theory: models, statistical methods and applications. Second
edition. Hoboken, New Jersey: John Wiley & Sons, inc.
Schneeweiss, W. G. 2001. Tutorial: Petri nets as a graphical
description medium for many reliability scenarios. IEEE
transactions on reliability 50(2): 159164.
Siu, N. 1994. Risk assessment for dynamic systems: an
overview. Reliability Engineering and Systems Safety 43:
4373.
Teixeira, A. P. & Guedes Soares, C. 2009. Modelling and
analysis of the availability of production systems by
stochastic Petri nets (in Portuguese). Riscos industriais e
emergentes 1: 469488. Lisboa: Edies Salamandra, lda.
Wang, J. 1998. Timed Petri nets: theory and application.
Kluwer Academic Publishers.
450

Reliability assessment of a plate subjected to randomly non-uniform

corrosion and compressive load
J.E. Silva, Y. Garbatov & C. Guedes Soares
ABSTRACT: The objective of this work is to evaluate the reliability of a nonlinear time variant randomly
non-uniform corroded rectangular plate subjected to compressive load. Based on the derived relationship of the
ultimate strength assessment of a rectangular plate with surface geometries deteriorated by different degree of
random corrosion, location and ages, FORM/SORM techniques are applied to assess the structural reliability
accounting for the global ship hull deterioration. The structural system, composed by two different correlated
events, is evaluated as a series system.
INTRODUCTION
Structural analysis and design have traditionally been

based on deterministic methods. However, uncertainties in the loads, strengths and in the modeling of the
structures require that methods based on probabilistic
techniques in a number of situations have to be used.
A structure is usually required to have a satisfactory
performance in the expected lifetime, i.e. it is required
that it does not collapse or becomes unsafe and that it
fulfils certain functional requirements.
The reliability estimated as a measure of the safety
of a structure can be used in a decision process. A
lower level of the reliability can be used as a constraint in an optimal design problem. The lower level
of the reliability can be obtained by analyzing similar
structures designed after current design practice or it
can be determined as the reliability level giving the
largest utility (benefits costs) when solving a decision problem where all possible costs and benefits in
the expected lifetime of the structural components are
taken into account.
Generally, the main steps in a reliability analysis include: select a target reliability level; identify
the significant failure modes of the structural component; decompose the failure modes in series systems of parallel systems of single components (only
needed if the failure modes consist of more than one
component); formulate failure functions (limit state
functions) corresponding to each component in the
failure modes; identify the stochastic variables and
the deterministic parameters in the failure functions
and further specify the distribution types and statistical parameters for the stochastic variables and the
dependencies between them; estimate the reliability of
each failure mode; compare with the target reliability;
evaluate the reliability result by performing sensitivity

analyses.
The failure modes (limit states) are generally
divided in: ultimate limit states correspond to the
maximum load carrying capacity, which can be related
to e.g. formation of a mechanism in the structure,
excessive plasticity, rupture due to fatigue and buckling; conditional limit states correspond to the
load-carrying capacity if a local part of the structure
has failed. The conditional limit states can be related
to e.g. formation of a mechanism in the structure, exceeding of the material strength or buckling;
serviceability limit states related to normal use of
the structure, e.g. excessive deflections, local damage
and excessive vibrations.
The fundamental quantities that characterize the
behavior of a structure are the basic variables, denoted
as X = (X1 , X2 , X3 , . . . , Xn ), where n is the number
of basic stochastic variables. Typical examples of
basic variables are loads, strengths, dimensions and
materials.
The reliability analysis presented here is using
FORM/SORM techniques to identify a set of basic
random variables, which influence the failure mode
or the limit-state under consideration. The limit-state
function is formulated in terms of the n basic variables
given as:
This function defines a failure surface when equals

to 0. It defines an (n 1) dimensional surface in the
space of n basic variables. This surface divides the
basic variable space into a safe region, where g(x) > 0
451
and an unsafe region where g(x) < 0. The failure probability of a structural component with respect to a
single failure mode can formally be written as:
where fX () is the joint probability density function

of the n basic variables and Pf denotes probability of
failure. The n-dimensional integral is defined over the
failure region.
In practical applications, the reliability cannot be
evaluated in the exact manner given by Eqn (2). This is
because enough statistical data is usually not available
to develop the n-dimensional joint density function
of the basic variables. Secondly, even when the joint
density function is available, analytical or numerical
integration is possible only for a few simple cases.
The FORM/SORM methods provide a way of evaluating the reliability efficiently with reasonably good
accuracy, which is adequate for practical applications (Hasofer and Lind 1974, Rackwitz and Fiessler
1978, Ditlevsen 1979 and Hohenbichler and Rackwitz
1986).
The time variant formulation of ship reliability
results from modeling the problem with stochastic
processes that represent the random nature of the load
and strength parameters.
Two main corrosion mechanisms are generally
present in steel plates. One is a general wastage that
is reflected in a generalised decrease of plate thickness. Another mechanism is pitting, which consists of
much localised corrosion with very deep holes
appearing in the plate.
Various models of corrosion deterioration have
been proposed. (Melchers 1998) suggested a steadystate trilinear model for corrosion wastage thickness.
In fact, experimental evidence of corrosion reported
by various authors show that non-linear models are
appropriate. (Yamamoto and Ikagaki 1998) proposed
a corrosion model based on analyzing collected from
plate thickness measurements. (Guedes Soares and
Garbatov 1999), proposed one model that describes
the growth of corrosion wastage by a non-linear function of time in three phases. The corrosion model of
(Paik, Lee et al. 2003), also categorizes the corrosion behaviour into three phases, durability of coating,
transition to visibly obvious corrosion, and progress
of such corrosion.
Engineering systems such as ship structures are
designed to ensure an economical operation throughout the anticipated service life in compliance with
given requirements and acceptance criteria. Deterioration processes such as fatigue crack growth and
corrosion are always present to some degree and
depending on the adapted design philosophy in terms
of degradation allowance and protective measures the
deterioration processes may reduce the performance
of the system beyond what is acceptable.
In order to ensure that the given acceptance criteria are fulfilled throughout the service life of the
engineering systems it is necessary to control the

development of deterioration and, if required, to install
corrective maintenance measures resulting in inspection in the most relevant and effective means of
deterioration control.
The initial formulations of the time variant approach
to ship structural reliability were developed in
connection with the fatigue problem, in particular to
be able to deal with the time degradation of reliability by (Guedes Soares and Garbatov 1996) with
the improvements made by maintenance actions by
(Guedes Soares and Garbatov 1996) and for corrosion
deterioration by (Guedes Soares and Garbatov 1998).
The present paper deals with the ultimate strength
assessment of ageing steel plates subjected to compressive load and randomly non-uniformly distributed
corrosion wastage.This work is continuation of the one
presented by (Silva, Garbatov et al. 2011), where the
effect of non-linear randomly distributed non-uniform
corrosion on the ultimate strength of unstiffened
rectangular plate subjected to axial compressive loading has been studied. A series of 570 plate surface
geometries are generated by Monte Carlo simulation
for different degree of corrosion, location and ages
and nonlinear finite element analyses were carried out,
using a commercial finite element code and based
on a regression analysis, empirical formulae to predict strength reduction because of corrosion have been
derived demonstrating a good accuracy.
2
RELIABILITY ASSESSMENT
The reliability assessment conducted in the present

study deals with the structural failure hypothesis due
to corrosion deterioration in the service life of a certain ship. The corrosion degradation can affect the
ship integrity either due to improper surveillance or
due to strength loss. The resulting events are modeled as series system assuming that the inspections
on corrosion may be ineffective, or the structure will
collapse due to strength degradation accounting for the
correlation between the referred failure modes.
2.1 Corrosion deterioration failure mode, E1 (t)
The time dependent limit state function that reflects
corrosion degradation along the years, of a steel deck
plate takes is based on the corrosion model proposed by (Guedes Soares and Garbatov 1999) and the
demands of classification societies for replacing any
structural element affected by corrosion. The implicit
limit state function is defined as:
where DCS is the demands of classification societies

for replacing any corroded structural element, which
is modeled as a log-normal distribution with mean
value and standard deviation of 2 mm and 0.1 mm
respectively, accounting for the fact that the studied
452
strength, x SW is the uncertainty in the model of predicting the still water bending moment, x w is the error
in the wave bending moment due to linear sea keeping analysis and x s takes into account nonlinearities in
sagging.
The full statistical description of parameters
involved in the limit state function presented by
Eqn (7) are given as (Mansour, Lin et al. 1993):
Figure 1. Reliability index and correspondent probability

of failure of E1 (t).
plate is located on deck. Dcorr is the corrosion depth,

modeled as a log-normal distribution:
where the mean value and standard deviation are

defined by (Guedes Soares and Garbatov 1999) as:
and d = 1.85 mm is the long term corrosion depth

for a steel deck plate, C = 10.54 years is the coating
life and t = 11.14 years is the transition period.
Reliability index and probability of failure defined
by the limit state function expressed by Eqn (3) based
on FORM can be found in Figure 1.
2.2
where N denotes the normal distribution function

and the first and second indicator inside the brackets refer to the mean value and standard deviation
respectively.
The midship section modulus is modeled based on
the minimum requirements of Classification Societies.
It is assumed here that the ship section modulus as
built, accounting for the additional corrosion plate
as built , is considered as 25% larger than
thickness, S M
the minimum required one. However, along the ship
service life, the midship section will be degraded and
consequently the section modulus will be also reduced.
The corrosion degradation of the midship section
modulus along the service life is assumed here as an
example to be linear and will be modeled as a function time by a reduction factor . The midship section
modulus is assumed here as a Log-Normal distribution
function
with the mean value at the beginning of the service life
equal to 25% of minimum required by Classification
Society and a covariance of 4% as:
where
Ultimate strength failure mode, of E2 (t)
The limit state function developed for ultimate strength

of a deck plate subjected to compressive load for
sagging condition is defined as:
(t) is the midship section modulus,

cr (t)
where S M
SW is the still water
is the critical failure stress, M
w is the wave induced bending
bending moment, M
moment, x U is the model uncertainty on ultimate
where 25 is the section modulus reduction factor at

the 25th year of service live.
Four severity levels of corrosion degradation are
analyzed here: 25 = 0.00, 0.05, 0.10, 0.15 i.e. corresponding to a section modulus reduction of 0%, 5%,
10%, 15% at the 25th year of service life respectively.
(Silva, Garbatov et al. 2011) derived a formula,
based on a Monte Carlo simulation, to predict the
ultimate strength reduction of a corroded steel plate
453
subjected to a compressive load, the ultimate strength

is modeled by a Log-normal distribution function
with a mean value and standard deviation presented
in Eqn (12) and Eqn (13) respectively.
where Y = 352 MPa is the yield stress, C,U =

10.54 years is the coating life, C,U
= 13.94 years is
the time from which the standard deviation starts to be
significant, t,U = 49.92 years is the transition period
in years, and n = 1.42 is the time decay capacity of the
ultimate strength.
The still water bending moment is described by
a normal distribution. It is assumed here that the
value given by Classification Societies is the maximum value with a probability of exceedance of 5%.
The large variability in the still water bending moment
results in a coefficient of variation of 40%, which
gives the mean value of the distribution to be 60% of
SW .CS thus (Guedes Soares and Moan 1982; Guedes
M
Soares and Moan 1988):

of failure of E2 (t) for 25 = 0.
If the wave induced loads can be represented as a stationary Gaussian process (short-term analysis), then
various methods may be used to define the probability
density function of the maximum load. The extreme
value distribution is based on up crossing analysis.
The wave induced bending moment given by
Classification societys rules is modeled as an extreme
value following the distribution function (Mansour
1990):
of failure of E2 (t) for 25 = 0.05.
Eqn (12) and (13) are derived for a deck plate with
a thickness of 10 mm, which is a part of a ship with
a length of 128 m between perpendiculars, 19.2 m
beam, 10.4 m depth, moulded draught of 8.5 m and
a block coefficient of 0.704. For the ship used in the
present study the required design bending moments
and section modulus are (DnV 2001):
where w , is the mean value of the distribution and
w is the standard deviation. N is the number of wave
bending moment peaks and 0 is the mean square of
the wave bending moment process. The value given by
Class societys rules is assumed to be the mean value
and choosing N to be 1000, which is equivalent to a
3 hours storm, gives a coefficient of variation of 9%.
For the different corrosion severities, values of 25 , the

FORM results are shown in Fig. 2 to 5.
454
The occurrence of any failure event Ei , for series

system of events, will cause the failure of the entire
system. The failure event of the entire system, E is
a union of all possible failure modes, which can be
expressed as follows:
The failure probability can be computed either by

using the Morgans law, Eqn (21), or by solving the
multidimensional integral based on Eqn (2).

The n-dimensional integral presented in Eqn (2) can

be solved only for some limited case and to overcome
this problem some bounding techniques may need to
be applied, such as the ones proposed by (Cornell
1969) and by (Ditlevsen 1979). Ditlevsen proposed
narrow bounds on the probability of failure of a
series system, given by:
where the joints probabilities of failure, P(Eij ) can be

expressed (Ang and Tang 1984) as:

As can be observed from Fig. 2 to 5, there is an

abrupt decrease in the reliability index from the 13th
to the 14th year of service life, in comparison to what
has been observed in the corrosion degradation failure mode, E1 (t) as can be seen in Figure 1. This can
be explained with the fact that the structure will only
respond to the effects of corrosion degradation a few
years after the coating time. Moreover, prior to the 13th
year the uncertainties on structural collapse are still
considerably low, when compared to the ones present
on following years.
3
where i and j are the reliability indices corresponding to the ith and jth failure modes, respectively,
ij is the correlation coefficient between the ith and
jth failure modes and 2 () and () are the probability density function and cumulative distribution
functions, respectively, of the 2D standard normal
distribution.
Eqn (22) represents the narrow bounds for the
system probability of failure, and they still regard the
joint failure modes. (Ditlevsen 1979) also proposed
a method for bounding joint probability of failure,
P(Eij ) as
SERIES SYSTEM RELIABILITY
Consider a series of events with i possible failure

modes, with the limit state functions defined by
Eqn (1). The failure event for a failure mode i is:
where
455
Figure 8. Joint Probability of failure, P(E12 ), for = 0.3.

Figure 6. Joint probability of failure, P(E12 ), for = 0.1.

Figure 7. Joint Probability of failure, P(E12 ), for = 0.2.
Taking into consideration that, in the present problem,

P(E1 ) is always greater than P(E2 ) then the bounds
for the system probability of failure, expressed by
Eqn (22) to (27b), may be written as:
Although, Ditlevsen (1979) bounds demonstrated to be
more practical and usable than the exact expressions.
3.1
Series system composed by two failure modes
where
The probability of failure of the series system, composed by the two events may be defined as:
456
in the same period, of 61.48%, 68.87%, 76.81% and

85.33% for 25 = 0, 0.05, 0.10, 0.15 respectively.
ACKNOWLEDGEMENTS
This work has been carried out under the pluriannual
funding of the Portuguese Foundation of Science and
Technology (Fundao para Cincia e Tecnologia) to
the Centre for Marine Technology and Engineering
(CENTEC).
The work reported here is a contribution to the activities of the MARSTRUCT VIRTUAL INSTITUTE,
(www.marstruct-vi.com) in particular its Technical
Subcommittee V on Risk Based maintenance.
REFERENCES
For the two events problem, PA + PB is always less

than P(E2 ), then one can easily see that Eqn (29)
and Eqn (30) becomes equivalent to Eqn (25). Eqn
(23) is an exact solution for the joints probabilities of
failure P(Eij ) and a numerical integration is needed to
be performed.
For the evaluation of the joint probability of failure, P(E12 ), only the case where 25 = 0, Eqn (10),
will only be considered here. For different correlations
between the studied events accounting for the
Ditlevsen bounds, Eqn (25), the joint probability of
failure is shown on Figures 6 to 10.
Figures 6 to 10 show how the joint probability
of failure, P(E12 ), lie inside the bounds, which represent the most and less conservative probability of
failure for a certain correlation factor, . The figures
also include the case where the events are mutually
exclusive.
CONCLUSIONS
The present work studied the reliability of a series

system composed by two time variant events, corrosion deterioration and ultimate strength assessment of
rectangular plate, where the correlation between them
was considered as unknown.
FORM techniques were applied in order to analyze
the variation of reliability index in time. Bounding
techniques were also used to evaluate the joint failure
mode for different correlations.
The corrosion degradation limit state function presented a decrease, from the 11th year to 25th year,
of 58.53% on the reliability index. While reliability
index of the ultimate strength failure mode decreases
Ang, A. and H. Tang (1984). Probability Concepts in Engineering Planning an Design, John Wiley and Sons, Inc.
Cornell, C. A. (1969). A Probability Based Structural Code.
American Concrete Institute Journal 66: 974985.
Ditlevsen, O. (1979). Generalised Second Moment Reliability Index. Journal of Structural Mechanics 7: 435451.
Ditlevsen, O. (1979). Narrow Reliability Bounds for
Structural Systems. Journal of Structural Mechanics 7:
453472.
DnV (2001). Hull Structural Design, Ships with Length 100
Meters and Above. Norway, Det norske Veritas. Part 3,
Chapter 1.
Guedes Soares, C. and Y. Garbatov (1996). Fatigue Reliability of the Ship Hull Girder. Marine Structures 9:
495516.
Guedes Soares, C. and Y. Garbatov (1996). Fatigue Reliability of the Ship Hull Girder Accounting for Inspection
and Repair. Reliability Engineering & System Safety 51:
341351.
Guedes Soares, C. and Y. Garbatov (1998). Reliability
of Maintained Ship Hull Subjected to Corrosion and
Fatigue. Structural Safety 20: 201219.
Guedes Soares, C. and Y. Garbatov (1999). Reliability
of Maintained, Corrosion Protected Plates Subjected to
NonLinear Corrosion and Compressive Loads. Marine
Structures: 425445.
Guedes Soares, C. and T. Moan (1982). Statistical Analysis of Still Water Bending Moments and Shear Forces
on Tankers and Bulk Carriers. Norwegian Maritime
Research. 10: 3347.
Guedes Soares, C. and T. Moan (1988). Statistical Analysis
of Still-Water Load. Effects in Ship Structures. SNAME
Transactions 96: 129156.
Hasofer, A. M. and N. C. Lind (1974). An Exact and
Invariant First-Order Reliability Format. Journal of
Engineering Mechanics Division 100: 111121.
Hohenbichler, M. and R. Rackwitz (1986). Asymptotic
Outcrossing Rate of Gaussian Vector Process into Intersection of Failure Domains. Probabilistic Engineering
Mechanics 1(3): 177179.
Mansour, A. (1990). An Introduction to Structural Reliability
Theory.
Mansour, A., M. Lin, et al. (1993). Probability-based Ship
Design Preocedures: a Demostration.
Melchers, R. (1998). Probabilistic Modelling of Immersion Marine Corrosion. Structural Safety and Reliability. N. Shiraishi, M. Shinozuka and Y. Wen, Balkema:
11431149.
457
Paik, J., J. Lee, et al. (2003). Time-Variant Ultimate

Longitudinal Strength of Corroded Bulk Carriers.
Marine Structures 16: 567600.
Rackwitz, R. and B. Fiessler (1978). Structural Reliability
Under Combined Random Load Sequences. Computers
and Structures.
Silva, J. E., Y. Garbatov, et al. (2011). Ultimate strength
assessment of ageing steel plates subjected to random
non-uniform corrosion wastage. Advances in Marine

Structures. C. Guedes Soares and W. Fricke, Taylor &
Francis: 213220.
Yamamoto, N. and Ikagaki, K., (1998). A Study on the
Degradation of Coating and Corrosion on Ships Hull
Based on the Probabilistic Approach. Journal of Offshore
Mechanics and Arctic Engineering 120: 121128.
458

Vertical breakwater reliability analysis with direct integration method

Lei Zhang
College of Engineering, Ocean University of China, Qingdao, Shandong, China
Sheng Dong
College of Engineering, Ocean University of China, Qingdao, Shandong, China
C. Guedes Soares
ABSTRACT: In this paper, the direct integration method is applied to calculate the failure probability of the
vertical breakwaters. The calculation precision is high by giving a full consideration of the correlation of the
wave pressure and the uplifting pressure. The integral domain is processed by transforming an infinite integral to
a limited regional integral. The joint density function of the wave pressure and the uplifting pressure is selected
after comparison. The failure probability is calculated using a numerical integration method. The reliability index
can be estimated by checking point method proposed by Lind. Each non-normal random variable is individually
approximated by a normal distribution at the checking point. This method is recommended by Joint Committee
on Structural Safety (JCSS), which is also called JC method. In this paper the calculation results are compared
with the results calculated by both the variable-correlated JC method and the variable-uncorrelated JC method.
Calculation results of practical case show that the direct integration method and the variable-correlated JC method
have similar results, whereas the variable-uncorrelated JC methods calculation results are obviously lower.
1
INTRODUCTION
The Full Probability Method (Level III Method) is

theoretically correct approach for the calculation of
structural reliability. Unfortunately, the calculation of
failure probability is difficult. Therefore, the Approximate Probability Method (Level II Method) have
developed rapidly because of its simple calculation
and accuracy to meet most of the projects, in which
the First Order Reliability Methods (FORM) transform the initial set of variables into an independent
multi-normal set and use a linear approximation. For
a complete description of some of these methods
and some illustrative examples see Hasofer and Lind
(1974), Madsen et al. (1986), Ditlevsen and Madsen
(1996), or Melchers (1999).
The probability-based design of composite breakwaters has been studied by many researchers, such
as Christiani et al. (1996), Burcharth and Sorensen
(1998), Sorensen and Burcharth (2000), as well as
in the European project PROVERBS (see Oumeraci
et al., 2001) and the Working Group 28 on Breakwaters with Vertical and Inclined Concrete Walls PIANC
(2003).
In the reliability analysis for breakwater design,
different modes of failure have been considered, in
which sliding failure and over turning failure are two
main failure forms for vertical breakwaters (Castillo

et al., 2005). Reliability Code of Port projects structures of China recommends JC Method and Monte
Carlo Method to calculate the reliability of vertical
breakwaters.
The main loads are the horizontal wave pressure
and the uplifting pressure when calculating the reliability of a breakwater. The calculation accuracy of
variable-uncorrelated JC method is not high because it
does not consider the correlation of the wave pressure
and the uplifting pressure.
Multi-dimensional integration plays an important role in reliability calculation. More and more
researches are focus on study of multivariate distribution model with multidimensional, non-Gaussian
and correlative variables. Bitner-Gregersen and Haver
(1991) set up joint environmental model for reliability calculation. Coles and Tawn (1994) proposed
the negative bivariate logistic model to consider
the joint distribution of extreme wave height, wave
period and storm surge in the east England coast.
Athanassoulis et al. (1994) have proposed the application of the Plackett bivariate structure to fit bivariate distributions functions to significant wave height
and peak period data. Prince-Wright (1995) proposed
maximum likelihood models of joint environmental data for TLP design. Zachary (1998) introduced
459
the three-dimensional Logistic model to compute the

joint occurrence probability of wave height, its corresponding period and wind speed. Yue (2001) used
the bivariate mixed Gumbel distribution and Gumbel
Logistic distribution in hydrology probability analysis. Ferreira and Guedes Soares (2002) established
bivariate distributions of significant wave height and
mean wave period. To model the data after transformation, the authors have chosen the Box-Cox transformation in order to put the variables close to normally distributed. Bitner-Gregersen (2005) gave a joint
probabilistic description for combined seas. On the
basis of the reliable simultaneous databases and multivariate models established since the 1990-ties, use of
joint probabilities has started to be permitted in design
(API 1995, DNV 2007).
The breakwater in Qinhuangdao is taken as an
example to apply the direct integration method. Based
on the analysis of the wave frequency, Xie and
Liu (1992) proposed that Log-normal distribution or
Gumbel distribution can be used to describe the long
term distribution of the wave loads.
In this paper, both the wave pressure and the uplifting pressure use Gumbel distribution as the long
term distribution. Thus, the bivariate Gumbel Logistic Distribution is used as the joint density function
to describe wave pressure and uplifting pressure. The
direct integration method is carried out to reliability calculation for vertical breakwater. The results are
also compared with those computed by both variablecorrelated JC method and variable-uncorrelated JC
method.
STRUCTURAL RELIABILITY THEORY
Considering one random vector X = (X1 , X2 , . . . , Xn ),

of the basic variables and their joint density function is fX (x1 , x2 , . . . , xn ) = fX (X ), the structure limit
state function is expressed by Z = g(X ), the failure probability of this structure can be expressed as
follows
where F = { x|g(x) < 0} represents the failure domain

of the structure.
As described by Eq. 1, the key step of calculating
the failure probability is solving the multi-dimensional
integral. Because the joint density function of the variables is frequently unknown, Eq. 1 has no theoretical
solution. Multi-dimensional numerical integral needs
large computation work. It is inconvenient for practical use. Some scholars propose transforming irregular
integral domain into regulars. The joint density function is established based on multivariate joint normal
distribution density function.
In this paper, the load variables of the vertical
breakwaters are horizontal wave pressure and uplifting
pressure. Horizontal wave pressure and uplifting pressure can both be described by the Gumbel distribution
character as proposed by Xie and Liu (1992).
Estimated by moment method, the mean value

and the
variance of X are = 0.57722/ + and
= / 6, respectively, in which and are scale
parameters and location parameters respectively.
The bivariate Logic Gumbel Distribution is adopted
as the joint distribution function of horizontal wave
pressure and uplifting pressure. It can be expressed as
follows
where is the correlation

parameter of x and y,

it can be estimated by 1 1,2 , 1,2 is the linear correlation coefficient. The marginal distribution of x and
y can respectively be expressed as follows
where i , i (i = 1, 2) are the scale and location parameters of the variables. In the frequency analysis of
coastal and ocean engineering, these two unknown
parameters can be estimated by moment method, maximum likelihood method, least square method for a
given statistical series. The joint density function of x
and y can be defined by differentiating Eq. 3 on x and y.
where
Then the failure probability can be calculated by

using
where D is the failure domain.
460
NUMERICAL INTEGRATION
The infinite integral domain is transformed into definite integral domain due to the accuracy of integral.
where e is the accuracy of integral, a, b, m, and n are

given according to e.
The primitive function of g(x, y) is hard to get,
Eq. 7 can be approximately expressed as
where, Kx = n m/xi , Ky = b a/yi . xi and yi

are the integration steps of the X axis and Y axis,
respectively.
where (xiD , yjD ) represents this point (i, j) is in failure domain D. That means this point can satisfy the
equation Z(xiD , yjD ) < 0.
Failure probability Pf can be calculated accounting
for the fact that the integral value of g(x, y) in the real
space is 1.
Figure 1. The flow diagram of computation procedure for

failure probability Pf .
Failure probability calculating process is demonstrated

in Fig. 1.
Figure 2. The cross section of vertical breakwater at
Qinhuangdao Port.
RELIABILITY OF QINGHUANGDAO
BREAKWATER
The parameters of the vertical breakwater are obtained

by Qie and Li (2004). The caisson breakwater of a fuel
base at the Qinhuangdao port was built in 1974, whose
total length is 250 m. The inner side of the breakwater
was used as a wharf. Fig. 2 is the cross section of the
caisson breakwater. Each reinforced concrete caisson
of the breakwater is 12.5 m in length. It is divided into
6 compartments filled with stone and cobble. The elevation of the concrete superstructure is 3.5 m and the
wave prevention wall is 6.0 m. The elevation of the sea
bottom in front of the vertical wall is 5.5 m. According to the design document, the dangerous conditions
for the stability of the breakwater against sliding are
the design high water level and the extreme high water
level which are 1.75 m and 2.54 m, respectively. The
above two water levels are adopted in the calculation
of failure probability. By use of the measured wave
data for 23 successive years from Qinhuangdao ocean
hydrology station, the annual frequency analysis for
H1% and for the relevant average wave period T are
performed for the main wave direction of SE. The
50-year return period wave height H1% = 4.2 m, and
the relevant period T = 8.7 s. According to the wave
force formulae stipulated in the Code of Hydrology
for Sea Harbors (Ministry of Communications, China,
1998), the annual extreme values of H1% and the relevant period T are inputted for the computation of
annual extreme value series of wave pressure acting
461
Table 1.
Statistic parameters of loads.
Water level /m
2.54
1.75
G (KN/m)
MG (KN m/m)
P (KN/m)
945.46
5923.59
360.87
81.74
146.13
34.52
2011.37
457.39
1085.12
256.30
1017.75
6326.61
337.30
74.99
132.29
29.43
1752.81
395.22
982.43
218.56
PU (KN/m)
MP (KN m/m)
MPU (KN m/m)
Table 2.
P
P
PU
PU
MP
MP
MPU
MPU
The statistical parameters of loads are listed in

Table 1. The linear correlation coefficient between
the correlated variables is = 0.97. The reliability
calculation results are shown in Table 2.
From Table 2, it can be seen that the failure
probability calculated by direct integration method
are close to the failure probability calculated by
variable-correlated JC method, the failure probability calculated by variable-uncorrelated JC method are
obviously lower.
With the development of computer technology and

computation methods, the application of the direct
integration method will be applied much more widely
on the reliability of marine structures under the
action of multi-loads (wind, waves and currents).
At present, the joint density functions that can be
explicitly defined are not utilized universally. In case
when the joint density function is unknown, it can
be constructed by NATAF transformation. Then the
numerical integration method can be used to calculate
failure probability.
Calculation results.
Water level /m
Sliding
failure
Overturning
failure
2.54
1.75
CONCLUSIONS
Variableuncorrelated
JC method
Variablecorrelated
JC method
Direct
integration
method
Pf
1.37
0.0848
2.06
0.0198
Pf
1.18
0.1181
1.79
0.0369
Pf
1.18
0.1183
1.78
0.0372
Variableuncorrelated
JC method
Variablecorrelated
JC method
Direct
integration
method
Pf
3.50
0.00023
4.39
0.0000056
Pf
2.72
0.00329
3.45
0.00028
Pf
2.71
0.00340
3.42
0.00030
ACKNOWLEDGEMENTS
The study was financially supported by the National
Natural Science Foundation of China (50879085)
and the Program for New Century Excellent Talents
in University (NCET-07-0778).
REFERENCES
on the vertical breakwater for two high water wave

levels.
Horizontal wave pressure and uplifting pressure
can both use the Gumbel distribution. The bivariate
Logic Gumbel Distribution is adopted as the joint
distribution function of horizontal wave pressure and
uplifting pressure.
The limit state equation of the caisson breakwater
against sliding is expressed as follows:
where, G is weight of the breakwater. Pij is uplifting pressure. P is horizontal wave pressure. f is the
coefficient of friction between the break water and the
rubble foundation whose value is 0.6
The limit state equation of the caisson breakwater
against sliding is expressed as follows:
where, MG is the moment of stability induced by G.

MP is overturning moment induced by P. MPU is
overturning moment induced by PU .
API RP 2A-LRFD. 1995. Planning, Designing and Constructing for Fixed Offshore Platforms Load and Resistance Factor Design. ISO13819-2, (E).
Athanassoulis, G.A., Sarsoulis, E.K. & Belibassakis, K.A.
1994. Bivariate distributions with given marginals with
an application to wave climate description. Applied Ocean
Research, 16: 117.
Bitner-Gregersen, E.M. 2005. Joint probabilistic description for combined seas. Proc. OMAE2005 Conference,
Halkidiki, Greece.
Bitner-Gregersen, E.M. & Haver, S. 1991. Joint environmental model for reliability calculations. Proc. ISOPE91
Conference, Edinburg, UK, 1: 246253.
Burcharth, H.F. & Sorensen, J.D. 1998. Design of vertical
wall caisson breakwaters using partial safety factors. Proceedings of the 25th International Coastal Engineering,
ASCE, Copenhagen, Denmark, 2: 21382151.
Castillo, C., Minguez, R., Castillo, E. & Losada, M.A.
2005. An optimal engineering design method with failure rate constraints and sensitivity analysis. Application to
composite breakwaters. Coastal Engineering, 53: 125.
Christiani, E., Burcharth, H.F. & Sorensen, J.D. 1996. Reliability based optimal design of vertical breakwaters modelled as a series system of failure. Proceedings of the 25th
International Coastal Engineering Conference (ICCE),
ASCE, Orlando, Florida (USA), 2: 15891602.
462
Coles, S.G. & Tawn, J.A. 1994. Statistical methods for

multivariate extremes: an application to structural design.
Appl Statist, 43(1): 148.
Ditlevsen, O. & Madsen, H.O. 1996. Structural Reliability
Methods. New York: Wiley, Chichester.
DNV RP-C205. 2007. Environmental Conditions and
Environmental Loads. DNV, Norway.
Dong, S. & Kong, L.S. 2005. Introduction to marine
engineering environment. Qingdao: Press of Ocean
University of China. (in Chinese)
Ferreira, J. A. & Guedes Soares, C. 2002. Modelling
bivariate distributions of significant wave height and mean
wave period. Applied Ocean Research, 24(2): 3145.
Hasofer, A.M. & Lind, N.C. 1974. Exact and invariant second
moment code format. Journal of Engineering Mechanics,
100 (1): 111121.
Lind, N.C. 1977. Formulation of probabilistic design.
Journal of Engineering Mechanics, 103(2): 273284.
Liu,Y. & Xie, S.L. 1993. Determination of partial coefficients
of vertical breakwaters. Port Engineering Technology, (4):
1117. (in Chinese)
Madsen, H.D., Krenk, S. & Lind, N.C. 1986. Methods of
Structural Safety (2nd Edition). New York: Prentice Hall,
Inc., Englewood Cliffs.
Melchers, R.E. 1999. Structural Reliability Analysis and
Prediction (2nd Edition). New York: John Wiley & Sons.
Ministry of Communications. 1998. Code of design and
construction for breakwaters. Beijing: China Communications Press.
Nowak, A.S. & Collins, K.R. 2000. Reliability of Structures.
New York: The McGraw-Hill Companies, Inc.
Oumeraci, H., Kortenhaus, A., Allsop, W., de Groot, M.,

Crouch, R., Vrijling, H. & Voortman, H. 2001. Probabilistic Design Tools for Vertical Breakwaters. Balkema
Publishers, New York.
PIANC, W.G. 2003. Breakwaters with vertical and inclined
concrete walls. Report, Maritime Navigation Commission
(MarCom).
Prince-Wright, R. 1995. Maximum likelihood models
of joint environmental data for TLP design, Proc.
OMAE1995 Conference, 2.
Qie, L.W. & Li, Y.B. 2004. Reliability index of caisson
breakwaters for load variables correlated. China Ocean
Engineering, 18(4): 577584.
Sorensen, J.D. & Burcharth, H.F. 2000. Reliability analysis of
geotechnical failure modes for vertical wall breakwaters.
Computers and Geotechnics, 26: 225245.
The preparation group of Code of Reliability Design for
Sea Harbor Structures. 1992. Reliability design for sea
harbor structures. Beijing: China Communications Press.
Xie, S.L. & Liu, Y. 1992. Reliability based design of vertical
breakwaters. Proceedings of International Conference on
Port Development for the Next Millennium. Hong Kong:
Hong Kong Institution of Engineers: 419423.
Yue, S. 2001. The Gumbel logistic model for representing a
multivariate storm event. Advances in Water Resources,
24: 179185.
Zachary, S., Feld, G., Ward, G. & Wolfram, J. 1998.
Multivariate extrapolation in the offshore environment.
Applied Ocean Research, 20: 273295.
463
Wind & wave modelling

Assessment of wind and wave simulations for an enclosed sea

using satellite data
M. Bernardino & N. Salvao
Instituto de Meteorologia, Lisbon, Portugal
L. Rusu
ABSTRACT: An accurate simulation of the wave regimes of the Black Sea is of most importance in different
areas. The most efficient methodology to obtain wind and wave information for large areas is through the use of
physical models. In global models however. the spatial resolution of the forecasting fields is low and the wind
is underestimated, this problem is specially important in enclosed or semi enclosed seas as the Baltic, Adriatic
or the Black Sea. Also, the use of buoy data only allows a local assessment of the quality of the simulations.
The availability of satellite data in the region were the simulations were performed makes possible a broader
validation as several satellites regularly observe the Black Sea. The main purpose of the present work is the
evaluation of the high resolution wind and wave simulation using satellite information. The SWAN (Simulating
WAves Nearshore) model was implemented for the Black Sea region, forced by the 10 meter wind field simulated
by the WRF. The ERA-INTERIM reanalysis were used to drive the WRF.
INTRODUCTION
Since the Black Sea is an inland sea, the fetch is

considerably smaller than in open-ocean and, as a
consequence, the average wave conditions are usually less energetic. Nevertheless, very strong storms,
that are sometimes characteristic of that region, can
generate in the Black Sea waves comparable, in terms
of wave heights, with the big ocean waves.
It is well known the fact that, in enclosed basins
the waves have characteristics rather different than in
ocean environment. This is mainly due to their local
generation, and they depend more on the quality of
the wind fields (Ardhuin et al. 2007).
The most efficient methodology to obtain wind and
wave information for large areas is through the use of
physical models. The developments of third generation
global models by institutions like the European Centre for Medium Range Weather Forecast (ECMWF)
or the National Centre for Environmental Prediction
(NCEP/NCAR) made available global scale wind simulation that were used in wave models in order to
generate wave fields. However, the spatial resolution
of the forecasting fields is low and the wind variability
is underestimated, this problem is specially important in enclosed or semi enclosed seas as the Baltic,
Adriatic or the Black Sea (Cavaleri et al. 2010 and
references therein).
One of the reasons for this lack of accuracy in surface winds by global meteorological models is the
relatively coarse resolution with which they describe
the local geometry, in particular the orography

that surrounds the basin (Signell et al. 2005). Using
a regional atmospheric circulation model removes the
underestimation of the wind magnitude observed in
reanalysis data as the one simulated by NCEP/NCAR
global scale models (Komarovskaya et al. 2007).
The implementation of the SWAN model in the
Black Sea was described in some previous studies
(Rusu 2009, Rusu 2010a,b) when the necessity of
using wind fields with a higher accuracy and resolution able to provide a better representation of the wind
conditions at the entire sea level, occurred extremely
clear. In the above studies the model results were
validated by performing comparisons against in situ
data measured in various coastal locations from the
Black Sea basin. Unlike these studies the present work
performs a validation of the results for the entire basin
using remotely sensed data from AVISO.
The results of the wind and wave models were
interpolated in space with a bilinear interpolation, and
in time to collocate with the altimeter data. The corrections considered in relationship with the altimeter
data are those applied by Queffeulou (2004).
MODEL IMPLEMENTATION
2.1 The implementation of the WRF model

The Weather Research and Forecasting (WRF) Model
is a next-generation mesoscale numerical weather
467
prediction system designed to serve both operational forecasting and atmospheric research needs
(Michalakes et al. 2004). It features multiple dynamical cores, a 3-dimensional variational (3DVAR)
data assimilation system, and a software architecture
allowing for computational parallelism and system
extensibility. WRF is suitable for a broad spectrum
of applications across scales ranging from meters
to thousands of kilometers. The effort to develop
WRF has been a collaborative partnership, principally among the National Center for Atmospheric
Research (NCAR), the National Oceanic and Atmospheric Administration (the National Centers for Environmental Prediction (NCEP) and the Forecast Systems Laboratory (FSL), the Air Force Weather Agency
(AFWA), the Naval Research Laboratory, the University of Oklahoma, and the Federal Aviation Administration (FAA). WRF allows researchers the ability
to conduct simulations reflecting either real data or
idealized configurations. WRF provides operational
forecasting a model that is flexible and efficient computationally, while offering the advances in physics,
numerical, and data assimilation contributed by the
research community.
Used as a regional model WRF requires both initial and boundary conditions. This information is
usually obtained from global scale model running
operationally or from reanalysis data sets.
Over the past decade, reanalysis data has found
widespread application in many areas of research.
It is advantageous to use reanalysis data in certain
research areas where observational data are sparse,
when knowledge of the state of the atmosphere on a
uniform grid is required or as it the present case as
forcing for regional models.
The WRF system was implemented in the Black
Sea region using a grid resolution of 27 km but with a
terrestrial input data of 10 min resolution. The simulation used a total of 51 by 30 grid points. The physical
parameterization used in the simulation was the default
one and it is described in Table 1.
The ERA-Interim (Simmons et al. 2006), is the
latest reanalysis data produced by ECMWF, it uses
4D-variational analysis on a spectral grid with triangular truncation of 255 waves (corresponds to approximately 80 km) and a hybrid vertical coordinate system
with 60 levels. The ECMWF global model is used for
the forward integration in the 4D-variational analysis

and the temporal length of the variational window is
12 h. This reanalysis covers the period from 1989 to
present day.
Using the ERA-INTERIM data set interpolated to
0.5 , the WRF model was executed for 4 moths, from
January toApril 2002, with a 3 hours output resolution.
Among other variables the WRF provides wind speed
and wind direction at 10 meters and it was possible
to obtain the wind field over the area of the Black
Sea with a 27 km resolution and 3 hour resolution.
This data was later used as input for the wave model.
2.2 The implementation of the SWAN model

Presently, the most accurate estimation of the wave
fields is given by the spectral phase averaging models
(Rusu 2011). The SWAN model is considered as being
one of the state of the art models for wave generation
and coastal transformation, respectively. Like most
phase averaging models, SWAN solves the spectral
energy balance equation that describes the evolution of
the wave spectrum in time, geographical, and spectral
spaces.
The capacities of the SWAN model were substantially extended in the last years both in offshore and
nearshore directions and for this reason it seems to be
now an appropriate wave model for the sub oceanic
scales. The computational domain considered for the
present SWAN implementation in the Black Sea is
illustrated in Figure 1.
The system origin corresponds to the lower left
corner point and has the coordinates (27.5 , 41.0 ),
whereas the lengths are 14 in x-direction (longitude) and 6 in y-direction (latitude), covering both
the Black Sea and the Sea of Azov. In the geographical
space the computational grid was chosen identically
with the bathymetric grid with a spatial resolution of
x = y = 0.08 . In the spectral space, 36 directions
and 30 frequencies were assumed. The computations
were performed in the non stationary mode.
Model outputs with significant wave height fields
were produced hourly in order to be compared in the
following section to remote sensing observations.
Table 1. Physical parameterization of the WRF model

implementation for the Black Sea region.
Microphysics
WRF Single-Moment
3-class scheme
Long wave Radiation

Shortwave Radiation
Surface Layer
Land Surface
Urban Surface
Planetary Boundary layer
Cumulus Parameterization
RRTM scheme
Dudhia scheme
MM5 similarity
Noah Land Surface Model
no urban physics
Yonsei University scheme
Kain-Fritsch scheme
Figure 1. Bathymetric map of the Black Sea considered in

SWAN simulations and the satellite tracks over the sea in
the period analyzed (ERS2 yellow, Topex red, JASON1
green, GEOSAT Follow-On (GFO) black).
468
3 VALIDATION OF RESULTS
The wind magnitude was compared with local observations at the Gloria platform (44.3 N, 29.33 E) for
both the large scale simulation from the ECMWF and
from the WRF simulation, linearly interpolated to the
platform location. As the wind is locally measured
at 36 m the correction proposed by Hsu et al. (1994)
was used to reduce it to the 10 m.
The quality of the results is evaluated by performing
statistical analyses considering the following parameters: the average values for measurements (Mmed ) and
simulations (Smed ), the bias, root mean square error
(RMSE), scatter index (SI) and Pearsons Correlation
Coefficient (r). If Si represent the measured values, Mi
the simulated values and n the number of observations,
the fore mentioned statistics can be defined with the
relationships (1):
Table 2 shows the comparison between the observed

wind magnitude, reduced to 10 m at the Gloria platform and the corresponding simulations (ECMWF
and WRF) for the same location. It can be observed
that the bias is always positive but much higher in
for the ECMWF (1.64 m/s) than for the WRF simulation (0.06 m/s). The Gloria platform is near the
Romanian coast and a large scale meteorological
model is expected to under estimate the wind magnitude. The RMSE is also high (3.350 m/s) and the
correlation although positive is weak (51.7%). When
comparing the observations with the values obtained
using the WRF, it can be seen that results show a significant improvement. The model still underestimates the
observations but the RMSE is about one half of the previous value (1.754 m/s) and the correlation coefficient
is above 80%. It is clear that the use of a regional meteorological model introduced an increase in the quality
of the wind simulation at this specific location.
To illustrate the quality of the WRF simulation, the

time series of observed and simulated wind magnitude were plotted in Figure 2. It can be seen that the
simulation follows the observation but nevertheless
it underestimates the observations.
Observing the scatter plot in Figure 3, it can be
confirmed that this underestimation is not only due to
extreme events but it is present in winds faster than
4 m/s.
The comparison perform at a given location is however incomplete and allows only a local assessment
of the quality of the simulations. By comparing the
WRF and SWAN models results with satellite data is
possible a global validation of the modeling system
developed for the entire Black Sea basin.
The values of the statistical comparison between
the WRF simulated wind and the multi satellite data
are given in Table 3. The results are good, with global
correlation coefficients always above 70%, although
slightly lower than the value obtained for the local
comparison. Bias is negative with the model overestimating the observations and the RMSE is below
2 m/s. These results are similar to the ones observed
by Solvsteen & Hansen (2006) for the Baltic Sea and
Figure 2. Simulated (red) and observed (blue) wind

magnitude at Gloria platform.
Table 2. Statistics of the comparison between wind magnitude observed at Gloria and simulated for the same location
by the ECMWF and by the WRF.
ECMWF
WRF
Bias
RMSE
1.64
0.06
3.350
1.754
0.517
0.808
459
459
Figure 3. Wind speed scatter plot for WRF simulation at

Gloria platform. The blue line denotes the perfect fit to
model and observed values and the green line is a linear
regression to data set.
469
Table 3. Statistics of the comparison between simulated

wind field values and corresponding satellite data in the
Black Sea area.
Obs. data
Bias
RMSE
SI
Total data
ERS-2
TOPEX
Jason-1
GFO
0.3
0.5
0.3
0.4
0.08
1.94
1.93
1.96
1.99
1.97
0.31
0.31
0.34
0.31
0.34
0.74
0.78
0.74
0.71
0.75
19683
4752
6049
4291
4591
Figure 5. Significant wave height scatter plot for all data.

The blue line denotes the perfect fit to model and observed
values and the green line is a linear regression to data set.
the same range of accuracy with the results obtained

by Ardhuin et al. (2007) in Mediterranean Sea, and
Dykes et al. (2009) in Adriatic Sea.
The use of regional atmospheric and wave models provides a better representation of the wind and wave
regimes in the Black Sea.The accuracy values obtained
for the wind and wave simulation are in the same
range as the ones obtained by other authors for other
enclosed and semi-enclose seas. The under estimation
of observation that is typical of the use of large scale
models is in general corrected but it still persist in high
wind and waves.
Further work needs to be done, introducing higher
resolution grids in the meteorological model and
performing comparisons for longer periods of time.
Figure 4. Wind speed scatter plot for all data. The blue line
denotes the perfect fit to model and observed values and the
green line is a linear regression to data set.
significant wave height values and corresponding altimeter
data in the Black Sea area.
Obs. data
Bias
RMSE
SI
Total data
ERS-2
TOPEX
Jason-1
GFO
0.03
0.03
0.02
0.01
0.04
0.41
0.43
0.38
0.41
0.40
0.39
0.43
0.40
0.39
0.33
0.75
0.75
0.76
0.70
0.75
19683
4752
6049
4291
4591
CONCLUSIONS
REFERENCES
slightly better than the ones obtain by Signell et al.

(2005) for the Adriatic Sea.
Looking at the scatter plot represented in Figure 4,
we can see that the model underestimates high wind
magnitudes and overestimates the low ones.
The values of the statistical parameters computed
for the significant wave height are given in Table 4
for all multi-satellite data, but also separately for
each satellite. It is observed that all biases are negative, which means that the model overestimates the
measurements.
Nevertheless, the scatter plot of the significant
wave height (Figure 5) shows that, on the other hand,
some higher values are underestimated. The RMSE, SI
and r have in general reasonable values and they are in
Ardhuina, F., Bertotti, L., Bidlot, J.R., Cavaleri, L.,

Filipetto, V., Lefevre, J.M. & Wittmanne, P. 2007. Comparison of wind and wave measurements and models in the
Western Mediterranean Sea, Ocean Engineering 34 (34):
526541.
Cavaleri L., Bertotti L., Buizza R., Buzzi A., Masato V.,
Umgiesser G. & Zampieri M. 2010. Predictability of
extreme meteo-oceanographic events in the Adriatic Sea,
Q. J. R. Meteorol. Soc. 136: 400413.
Dykes, J.D., Wanga, D.W. & Booka, J.W. 2009. An evaluation
of a high-resolution operational wave forecasting system
in the Adriatic Sea, Journal of Marine Systems 78 (S1):
S255S271.
Hsu S.A., Meindl E.A. & Gilhousen D.B. 1994. Determining the Power-Law Wind-Profile Exponent under
Near-Neutral Stability Conditions at Sea, J. Appl. Meteor
33: 757765.
Komarovskaya, O.I., Efimov, V.V. & Barabanov, V.S. 2007.
Evaluation of wind fields for storm events in the black
Sea region using a numerical regional atmospheric
470
circulation model, Russian Meteorology and Hydrology

32 (7): 465469.
Michalakes, J., Dudhia, J., Gill, D., Henderson, T., Klemp, J.,
Skamarock, W. & Wang, W. 2004. The Weather Reseach
and Forecast Model: Software Architecture and Performance, proceedings of the 11th ECMWF Workshop on
the Use of High Performance Computing In Meteorology, 2529 October 2004, Reading U.K. Ed. George
Mozdzynski.
Queffeulou P. 2004. Long term validation of wave height
measurements from altimeters, Marine Geodesy 27:
495510.
Rusu, E. 2009. Wave energy assessments in the Black
Sea, Journal of Marine Science and Technology 14 (3):
359372.
Rusu, E. 2010a. Modeling of wave-current interactions at
the Danubes mouths, Journal of Marine Science and
Technology 15 (2): 143159.
Rusu, E. 2011. Strategies in using numerical wave models

in ocean/coastal applications, Journal of Marine Science
and Technology Taiwan 19 (1): 5875.
Rusu, L. 2010b.A system to evaluate the oil spills propagation
in the coastal environmental of the Black Sea, Journal of
Environmental Engineering and Landscape Management
18 (4): 288295.
Signell R.P., Carniel, S., Cavaleri, L., Chiggiato, J.,
Doyle, J.D., Pullen, J. & Sclavo, M. 2005. Assessment
of wind quality for oceanographic modeling in semienclosed basins, Journal of Marine Systems 53: 217233.
Simmons, A., Uppala, S., Dee, D. & Kobayashi, S. 2006.
ERA-Interim: New ECMWF reanalysis products from
1989 onwards, ECMWF Newsletter 110: 2635.
Solvsteen, C. & Hansen, C. 2006. Comparison of altimeter
wave and wind data with model and buoy data, Proceedings of 15 years in Progress in Radar Altimetry ESA
Symposium, Venice, 1318 March 2006.
471

Comparison of two wind and wave data sets from the North Atlantic
R.M. Campos & C. Guedes Soares
ABSTRACT: This paper compares two wave data sets: HIPOCAS and NOAA/NCEP. HIPOCAS is a hindcast developed under a European project and it is a result of the wave model WAM-cicle4 forced by REMO
surface wind fields. The wave data set of NOAA/NCEP is a forecast that has been continually performed by
WAVEWATCH III since 1997, forced by GFS surface wind fields. In order to compare the described data sets,
coincident time period and area were determined, which corresponds to 1997 to 2001, latitudes between 20 and
68 North and longitudes between 50 West and 04 East. The differences for the mean values and for the 99%
exceedance level were calculated. Small divergences between the two data sets are identified in standard weather
conditions in the North Atlantic. However, the longitudinal evolution of the differences is extremely marked in
both wave and wind fields. There is an important increase of differences for extreme situations, which is mainly
due to the differences of the meridional surface wind component, occurred during the winter.
INTRODUCTION
The knowledge of wave conditions, either as climatology or short-term forecast, is critical for all human
activities at sea, including shipping, fishing, oil extraction, naval operation and designs. Efforts have been
made in the past to produce consistent sets of wave data
that could be used to establish reliable wave conditions
for different locations and to obtain assessments of the
expected extreme and a long term trends. At the same
time, the operational wave forecasts have improved
substantially over the last two decades, as well as
data assimilation and computational wave modelling
capacity. Two different and broadly used wind-wave
data sets are compared in this paper.
Concerning this matter, the project Hindcast of
Dynamic Processes of the Ocean and Coastal Areas
of Europe (Guedes Soares, 2008) has performed a
44-year wave hindcast over the North Atlantic and
Coast of Portugal. The main intention was to improve
the wave data over the North EasternAtlantic and especially in the coastal waters of Europe, where the wind
fields are disturbed by the orography of the neighbouring terrain and the wave fields are modified by
complicated coastal bathymetry.
The wind input fields were generated by the regional
atmospheric model REMO (Jacob and Podzun, 1997;
Von Storch et al., 2000), which was forced with the
44-year atmospheric reanalysis, from 1958 to 2001,
carried out by National Centers for Environmental Prediction National Center for Atmospheric Research.
Additional description regarding wind modelling is
provided by Weisse et al. (2005). Afterwards WAM
model cycle 4 (WAMDI Group, 1988) was applied

using multi-grid mode. The procedure resulted in four
grids with increasing resolution towards the Iberian
Peninsula, as described by Pilar et al. (2008).
The second dataset compared is the forecast that has
been continually performed by the National Oceanic
and Atmospheric Administration National Centers
for Environmental Prediction NOAA/NCEP, which
is a well known centre that provides the parameters:
significant wave height, peak period, peak direction,
directional wave spectrum at some points and surface
wind fields. The period covered is from 1997 to 2011,
over many areas and resolutions.
The wave model applied is WAVEWATCH III
(Tolman, 2002) with 1 1.25 spatial resolution
and 3-hourly time resolution, forced by GFS surface
winds (nowadays with spatial resolution of 0.5 ).
An important question to highlight is the fact
that GFS-NOAA is a forecast (analysis), performed
day-by-day, while REMO-HIPOCAS is a hindcast
(reanalysis). An analysis has a commitment of producing results routinely, with available data assimilation
and a non-expensive computational cost method. For
a reanalysis, it is possible to apply the state-of-the-art
concerning wind and wave modeling. The modeling
time-step integration is no longer a limiting factor
in a hindcast as well. However, REMO is a climatic
model in which simplifications are made because it is
meant to be run for many years of data and the climatic averages will be an interesting output. These
slightly differences between procedures yield different results that can be noted when compared both
data-sets.
473
STATE-OF-THE-ART AND PREVIOUS

EVALUATIONS
Several evaluations and some comparisons regarding wave models and reanalysis have been performed
recently. Most of them considering the errors of the
wave fields as a consequence of the wind input, numerical wave model and resolution. Pilar et al. (2008)
describes HIPOCAS wave hindcast introduced above.
The final wind input has been modelled by REMO with
resolution of 0.5 and time step of 1 hour, instead of
rough resolution of NCEP/NCAR and ECMWF wind
hindcasts. The article has argued that this improvement
is important for an accurate description of the conditions in coastal areas and rapidly changing storms.
After WAM model simulation, the outcomes were
compared with buoys data which has shown an excellent agreement for locations in Portugal and Spain,
with bias less than 0.5 meters. For locations at the
entrance of English Channel and Irish Sea the calculations tend to overpredict the measurements. Gaslikova
and Weisse (2006) have also used HIPOCAS data to
apply downscaling techniques for extreme events. The
authors have found an overestimation of the significant
wave heights of HIPOCAS during severe storms and
they have identified waves with higher wave heights
occurring more often.
Ardhuin et al. (2007) performed a hindcast of wind
and wave conditions in the Mediterranean Sea for
two months. Four meteorological models (ECMWF,
ALADIN, COAMPS and ARPEGE) and three wave
models (WAM-cycle 4, VAG and WAVEWATCH III)
were used. The results were compared with satellite and buoys. They have also performed a threshold
analysis to understand the behaviour of simulations
on different storm intensities. Among the results, they
have found that the quality of wind and wave fields
degrades substantially close to coastal areas. For lower
resolutions the quality rapidly deteriorates, both off
and at the coast. They concluded that the accuracy of
wind modelling depends on how well the meteorological situation is defined. In well-extended storm areas
the models are more consistent to each other. According the article, winds are still the main source of error
in wave modelling. For low wave heights, WAM and
WAVEWATCH perform similar. However for growing significant wave heights there is a progressively
increasing underestimation by WAVEWATCH with
respect to WAM. Finally comparing WAVEWATCH
with buoys it seems to underestimate substantially the
largest wave heights in the Mediterranean Sea.
Feng et al. (2006) have studied the WAVEWATCH
III sensitivity to wind-forcing products. Four wind
fields have used to drive the model: NCEP/NCAR,
ECMWF, QuickSCAT and an enhanced ECMWF
wind field obtained by assimilating Special Sensor
Microwave/Imager (SSM/I)-retrieved wind speeds.
Comparing the tests, the WAVEWATCH simulation with NCEP/NCAR wind input accords best
with significant wave height measured from altimeter at both regional and global scales, although the
NCEP/NCAR reanalysis are consistently biased negatively with respect to altimeter or scatterometer
observed winds. The WAVEWATCH modelled with
the other three winds are all biased positively with the
highest bias of WAVEWATCH simulation with
QuickSCAT winds. They explain it is most likely
because WAVEWATCH model was formed using
winds similar to the NCEP/NCAR reanalysis winds.
Mackey et al. (2010) have not concerned exactly
wave modelling assessment but they have focused on
calibration techniques instead. An important conclusion is the seasonal and inter-annual effect on the hindcast evaluation, concluding that the wave simulation
results in distinct accuracies through the seasons.
3
DESCRIPTIONS OF DATA SETS AND

METHODOLOGY
In order to compare the described data sets, coincident

period and area were determined, which corresponds
to 1997 till 2001, latitudes between 20 and 65 North
and longitudes between 50 West and 0 . The resolution of NOAA/NCEP is 1.00 1.25 and HIPOCAS
is 1.00 1.00 (at the considered grid). In order to
avoid additional uncertainties due to interpolation,
the comparison has included only coincident grid
points. Hence the comparison regarding HIPOCAS
NCEP/NOAA has a final regular resolution of 5.00
degrees under a matrix with 110 coincident points
(10 lines and 11 columns) 89 non-excluded points.
Afterwards, the extreme values at each grid point
are considered, taking account the exceedance level
of 99%. The intermediate and shallow water points
as well as boundary points between ocean and ice
coverage, at high latitudes have been excluded. This
has been done to focus the analysis on deep water
and large-scale.
The results are shown concerning all comparisons
in time, named as mean comparison, and monthly
(monthly comparison). On the lower right corner of
each comparison maps is plotted the mean of 89 nonundefined values representing the total difference of
the area (spatial integration).
4 WIND AND WAVE COMPARISONS FOR
GENERAL CONDITIONS
This section deals with comparison of all time series.
The statistic parameters represent the differences,
percentage differences (or scatter index, which is the
difference divided by the reference value, between 0
and 1) and the correlation coefficient. Wind analyses
are divided into intensity and two components.
4.1 Mean wind and wave field comparison
4.1.1 Wind analysis
This sub-section represents the difference of REMO
minus GFS winds. Table 1 shows small differences of
474
Table 1.
Statistic results for general wind comparison.
Difference
(m/s)
Scatter
index
Table 2.
Correlation
Coefficient
Int.
Int.
Int.
0.60
0.19
1.47
0.43
0.42
0.80
0.55
0.90
0.05
Statistic results for general wave comparison.
Difference (m/s)
Scatter index
Correlation Coefficient
0.08
0.18
0.81
Figure 2. Comparison map representing the 89 point matrix

with the difference (on the left) and percentage difference
(scatter index on the right) for general wave analysis.
4.1.2 Wave analysis

As discussed above, the differences of significant wave
heights between data sets are very small. Table 2
shows NOAA/NCEP with very similar wave heights to
REMO. The scatter index is small and the correlation
is high.
Figure 2 illustrates small differences, with range
between 0.4 and 0.6 meters. The longitudinal evolution of differences is clear, with NOAA/NCEP higher
than HIPOCAS in the West and HIPOCAS higher than
NOAA/NCEP in the East, close to Europe. The disagreement is larger around 35 North, close to North
America.
4.2 Monthly wind and wave field comparison
with the difference (left column) and percentage difference
(scatter index, right column) for general wind analysis. First
line: Intensity. Second line: zonal component. Third line:
meridional component.
the intensity and zonal component (less than 0.20 m/s),

while the meridional component shows 1.47 m/s,
i.e., GFS with more intense winds than REMO. The
percentage difference (scatter index) is around 0.4
(40%) for intensity and zonal component, and very
high for V component (80%). The models are well
correlated to the zonal component and almost noncorrelated to the meridional component.
Analysing the spatial distribution (Figure 1) it is
possible to see, in a general sense, small differences
between REMO and NOAA/NCEP. The figures of
intensity show that GFS has higher values than REMO
at low latitudes and calm winds, while REMO is
stronger than NOAA/NCEP at high latitudes with
strong winds. The pattern of meridional wind differences demonstrates important divergences at high
latitudes, with scatter index around 0.8, i.e., REMO is
80% higher than GFS for the meridional component
of the wind. Analysing the range of differences, it is
possible to conclude that data sets are quite similar,
with intensity divergences less than 2 m/s.
This section shows the results for each month, analyzing the annual evolution. Graphics with values representing spatial integration of differences are adopted
instead of plotting twelve figures so as to save space.
June and December are shown in Figure 4 in order to
illustrate the spatial distribution of the differences in
the summer and winter.
4.2.1 Wind analysis
Figure 3 illustrates again the great agreement between
models for intensity and zonal component over the
year. This difference of intensity is a little higher
between April and June. The difference of meridional
component is constantly negative around 1.5 m/s.
Figure 4 shows that during the winter the differences of wind intensity increases close to United
Kingdom and North-eastern Atlantic. During the
summer the main divergences is in the South-eastern
part of the figure, close to Portugal and Africa.
4.2.2 Wave analysis
Figure 5 illustrates the differences between data
sets over the months. HIPOCAS and NOAA/NCEP
are almost equal between October and February.
NOAA/NCEP becomes higher than HIPOCAS in
March, which extends until September, i.e., mainly
during the spring and summer.
475

with the monthly wave difference for June (on the left) and
December (on the right), for general conditions.
Table 3.
Statistic results for general wind comparison.
Difference
(m/s)
Figure 3. Monthly evolution of general wind differences

between REMO and GFS.
Int.
Scatter
index
V
Int.
Correlation
Coefficient
V
Int.
2.24 2.39 12.06 0.30 0.20 0.71 0.14 0.32 0.02

with the monthly wind difference for June (on the left) and
December (on the right), for general conditions.
Figure 5. Monthly evolution of general wave differences

between HIPOCAS and NOAA/NCEP.
As commented above, Figure 6 shows the pattern of

differences varying with longitude, which is persistent
over the year. Figure 6 is extremely similar to Figure 4, illustrating the propagation of wind differences
through the wave field.
5 WIND AND WAVE COMPARISONS FOR
EXTREME SITUATIONS
5.1 Mean wind and wave field comparison
In this section the evaluation is performed using timeseries values above 99% significant level, applied
independently to 110 matrix points (89 non-excluded
points).
Figure 7. Comparison map representing the 89 point

matrix with the difference (first column) and percentage
difference (scatter index second column) for extreme wind
analysis. Fists line: Intensity. Second line: zonal component.
Third line: meridional component.
5.1.1 Wind analysis

Table 3 shows the increase of differences when
compared to the analysis concerning all data. The
mean difference of wind intensity for extreme events
is 2.24 m/s instead of 0.6 m/s, which represents an
increase of 273%. The divergence between REMO
and GFS for the meridional component of the wind
is even greater (12.06 m/s), the scatter index is high
and the correlation is weak.
Figure 7 shows the longitudinal effect of the difference in the intensity maps. Close to Europe, and
476
Table 4.
Statistic results for extreme wave comparison.
Difference (m/s)
Scatter index
Correlation Coefficient
1.24
0.19
0.26

with the difference (on the left) and percentage difference
(scatter index on the right) for extreme wave analysis.
especially at higher latitudes, REMO is up to 7 m/s

more intense than GFS. At the western part of the figure GFS winds tends to become similar or stronger
than REMO. The patter of the wind component differences is different, with a growing disagreement of
REMO with respect to GFS at higher latitudes, around
45 and 60 north. The meridional component of the
wind show differences about 16 m/s at this area.
5.1.2 Wave analysis
For extreme events, the wave response to the wind
model divergences is even more evident. Table 4 has
1.24 m/s of integrated differences between data sets,
the scatter index is reasonable and the correlation
coefficient is weak.
Figure 8 shows a large area with 2 meters differences. In general, HIPOCAS tends to have higher
waves than NOAA/NCEP in the North-eastern, while
NOAA/NCEP tends to be similar to HIPOCAS in the
South-western sector of the figure. The percentage difference figure shows greater divergences around North
Europe.
5.2
Monthly wind and wave field comparison
5.2.1 Wind analysis

Figure 9 shows all values above zero, i.e., REMO
winds stronger than GFS winds. The seasonality effect
is subtle for wind intensity and wind zonal component all months have values between 0 and 4 m/s.
The seasonality is evident for the meridional component of the wind, where the differences between data
sets reach values above 14 m/s during the winter. On
the other hand, the differences are considerably smaller
during the summer. Comparing Figure 9 with Figure 3
it is possible to conclude that the monthly variation
and seasonality are more evident for extreme events.
Figure 10 shows the spatial distribution of differences during June and December months. Due to small
number of events above 99% exceedance level, there is
a greater spatial variation in the figures. It is not clear
Figure 9. Monthly evolution of extreme wind differences

between REMO and GFS.

matrix with the monthly wind difference for June (on the
left) and December (on the right), for extreme conditions.
any area with much more divergences in June. However it is possible to see higher values between United
Kingdom and Iceland. In December, during the winter, the North-eastern Atlantic becomes preferable to
have greater differences that can reach 13 m/s.
5.2.2 Wave analysis
The evident seasonality of extreme winds analyzed
in the last section resulted in the annual cycle
of Figure 11. The mean differences of significant
wave heights between data sets are greater between
November and February, and very small (less than
1 meter) between June and August.
The first point to highlight in Figure 12 is the different range present in the figures of June and December.
During the June months, the differences of data sets
are distributed between 0.5 and 2.5 meters, which is
relatively small. One more time, even in the summer,
the area with higher disagreement is the North-eastern
Atlantic. There are a few points in the summer where
the waves of NOAA/NCEP are higher than HIPOCAS,
occurring in the Western Atlantic. During the winter
the wave differences have increased significantly. It
is possible to see some points with differences above
4 meters of significant wave height, which is a very
important difference to consider. The area where the
differences are above 2 meters is confined to the
North and North-eastern parts of the figure, with latitudes above 37 . The area where NOAA/NCEP is
higher than HIPOCAS is the South-western part of the
figure, below 40 .
477
Figure 11. Monthly evolution of extreme wave differences

between HIPOCAS and NOAA/NCEP.

matrix with the monthly wave difference for June (on the
left) and December (on the right), for extreme conditions.
CONCLUSIONS
From the analysis described it is possible to derive

some conclusions based on tables and graphics. For
standard weather conditions, taking the average of all
data, HIPOCAS and NOAA/NCEP are very similar in
the considered area. The small disagreement is mainly
due to the differences of the meridional component
of the wind, with 0.80 scatter index and a great variation among latitudes. According the literature, it is
well known that the circulation around the mid-high
latitudes is from West, with zonal winds. This effect
is a result of the rising arm of Ferrel cell. This is one
reason why the great disagreement of meridional wind
did not generate a great difference of wind intensity
and wave heights at the beginning.
An important aspect of wave comparison is the longitudinal evolution of differences, with NOAA/NCEP
higher than HIPOCAS in the West and HIOPOCAS
higher than NOAA/NCEP in the East. Considering all
data of the initial analysis, the effect of seasonality
is almost zero for the wind evaluation and small for
the wave evaluation. In the wave comparison it was
seen that the pattern of longitudinal evolution of the
differences is constant over the year, however there is
a tendency of intensification of divergences at high
latitudes during the winter.
When the focus is transferred to extreme events,
the analysis changes completely. During these events,
transient eddies become important, as well as the
intensification of low pressure system with cyclonic
circulation. The effect is an increased importance of

the meridional wind and, consequently, the associated
errors and differences. As a matter of fact, REMO
winds can differ up to 7 m/s from GFS winds, which is
propagated through the wave models to the significant
wave height fields. The result is a large area with wave
differences between data sets equal or above 2 meters
at high latitudes.
The physic aspects of the wave models, WAM and
WAVEWATCH must be also considered. As initially
described, Ardhuin et al. (2007) concluded that for
low wave heights WAM and WAVEWATCH perform
similarly and for growing significant wave heights
there is an underestimation by WAVEWATCH with
respect to WAM. That characteristic is well observed
in the present results, which amplifies the differences between wave data sets above the 99% level
of exceedance. Despite this fact, the main reason for
wave divergences is still the wind input differences,
which is possible to prove comparing Figures 8 and 7.
There is a direct response of wave model to the spatial distribution of great differences between REMO
and GFS winds. At the North-eastern part REMO
(and consequently HIPOCAS) has higher values than
GFS (NOAA/NCEP). This area, especially close to
North Europe, is considered the most important in
terms of data sets differences.
The seasonality becomes important in the extreme
analysis. The monthly variation of meridional wind
component and the significant wave height has a
wide variation. The great agreement between models
during the summer changes rapidly to an increase
difference during the winter. This leads to wind
intensity differences up to 13 m/s and wave height
differences up to 5 meters during the winter.
Finally it is possible to conclude that there is a
constant agreement between wind and wave data sets
for non-extreme events during all year. For extreme
events, the agreement is confined to April until
October and the divergences increase significantly
during the winter, especially at high latitudes.
ACKNOWLEDGEMENTS
The work reported here has been performed within
the EXTREME SEAS project www.mar.ist.utl.pt/
extremeseas, Design for Ship Safety in Extreme
Seas, which has been partially financed by the
European Union thought its 7th Framework program
under contract SCP8-GA-2009-24175.
REFERENCES
Abdalla, S. & Cavaleri, L., 2002. Effect of wind variability
and variable air density on wave modeling. Journal os
Geophysical Research, v. 107, no. C7, 3080.
Ardhuin, F., Bertotti, L., Bidlot, J., Cavaleri, L., Filipetto, V.,
Lefevre, J., Wittmann, P., 2007. Comparison of wind
and wave measurements and models in the Western
Mediterranean Sea. Ocean Eng., 34, 526541.
478
Feng, H., Vandemark, D., Quilfen, Y., Chapron, B., Beckley,

B., 2006. Assessment of wind-forcing impact on a global
wind-wave model using the TOPEX altimeter. Ocean
Eng., 33, 14311461.
Gaslikova, L. & Weisse, R., 2006. Estimating near-shore
wave statistics from regional hindcasts using downscaling
techiques. Ocean Dynamics, 56, 2635.
Guedes Soares, C., 2008. Hindcast of Dynamic Processes
of the Ocean and Coastal Areas of Europe, Coastal
Engineering, Vol. 55, pp. 825826.
Grotjahn, R., 2008. Different data, different general circulations? A comparison of selected fields in NCEP/DOE
AMIP-II and ECMWF ERA-40 reanalyses. Dynamics of
Atmospheres and Oceans 44, 108142.
Jacob, D., Podzun, R., 1997. Sensitivity studies with the
regional climate model REMO. Meteorol. Atmos. Phys.
63, 119129.
Pilar, P., Guedes Soares, C., Carretero, J. C., 2008. A 44year high resolution wave hindcast in the North Atlantic.
Coastal Eng. Vol. 55(11), pp. 861871.
Tolman, H. L., 2002, User manual and system documentation of WAVEWATCH III version 2.22. SAIC-GSO/EMC/
MMAB/NCEP/NOAA, MMAB Contribution No. 222.
Von Storch, H., Langenberg, H., Feser, F., 2000. A spectral
nudging technique for dynamical downscaling purposes.
Mon. Weather Rev. 128 (10), 36643673.
WAMDI Group, 1988, The WAM Model A third generation
ocean wave prediction model. J. Phys. Oceanogr., v. 18,
n. 12, pp. 17751810.
Weisse, R., Sortch, H. V., Feser, F., 2005. Northeast Atlantic
and North Sea storminess as simulated by a regional
climate model 19582001 and comparison with observations. J Climate 18, 465479.
479

Evaluation of the wave models SWAN and STWAVE in shallow water

using nested schemes
Marta Gonalves, Eugen Rusu & C. Guedes Soares
ABSTRACT: The objective of the study presented herewith is to evaluate two state of the art models based
on the spectrum concept, SWAN and STWAVE. This study is especially focused on the performance of the
two models in shallow and very shallow water. The target area considered is Porto Ferro, which is located on
the northwest coast of Sardinia. Comparisons were made as regards the spatial distribution of the main wave
parameters, based on the results obtained by both models. In the time domain comparisons are made for the period
614 February 2005. A wave-rider type buoy was used as offshore forcing source and the same nested scheme
was considered for both models to study the wave transformation towards the shore. As a general conclusion,
besides its simplicity, a major advantage brought by the STWAVE model is that almost no BC effect is introduced
as in SWAN. This is due to the scheme adopted for the wave propagation along a wave ray. Nevertheless this
feature represents an advantage mainly for one computational domain rather than in the case of nested schemes.
INTRODUCTION
It is a very important issue to have a good description

of wave propagation along the coast. For such purpose several equipment are located at strategic points
allowing a real-time observation of the various
parameters of sea waves, and these data are also
used to calibrate and validate the simulations that are
performed using several numerical models.
Nowadays, the wave models based on the spectrum
concept, or phase averaging models, are considered
among the most effective tools in estimating the wave
conditions in ocean and coastal environment.
The third generation wave models solve the energy
balance equation in spectral form and give a realistic estimation of the evolution of wave spectrum for
different conditions. They can be divided into generation models such as WAM (WAMDI Group, 1988)
and WW3 (Tolman, 1999), and transformation models such as SWAN (Booij et al., 1999) and STWAVE
(Smith et al., 2001).
In the last years many separate studies concerning
the reliability of SWAN and STWAVE models results
in the Portuguese nearshore have been carried out as
those of Guedes Soares et al. (2004), Rusu et al. (2005)
and Rusu et al. (2008) that used SWAN, and the work
of Mendona et al. (2005) for STWAVE.
Comparisons were also made between models
SWAN and STWAVE for the areas of Pinheiro da
Cruz, Figueira da Foz (Gonalves et al., 2008), bidos
(Gonalves et al., 2009) and Leixes (Gonalves et al.,
2010), revealing a good agreement between the two
models.
In this context, in addition to a comparative analysis between theoretical formulations and STWAVE
and SWAN, the objective of this work is to make
comparisons, in both space and time domains.
The target area selected for this study is the coastal
area of Porto Ferro, which is characterized by high
bathymetric gradients. This is located on the northwest
coast of Sardinia near the city of Alghero. A waverider type buoy is considered as offshore forcing source
for the SWAN and STWAVE model simulations. Two
Nortek gauges provided the nearshore wave measurements at about 1 meter water depth. Previous studies on
the wave conditions in the target area by performing
simulations only with the SWAN model were made
in Conley & Rusu (2005). Subsequently Rusu et al.
(2008) and Rusu & Guedes Soares (2010) made also
simulations with circulation modelling systems that
included also that area.
2 THEORETICAL FORMULATIONS
Both SWAN and STWAVE are third generation models focused on the wave transformation and based on
solving the spectral energy balance equation, although
they use different forms for the governing equations
and some formulations for the source terms are also
different.
The SWAN model is a third-generation stand-alone
(phase-averaged) wave model for the simulation of
waves in waters of deep, intermediate and finite depth.
This model simulates wave propagation in time and
481
space, shoaling, refraction due to current and depth,

frequency shifting due to currents and nonstationary
depth; wave generation by wind; nonlinear wave-wave
interactions (both quadruplets and triads); whitecapping, bottom friction, and depth-induced breaking;
blocking of waves by current. Diffraction is not explicitly modelled in SWAN but diffraction effects can
be simulated by applying directional spreading of the
waves. Reflections are also not included in SWAN.
STWAVE (STeady State spectral WAVE) is an
easy-to-apply, flexible, robust, half-plane model for
nearshore wind-wave growth and propagation, developed by the US Army Corps of Engineers (Smith
et al., 2001). STWAVE simulates depth-induced wave
refraction and shoaling, current-induced refraction
and shoaling, depth- and steepness-induced wave
breaking, diffraction, parametric wave growth because
of wind input, and wave-wave interaction and white
capping that redistribute and dissipate energy in a
growing wave field.
Phase averaged models solve an advection type
energy balance equation (Holthuijsen, 2007), that
describes the evolution of the wave spectrum in time,
geographical and spectral spaces:
The spectrum that is considered in most of the actual

wave models is the action density spectrum (N ), rather
than the energy density spectrum since in the presence
of currents action density is conserved whereas energy
density is not. The concept of wave action was introduced by Bretherton and Garret (1968), and the action
density is equal to the energy density (E) divided by
the relative frequency (). S from the right hand side
of the action balance equation represents the source
terms.
The radian relative frequency is related to the wave
number (k) by the dispersion relationship:
where g is the acceleration of gravity and d the water

depth. In the presence of currents the absolute radian
frequency () is given by the usual Doppler shift:
where the subscript i indicates the tensor notation for

the x and y axes and consequently (k)i , (U )i are the
corresponding components of the wave number and
current velocity vectors, respectively.
For coastal applications the Cartesian coordinates
are mostly used in the SWAN simulations and the
action balance equation becomes:
with the propagation velocities of the wave energy

in the spectral space defined as: C = ,
C = . The
components of the absolute group velocity are defined

using the standard formulation:
with (Cgr )i = /ki the components of the relative

group velocity.
The STWAVE model uses the governing equation
for steady-state conservation of spectral wave action
along a wave ray:
where C represents the absolute phase velocity

defined as:
with the relative phase velocity defined using the

standard formulation Cr = /k, is direction of the
current relative to the x-axis and is the wave direction (normal to the wave crest) relative to the same
x-axis.
The absolute group velocity vector defines the
direction of the wave ray ():
The left-hand side in the action balance equations

(4) and (6) represent their kinematic part. As regards
SWAN, the first term in equation (4) describes the
time evolution of the action density spectrum, the
second term denote the propagation of wave energy
in the two-dimensional geographical space, the third
term represents the effect of shifting of the radian frequency due to variations in depth and mean currents,
and finally the fourth term represents depth-induced
and current-induced refraction.
Unlike in SWAN, in STWAVE refraction and shoaling are implemented by applying the conservation of
wave action along backward traced wave rays. Rays
are traced in a piecewise manner, from one grid column to the next. The two-dimensional wave spectra are
set as input along the first grid column (the offshore
boundary). For a point on the second grid column, the
spectrum is calculated by back tracing a ray for each
frequency and direction component of the spectrum.
The ray direction () is determined by equation 8. The
same process is then repeated for the next columns.
3
CASE STUDY
As mentioned before, this experiment was carried out

in Porto Ferro, which is a nonconventional coastal
environment with strong bottom irregularities, located
on the northwest coast of Sardinia near the city of
Alghero, Figure 1.
Porto Ferro is a westward opening bay and it is fully
open to the strongest winter weather conditions in the
482
Figure 1. Target area Porto Ferro (Sardinia Island).

Figure 3. Comparison of the significant wave height
calculated for intermediate waters for both SWAN and
STWAVE.
Figure 2. Computational domain over the bathymetric map.

Table 1. Characteristics of the computational domains
defined for the two models.
Computational
domain SWAN/
STWAVE
Coordinates x y
Area A1
Area A2
(nesting)
Cartesian
Cartesian
Figure 4. Scatter plot for the significant wave height

calculated for intermediate waters.
both SWAN and STWAVE models against the offshore

buoy is also presented in Figure 3.
nf n nx*ny
50 50 m 30 36 133 230
25 25 m 30 36 124 166
western Mediterranean generated by the Mistral wind

blowing out of the Gulf of Lion.
A wave-rider type buoy was considered as offshore forcing source for the SWAN and STWAVE
model simulations. Two Nortek gauges provided the
nearshore wave measurements.
For this study two areas were considered, A1 as the
major area and A2 as a nested area, as illustrated in
Figure 2.
Simulations were carried out with both SWAN an
STWAVE models considering the same computational
domain as defined in Table 1.
A comparison of the significant wave height calculated for intermediate waters (about 50 m deep) for
EVALUATION IN THE TIME DOMAIN
Comparisons in the time domain were carried out for

the simulations performed with the two models against
the in situ measurements. For this study simulations
for a 9-day period starting on 2005/02/06/-10h were
considered.
As mentioned before, for this study, a wave-rider
type buoy was considered as offshore forcing source
for the SWAN and STWAVE model simulations and
two Nortek gauges provided the nearshore wave measurements (1 m and 0.90 m deep, correspondingly).
Figure 5 represents the comparisons of the model
simulations against the measured dada from the two
Nortek vectors.
The comparisons performed in the time domain
show in general a good agreement between the two
models as regards the significant wave heights.
483
Figure 7. Significant wave height fields and wave vectors

for the Swan and Stave simulations (Case study 1 Wave
input parameters: Hs = 4.66 m, Tp = 8.06 s and Dir = 284 ).
Figure 5. Comparisons of the computed (SWAN and
STWAVE) and measured (Nortek) significant wave height
for area A1.
Figure 8. Significant wave height fields and wave vectors for the Swan and Stave simulations (Case study 2
Hs = 1.23 m, Tp = 8.06 s and Dir = 284 ).
Figure 6. Comparisons of the computed (SWAN and

STWAVE) and measured (Nortek) significant wave height
for area A2.
As illustrated in Figure 5 and Figure 6 starting in

day 11/March for Nortek 1, and earlier for Nortek 2
the results of the significant wave height have a tendency to become constant, this is due to the wave
breaking, since the Norteks gauges are in the breaking
zone (1 m and 0.90 m deep, respectively).
Figure 9. Significant wave height differences for CS1 (at

the right) and CS2 (at the left).
models. In order to quantify these differences, the

significant wave height variation was computed:
5 EVALUATION IN THE GEOGRAPHICAL

SPACE
Figure 7 and Figure 8 illustrate the simulations results
with SWAN and STWAVE models in terms of significant wave height scalar fields and wave vectors.
For this study, two different cases of wave propagation were considered. In the first case (CS-1)
extreme conditions are considered with Hs = 4.66 m,
Tp = 8.06 s and Dir = 284 . For the second case (CS-2)
average conditions of wave propagation are considered
with Hs = 1.23 m, Tp = 8.06 s and Dir = 284 .
As shown in these figures, considerable differences
occur between the wave fields provided by the two
and is indicated in Figure 9.
DISCUSSION OF THE RESULTS
The study in the geographical space carried out in

Section 5 is focused on two situations often encountered in many coastal applications when only data in
one offshore point are known and these data are used
484
to assess the nearshore wave propagation in an entire

coastal region.
The results presented in Figure 7 and Figure 8
show that, by imposing constant lateral boundary conditions in SWAN and STWAVE models, less errors are
generally induced in the wave height field.
On the other hand, as illustrated by Figure 7 and
Figure 8, the affected area increases when the angle
between the mean wave directions relatively to the
normal to the coastline is enhanced.
The comparisons performed in the time domain
showed in general a good agreement between the two
models as regards the significant wave heights. As
illustrated in Figure 5 starting in day 11/March (for
Nortek 1) the results of the significant wave height
tend to become constant, this is due to the wave
breaking, since the Nortek gauges are in the breaking
zone.
7
CONCLUSIONS
Some general remarks in relation with the two wave

models evaluated in the present work can be also
made. It has to be mentioned first that both SWAN
and STWAVE are spectral wave models designed for
coastal areas and based on the integration of the
action balance equation with finite difference schemes
in order to provide realistic estimates of the most
relevant wave parameters.
SWAN is a more complex model and much many
options are available. The model can work either
in Cartesian or spherical coordinates, in stationary
or non stationary modes and its applicability has
been extended from nearshore areas to quasi oceanic
scales. SWAN accounts also for some physical processes considered relevant in shallow water as triads
and wave induced set up that are not included in the
current version of STWAVE.
On the other hand, STWAVE is a simpler model that
accounts for the main coastal processes (refraction,
diffraction, wind input, wave-wave and wave-current
interactions, dissipation by whitecapping and by depth
induced wave breaking) and is very effective for
many coastal problems.
In addition to the simplicity, a major advantage of
the STWAVE model is that almost no boundary conditions effect is introduced as in SWAN when using
simple schemes based on data provided by a single
source. As mentioned before, this is due to the fact
that STWAVE have a zero flux condition on the lateral
boundaries that provides better boundary conditions.
ACKNOWLEDGEMENTS
The work presented has been made in the scope
of the project NEARPORT (Development of a
real-time nearshore wave prediction system for the
Portuguese ports) supported by the Portuguese Foundation for Science and Technology under contract
PTDC/ECM/64373/2006.
REFERENCES
Booij, N., Ris, R. C. and Holthuijsen, L. H., (1999). A third
generation wave model for coastal regions. Part 1: Model
description and validation, J. Geophys. Res. 104, C4,
pp. 76497666.
Bretherton, F.P., Garrett, C.J.R., (1968). Wavetrains in
inhomogeneous moving media. Proceedings of the Royal
Society, England, Series A, Vol. 302, No. 1471, January,
529554.
Conley, D.C. and Rusu, E., 2005, Tests of wave shoaling
and surf models in a partially enclosed basin, Maritime
Transportation and Exploitation of Ocean and Coastal
Resources, C. Guedes Soares, Y. Garbatov & N. Fonseca
(Eds.), Francis & Taylor, London, pp. 10151021.
Gonalves, M., Pilar, P., Rusu, E. and Guedes Soares, C.,
2008, STWAVE model simulations on the Portuguese
nearshore, (in Portuguese), As Actividades Martimas e
a Engenharia, C. Guedes Soares & V. Gonalves de Brito
(Eds.), Edies Salamandra, Lisboa, pp. 427439.
Gonalves, M., Rusu, E. and Guedes Soares, C., 2009.
Comparison between the SWAN and STWAVE models
in the bidos nearshore, (in Portuguese), Actas das 6as
Jornadas Portuguesas de Engenharia Costeira e Porturia,
Funchal, (cd-rom).
Gonalves, M., Rusu, E. and Guedes Soares, C., 2010,
Comparison between the SWAN and STWAVE models
in the Leixes Port nearshore, (in Portuguese), Actas das
1as Jornadas de Engenharia Hidrogrfica, Lisboa.
Guedes Soares, C., Rusu, L. and Pilar, P., (2004). Wave
hindcasting on the Portuguese continental nearshore, (in
Portuguese), As Actividades Martimas e a Engenharia,
C. Guedes Soares & V. Gonalves de Brito (Eds), Edies
Salamandra, pp. 7382.
Holthuijsen, H., (2007). Waves in Oceanic and Coastal
Waters, Cambridge University Press, pp. 387.
Mendona, A., Teixeira, A.T., Oliveira, I.M., Silva, A.P.,
(2005), Characterization of the tranquillity conditions in
the area of the Leixes port entrance in view of short
period waves (10 to 18s), (in Portuguese), Actas 4as
Jornadas Portuguesas de Engenharia Costeira e Porturia,
Angra do Herosmo, (CD-ROM).
Rusu, L., Pilar, P., and Guedes Soares, C., (2005). Hindcasts of the Wave Conditions in Approaches to Ports of
the North of Portugal, Proceedings Fifth International
Symposium on Ocean Wave Measurement and Analysis
(WAVES 2005), Madrid, Spain, (cd-rom).
Rusu, L., Pilar, P., Guedes Soares, C., (2008). Hindcast of
the wave conditions along the west Iberian coast, Coastal
Engineering, vol. 55, pp. 906919.
Rusu, E., Conley, D.C. and Coelho, E.F., 2008. A Hybrid
Framework for Predicting Waves and Longshore Currents,
Journal of Marine Systems, Vol. 69, pp. 5973.
Rusu, E. and Guedes Soares, C., 2010. Validation of Two
Wave and Nearshore Current Models, Journal of Waterway, Port, Coastal, and Ocean Engineering, Vol. 136,
pp. 2745.
Smith, J., Sherlock, A., Resio, D., (2001), STWAVE:
Steady-State Spectral Wave Model. Users Manual for
STWAVE, Version 3.0; ERDC/CHL, US Army Corps of
Engineers; Washington, DC
Tolman, H. L., (1999). User manual and system documentation of WAVEWATCH III, Tech. Note 166,
NOAA/NWS/NCEP/OMB, 99 pp.
WAMDI group, (1988). The WAM model a third generation ocean wave prediction model, Journal Physical
Oceanography, Vol. 18, pp. 17751810.
485

Assessment of wave height extreme quantiles in North Atlantic using

regional frequency analysis
Cludia Lucas, G. Muraleedharan & C. Guedes Soares
ABSTRACT: An approach based on L-moments is applied to the daily maximum significant wave heights of
a region in the North Atlantic Ocean for the winter season in order to identify the discordant sites, formation
of homogeneous regions, selection of appropriate regional frequency distributions and estimation of regional
extreme quantiles. A previous study in the offshore region off Portugal demonstrated the usefulness of regional
frequency analysis (RFA). This study examines the performance of RFA for estimation of regional extreme
quantile as a true representative of at-site extreme quantiles. This study is applied to a larger ocean area, (40
60 N, 33 13W), using data from a period from 1958 to 1978 (21 years) in the winter season (December to
February) extracted from the HIPOCAS data base.
1
INTRODUCTION
The knowledge of the frequency of extreme wave

heights, i.e., the extreme quantiles, is highly essential for the design of civil engineering structures and
for the operation and safety of shipping. Estimation of
these frequencies is difficult because extreme quantiles are rare. This problem is resolved by the regional
frequency analysis (RFA), joining data from several
sites of identical wave characteristics for estimating
the event frequencies at any one site. L-moments are
utilized in the estimation process of regional frequency
analysis (Hosking and Wallis, 1997). From the different methods of regional frequency analysis that exist,
the regional probability weighted moments (PWM )
algorithm is considered to be more efficient (Cunnane,
1988). The L-moments are statistical quantities derived
from probability weighted moments and they increase
the accuracy and use of the analyses based on these
PWMs.
The initial step in regional frequency analysis is to
identify discordant sites, as those sites that are misfit
with the set as a whole and are to be discarded. The following step is to form sets of sites that approximately
satisfy the homogeneity condition, with frequency distributions that are identical apart from a site specific
scale factor. This is possible by grouping the sites into
different sub-regions. Then the selection of a regional
frequency distribution is necessary to fit the pooled
data from several similar sites identified by their sitestatistics. By this approach there will be no single true
distribution that applies to each site. Even when a distribution is found to be a close fit to the observed data,
there is no justification that future values will match
the past, particularly for the daily maximum significant wave height frequency distributions that give rise
to outlying values far away from the size of the data.

Therefore the aim is not to identify the true distribution but to find a distribution that will yield accurate
quantile estimates for each site.
It is essential to fit a probability model to the
observed daily maximum significant wave height for
the evaluation of the extreme quantiles of given return
periods.
The process of using data from several identical sites to estimate the frequency distribution of a
homogeneous region is known as regional frequency
analysis (RFA). Data is pooled from statistically identical sites (homogeneous region) other than the site
at which quantile estimates are required. Since more
information is used than in an at-site analysis using
only a single sites data, there is potential for greater
accuracy in the final quantile estimates. Even though
a region may be moderately heterogeneous, regional
analysis will still yield much more accurate quantile
estimates than at-site analysis (Lettenmaier and Potter,
1985; Lettenmaier et al., 1987; Hosking and Wallis,
1988; Potter and Lettennaier, 1990).
Lucas et al. (2011) applied regional frequency
analysis for the daily maximum significant wave
heights for the period extending from 1958-1978 and
extracted from 44 years North Atlantic Ocean wave
data (HIPOCAS) for 15 equally and closely spaced
(0.75 ) and finer grids (0.25 0.25 ) with geographical locations (35 45 N, 9.5 11W) for the
winter month January, offshore Portugal.
This study expands the previous one by studying a large region in the North Atlantic Ocean
extracted from the HIPOCAS database (Pilar et al.,
2008) and applies the regional frequency analysis
approach based on L-moments to identify discordant
sites, formation of homogeneous regions, selection of
487
appropriate regional frequency distribution and finally

and with a vital importance in this study, the estimation of extreme quantiles for designated return periods.
The comparability between the at-site quantiles of a
homogeneous sub-region and the respective regional
quantiles is also carried out to show that the regional
frequency analysis has a good performance on both,
due to the very similar values estimated.
2
EXTREME QUANTILE FUNCTION
Being Q a random variable with positive values, this

variable represents the magnitude of an event that can
occur at a given site. The frequency distribution is
the principal quantity of statistical frequency analysis because it specifies how frequently the values of
Q can occur. F(z) is the probability that the value
of Q is lower or equal to z:
The cumulative distribution of the frequency distribution is F(z), the quantile function or the inverse
function of cumulative distribution function is z(F)
which expresses the magnitude of an event in terms of
its non-exceedance probability F.
The extreme wave height in the upper tail of the
frequency distribution QT , is the quantile of the return
period T , is represented by equation:
Obtaining a good estimate of the quantile QT for a

practical return period is the aim of the frequency
analysis.
An alternative expression is suggested by
(Muraleedharan et al., 2007), a similar approach is
adopted for the analysis of extreme wave heights and
for return periods which is explained hereafter.
If Z is a daily maximum significant wave height
random variable with a distribution function of F(z),
Z1 , Z2 , . . . , Zn is a sample of n daily maximum significant wave heights. The probability for the largest
maximum significant wave height that does not exceed
zL is given by:
The mean value of the observation k is:
where E is the usual expectation operator.

The average number of observations included
between two adjacent wave heights that exceed zL is
E(k). As we are observing the largest wave height that
occurs every day, the number of observations E(k)
equals the number that lies between the appearance
of maximum significant wave heights that exceeds zL .
Therefore E(k) represents a period (T ) in which an
extreme quantile is observed. Therefore
L-MOMENTS OF PROBABILITY
DISTRIBUTIONS
Probability weighted moments of a random variable

Z with cumulative distribution function F() are
defined by:
For a distribution that has a quantile function z(u),

eq. (8) gives
In terms of probability weighted moments, L-moments

are given by
The L-moment ratios are defined as
This implies that the probability for the specified

value zL exceeded is:
A series of observations on the daily maximum significant wave height the probability that the kth observation is the first value that exceeds zL is given by the
geometric law:
where 3 is the L-skewness and 4 is the L-kurtosis
is the Lcoefficient of variation. It is analogous to

the ordinary coefficient of variation. L-moment ratios
assess the shape of a distribution independently of its
scale of measurement.
488
SAMPLE L-MOMENTS
Let z1,n z2,n . . . , zn,n be the ordered sample of

size narranged in ascending order of magnitude.
Then the unbiased estimates of r and r are respectively ar and br .
The sample L-moment lr is an unbiased estimator of

r . Then the sample L-moment ratios are defined as
sites with what would be expected for a homogeneous

region. The between-site variation in L-CV has a much
larger effect than variation in L-skewness or L-kurtosis.
Let the proposed region have N sites, with site i
having sample size ni and sample L-moment ratios
(i)
(i)
t (i) , t3 and t4 .
Let regional average L-CV, L-skewness and
L-kurtosis be denoted by t R , t3R and t4R weighted
proportionally to the sites sample size. I.e.
Calculate the weighted standard deviation of the

at-site sample L-CV s,
and the sample L-CV is:
STEPS IN REGIONAL FREQUENCY

ANALYSIS
The Regional Frequency Analysis, an approach based

on L-moments, introduced by Hosking and Wallis,
(1997) is briefly described here. The various steps in
regional frequency analysis are:
(a)] Screening of the data
Data should be screened for homogeneity (stationary)
over time. The NCEP global atmospheric re-analysis
wind data set were used to produce 44 years homogeneous and high resolution atmospheric data sets
(Guedes Soares, 2008). These have been used to
force the numerical wave model WAM to generate
the wave data of the ocean and the coastal areas of
Europe. Hence the wave data can be assumed to be
homogeneous over time.
(b) Identification of homogeneous regions
The next step in regional frequency analysis is the
assignment of sites to regions. A region consists of
sites whose frequency distributions are considered to
be approximately the same.Approximate homogeneity
is sufficient to ensure that regional frequency analysis is much more accurate than at-site analysis. Sites
with similar site statistics (L-moment ratios) constitute
a region.
(c) A heterogeneity measure
The heterogeneity measure compares the betweensite variations in sample L-moments for the group of
A kappa distribution is fitted to the regional average

L-moment ratios, it is a four parameter distribution
which includes the distributions generalized logistic,
generalized extreme-value and generalized Pareto as
special cases. Simulate a large number Nsim of realizations of a region with N sites, each having this kappa
distribution as its frequency distribution. The simulated regions are homogeneous and sites have the same
sample size as their original data. For each simulated
region V is calculated. The mean (v ) and standard
deviation ( v ) of the Nsim values of V are computed.
The heterogeneity H measure is calculated as:
The region is acceptably homogeneous if 2< H < 1,

possibly heterogeneous if 1 H < 2, and definitely
heterogeneous if H 2 (Hosking and Wallis, 1997).
If the region is not acceptably homogeneous, the
region could be divided into two or more subregions, some sites could be removed from the region
or a completely different assignment of sites to regions
could be tried.
(d) Choice of a frequency distribution
The selection of an appropriate frequency distribution by goodness-of-fit test, which involves computing
summary statistics from the data and testing with the
values from postulated distribution is the final stage
in regional frequency analysis.
Quality of fit is judged by the difference between
the L-kurtosis.
489
Table 1. Sample L-moment ratios and discordance measure

for December.
Figure 1. Area under study, 25 sites of grid size 0.5 0.5 .
A goodness-of-fit measure based on L-kurtosis is

given as:
Z fd Goodness-of-fit measure for the frequency

distribution.
t4R Regional average L-kurtosis
fd
4 L-kurtosis of the fitted frequency distribution
4 The standard deviation of t4R , which is obtained
by repeated simulation (Monte Carlo) of a homogeneous region whose sites have the fitted frequency
distribution and sample size the same as those of the
observed data. Z statistics were calculated from 500
simulations. The Z statistic has approximately a standard normal distribution. The criterion |Z| 1.64 then
corresponds to acceptance of the postulated distribution at a confidence level of 90%. The three-parameter
(location, scale, and shape) distributions subjected to
the goodness of fit test statistic are: generalized logistic, generalized extreme value, generalized normal,
Pearson type-III and generalized Pareto.
(e) Estimation of the frequency distribution
Estimation of the regional frequency distribution can
be achieved by estimating the distribution separately
at each site and combining the at-site estimates to give
a regional average.
6
Deep water daily maximum significant wave heights

from 25 sites (0.5 0.5 ) (Figure 1) of December, January and February, the rough winter season
in the North Atlantic Ocean off Portugal extend
from 19581978 (21 years) and extracted from 44
years HIPOCAS database (Guedes Soares, 2008) are
subjected to regional frequency analysis (RFA). The
Site
L-cv
L-skew
L-kurt
D(I)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Reg. Avg.
1.130
1.194
1.272
1.300
1.314
1.297
1.363
1.359
1.358
1.345
1.431
1.459
1.473
1.462
1.381
1.355
1.378
1.365
1.317
1.296
1.195
1.184
1.175
1.176
1.116
1.308
0.212
0.201
0.194
0.199
0.175
0.185
0.191
0.185
0.174
0.187
0.234
0.216
0.227
0.222
0.221
0.234
0.223
0.242
0.239
0.228
0.264
0.240
0.241
0.242
0.219
0.216
0.172
0.165
0.164
0.157
0.134
0.119
0.136
0.147
0.137
0.142
0.168
0.157
0.145
0.154
0.160
0.163
0.161
0.168
0.162
0.161
0.160
0.155
0.148
0.150
0.166
0.154
1.93
1.24
1.08
0.40
1.21
2.75
0.71
0.62
1.08
0.55
1.12
0.83
1.41
0.85
0.31
0.35
0.33
0.76
0.34
0.13
1.85
0.91
1.48
1.34
1.43
sample size of December and January is 651 (=21*31)

and of February is 593 (=28*16 + 29*5). The sample L-moment ratios computed are L-CV, L-skewness,
L-kurtosis and the ratio t5 . The regional analysis is
carried out with this at-site statistics and the at-site
mean. In December the region is homogeneous as
the heterogeneity measure H is lower than 2. There
are no discordant sites in December as the discordant measure D(I) < 3, critical value for a region with
more than 15 sites. The 25 sites form a region and
hence there are no sub-region formations in December.
Sample L-moment ratios are given in table 1. In
January, the region under investigation also forms a
region but with a discordant site as illustrated in Figure 2. The site 5 is a discordant site (D(I) = 3.65)
and is discarded. For February, several tests were performed with the 25 sites, they were grouped in various
ways according to the similarity in at-site statistics
(L-moment ratios).
The values for the sample L-CV, L-skewness,
L-kurtosis and the discordance measures for January
are shown in Table 2.
In February, there are two sub-regions, sub-region
1 which includes the sites 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
(H = 0.34) and sub-region 2 with the sites 11, 12,
13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24 and 25
(H = 0.63). The sub-region formations are shown in
Figure 3 and the sample L-moment ratios are given in
Table 3.
The three-parameter (location, scale and shape)
distributions subjected to the goodness-of-fit test
490
Figure 3. Sub-regions 1 and 2 for February.
Figure 2. Sub-regions for January.

for February.

for January.
Site
L-cv
L-skew
L-kurt
D(I)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Reg. Avg.
1.404
1.454
1.466
1.433
1.390
1.461
1.498
1.447
1.427
1.379
1.400
1.474
1.435
1.416
1.351
1.371
1.412
1.346
1.285
1.231
1.240
1.224
1.128
1.079
1.059
1.352
0.194
0.201
0.195
0.184
0.153
0.225
0.212
0.201
0.193
0.178
0.197
0.209
0.200
0.212
0.186
0.185
0.187
0.193
0.201
0.189
0.176
0.176
0.164
0.163
0.172
0.190
0.141
0.156
0.154
0.153
0.137
0.154
0.161
0.167
0.145
0.128
0.157
0.138
0.160
0.172
0.154
0.146
0.132
0.138
0.141
0.147
0.140
0.142
0.131
0.151
0.154
0.148
0.28
0.34
0.43
0.77
3.65*
1.82
0.78
1.15
0.20
1.26
0.25
1.39
0.47
1.59
0.22
0.12
1.00
0.54
1.40
0.56
0.36
0.39
1.51
2.11
2.41
Site
L-cv
L-skew
L-kurt
D(I)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Reg. Avg.
1.445
1.564
1.610
1.502
1.438
1.289
1.315
1.376
1.368
1.324
1.364
1.364
1.346
1.299
1.260
1.457
1.383
1.306
1.234
1.193
1.318
1.276
1.220
1.117
1.056
1.337
0.212
0.209
0.217
0.209
0.202
0.180
0.202
0.221
0.214
0.185
0.247
0.257
0.247
0.242
0.231
0.262
0.252
0.233
0.256
0.255
0.247
0.227
0.238
0.239
0.237
0.229
0.141
0.120
0.140
0.147
0.135
0.136
0.148
0.176
0.156
0.141
0.190
0.183
0.149
0.173
0.183
0.172
0.178
0.154
0.170
0.173
0.139
0.143
0.166
0.167
0.153
0.157
0.40
2.06
1.66
0.70
0.64
1.98
0.59
1.15
0.26
1.47
1.45
0.93
0.84
0.26
1.12
1.42
0.71
0.11
0.60
0.70
1.80
0.56
0.31
1.05
2.21
*Flagged value.
statistic are the generalized logistic, generalized

extreme value, generalized normal, Pearson type III
and generalized Pareto.
Generalized normal is the regional frequency distribution for the region in December. The regional
frequency distributions at sub-regions 1, 2 and 3 for
January are generalized normal but in the case of
February, for sub-region 1, Pearson type III is a better
fit and generalized normal for sub-region 2.
Tables 4, 5 and 6 present the at-site L-moment

estimates of the model parameters of the appropriate distributions of the sub-regions for its member
sites.
The generalized normal distribution is in fact the
three parameter lognormal, this distribution combines the lognormal distributions (positively skewed),
reflected lognormal distributions (negatively skewed)
and the normal distribution (zero skewness). The
491
Table 4. At site parameter estimates (by L-moments) for

December.
Table 6. At site parameter estimates (by L-moments) for the

distributions for sub-regions 1 and 2 in February.
Generalized normal
Sub-region 1
Pearson type III
Sites
Location
Scale
Shape
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
0.57
0.57
0.56
0.54
0.59
0.57
0.54
0.55
0.58
0.55
0.41
0.44
0.41
0.42
0.46
0.44
0.46
0.42
0.44
0.48
0.45
0.50
0.50
0.50
0.57
1.85
1.97
2.11
2.15
2.21
2.16
2.26
2.27
2.28
2.24
2.30
2.38
2.38
2.37
2.24
2.18
2.24
2.18
2.11
2.09
1.87
1.89
1.88
1.88
1.82
0.44
0.42
0.40
0.41
0.36
0.38
0.39
0.38
0.36
0.38
0.49
0.45
0.47
0.46
0.46
0.48
0.46
0.50
0.50
0.47
0.55
0.50
0.50
0.50
0.45
Table 5. At site parameter estimates (by L-moments) for

January.
Generalized normal
Sites
Location
Scale
Shape
1
2
3
4
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
0.52
0.48
0.49
0.53
0.42
0.44
0.48
0.51
0.56
0.51
0.45
0.49
0.47
0.55
0.55
0.53
0.54
0.54
0.59
0.61
0.62
0.67
0.69
0.67
2.33
2.40
2.43
2.39
2.37
2.45
2.39
2.37
2.31
2.32
2.42
2.37
2.32
2.25
2.29
2.35
2.23
2.12
2.05
2.08
2.05
1.91
1.82
1.78
0.40
0.41
0.40
0.38
0.47
0.44
0.42
0.40
0.37
0.41
0.43
0.41
0.44
0.38
0.38
0.39
0.40
0.42
0.39
0.36
0.36
0.34
0.34
0.35
Sub-region 2
Gen. Normal
Sites Location Scale Shape Location Scale Shape

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
2.69
2.91
3.01
2.80
2.67
2.37
2.44
2.58
2.55
2.44
1.28
1.26
1.31
1.26
1.22
1.09
1.12
1.34
1.29
1.12
0.41
0.38
0.41
0.45
0.49
0.33
0.39
0.46
0.44
0.47
0.43
0.49
0.49
0.53
0.56
2.17
2.15
2.14
2.07
2.03
2.28
2.19
2.10
1.94
1.88
2.09
2.06
1.95
1.79
1.69
0.51
0.53
0.51
0.50
0.48
0.55
0.52
0.48
0.53
0.53
0.51
0.47
0.49
0.50
0.49
Table 7. Extreme significant wave height for given return

periods estimated by generalized normal distribution for the
region in December.
T
(years)
Region (25 sites)

height (m)
10
20
50
100
19.51
21.54
24.28
26.42

region in January.
T
(years)
Region (25 sites)

height (m)
10
20
50
100
18.36
20.12
22.48
24.30
cumulative distribution function of the three parameter lognormal distribution is a generalized normal
distribution:
492

sub-regions in February.
T (years)
Sub-region 1
height (m)
Sub-region 2
height (m)
10
20
50
100
17.31
18.64
20.35
21.63
21.34
23.77
27.09
29.72
Table 11. Difference of at site extreme significant wave

heights (m) for the given return periods (T) from regional
quantiles estimated by generalized normal for the region in
January.
Return periods (T years)

quantiles estimated by generalized normal for the region in
December.
Return Periods (T years)
Sites
10
20
50
100
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
+2.75
+2.55
+1.97
+1.32
+2.69
+2.23
+1.04
+1.49
+2.17
+1.55
3.31
2.17
3.31
2.70
1.47
2.07
1.54
2.98
1.95
0.78
2.04
0.02
+0.15
+0.04
+2.53
+3.07
+2.91
+2.30
+1.57
+3.20
+2.64
+1.30
+1.83
+2.63
+1.88
3.81
2.41
3.75
3.04
1.67
2.42
1.75
3.50
2.33
0.94
2.61
0.13
+0.01
0.12
+2.79
+3.50
+3.40
+2.77
+1.91
+3.91
+3.22
+1.68
+2.31
+3.28
+2.35
4.52
2.74
4.37
3.51
1.94
2.92
2.05
4.24
2.87
1.18
3.45
0.37
0.21
0.36
+3.14
+3.84
+3.80
+3.15
+2.09
+4.50
+4.42
+1.98
+2.71
+3.82
+2.74
5.09
2.99
4.87
7.89
2.16
3.33
2.30
4.86
3.32
1.39
4.17
0.59
0.41
0.59
+3.41
where k, , are the shape, scale and location

parameters respectively.
Sites
10
20
50
100
1
2
3
4
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
0.86
2.03
1.75
0.55
4.06
3.49
1.98
1.09
+0.51
1.05
2.92
1.74
2.35
+0.34
+0.20
0.48
0.03
+0.17
+1.70
+2.34
+2.57
+4.34
+4.94
+4.74
0.96
2.30
1.95
0.57
4.71
3.99
2.24
1.21
+0.63
1.19
3.33
1.97
2.72
+0.40
+0.25
0.50
0.05
+0.13
+1.88
+2.64
+2.90
+4.88
+5.55
+5.29
1.10
2.67
2.22
0.58
5.62
4.67
2.60
1.37
+0.81
1.38
3.90
2.29
3.48
+0.49
+0.34
0.53
0.06
+0.07
+2.14
+3.07
+3.36
+5.64
+6.39
+6.06
1.22
2.96
2.44
0.59
6.36
5.22
2.90
1.50
+0.94
1.54
4.37
2.55
3.66
+0.56
+0.40
0.55
0.08
+0.01
+2.96
+3.40
+3.72
+6.24
+7.05
+6.65
The extreme quantiles of the designated return periods 10, 20, 50 and 100 for the regions are given in
Tables 7, 8 and 9. At site extreme quantile analysis
is also executed along with the regional frequency
analysis (RFA), to emphasize reliability on regional
extreme quantiles and to identify the sites with extreme
quantiles far away from regional values.
Tables 1012 give the difference of at-site extreme
quantiles of a region from its regional value. In
December and January, even though the 25 sites
together form a region and there are no any discordant sites, certain at-site extreme quantiles are
apart from its regional values. These findings indicate that the region is to be further divided into
sub-regions.
In February the at-site extreme quantile disparity
for the sites 3, 6, 12, 16, 17, 24 and 25 from their
regional quantiles are comparatively high. Coarser
grids (0.5 0.5 ) and greatly spaced (5.0 ) grids
attribute to this discrepancy. Taking into account the
results of this case study and the deep water rough winter weather conditions, it is necessary to consider finer
and more closely spaced grids for regional frequency
analysis.
493

quantiles estimated by Pearson type III (sub-region 1) and
generalized normal (sub-region 2) for the sub-regions in
February.
Return periods (T years)
Sub-regions
Sub-region 1
1
2
3
4
5
6
7
8
9
10
Sub-region 2
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
10
20
50
100
0.53
1.79
2.69
1.08
0.03
+2.44
+1.35
0.10
+0.30
+1.87
0.59
1.94
2.94
1.18
0.03
+2.68
+1.47
0.14
+0.30
+2.05
0.66
2.14
3.25
1.30
0.03
+2.98
+1.61
0.19
+0.31
+2.29
0.71
2.28
3.48
1.39
0.02
+3.22
+1.73
0.22
+0.33
+2.47
1.56
2.47
1.27
0.05
+1.41
4.52
2.30
+0.53
0.19
+0.59
0.84
+1.46
+1.44
+2.97
+4.10
+1.23
2.87
1.44
0.04
+1.67
5.21
2.64
+0.67
0.29
+0.59
0.96
+1.75
+1.66
+3.36
+4.63
2.06
3.43
1.68
0.02
+2.04
6.19
3.11
+0.88
0.46
+0.57
1.13
+2.16
+1.95
+3.90
+5.37
2.29
3.90
1.87
+0.01
+2.35
7.00
3.49
+1.05
0.60
+0.55
1.26
+2.51
+2.20
+4.33
+5.96
Figure 5. Extreme quantiles vs return periods for January.
Figure 6. Extreme quantile vs return periods for the points

of sub-region 1.
Figure 4. Extreme quantiles vs return periods for December.
The Figures 47 illustrate the extreme quantiles

versus return periods at the member sites of the
sub-regions and respective regional extreme quantiles.
The distributions are logarithmic.
Figure 7. Extreme quantiles vs return periods for the points

of sub-region 2.
494
CONCLUSIONS
Regional frequency analysis is found to be effective

in identifying the discordant sites, formation of
homogeneous regions, selection of appropriate
regional frequency distributions and estimation of
regional extreme quantiles. The mean of the at-site
extreme quantiles is its regional quantile. At-site analysis enables us to understand the spread of the at-site
values from its mean regional values. Selection of
finer and more closely spaced grids for rough winter weather conditions for regional frequency analysis
will reduce the disparities of at-site extreme quantiles
from its regional values. RFA also enables to perform
more accurate at-site analysis by appropriate selection
and estimation (by L-moments) of frequency distributions at the sites. This case study highlights the
necessity of performing regional frequency analysis
for clarification of the ocean wave climate of a region
of interest.
ACKNOWLEDGEMENTS
project EXTREME SEAS, Design for Ship Safety
in Extreme Seas, which has been partially financed
by the European Union thought its 7th Framework program under contract SCP8-GA-2009-24175.
(http://www.mar.ist.utl.pt/extremeseas/)
REFERENCES
Cunnane, C., 1988. Methods and merits of regional flood
frequency analysis. J. Hyd.,100, 269290.
Guedes Soares, C., 2008. Hindcast of Dynamic Processes
of the Ocean and Coastal Areas of Europe. Coastal
Engineering, Vol. 55, pp. 825826.
495
Hosking J.R.M. & Wallis, J.R., 1988. The effect of inter

dependence on regional flood frequency analysis. Water
Resources Research, 24, 588600.
Hosking, J.R.M. & Wallis, J.R., 1997. Regional Frequency
Analysis: An Approach Based on L-moments. Cambridge
University Press, Cambridge.
Lettenmaier, D.P. & Potter, K.W., 1985. Testing flood frequency estimation methods using a regional flood generation model. Water Resources Research, 21, 19031914.
Lettenmaier, D.P., Wallis, J.R. & Wood, E.F., 1987. Effect
of regional heterogeneity on flood frequency estimation.
Water Resources Research, 23, 313323.
Lucas, C., Muraleedharan,G. & Guedes Soares, C., 2011.
Application of regional frequency analysis for identification of homogeneous regions of design wave conditions offshore Portugal. 30th International Conference on
Ocean, Offshore and Arctic Engineering (OMAE2011),
June 1924, 2011, Rotterdam, The Netherlands. Paper
OMAE2011-50214.
Muraleedharan, G., Rao, A.D., Kurup, P.G., Unnikrishnan
N. & Sinha, M., 2007. Modified Weibull distribution for
Maximum and Significant Wave Height Simulation and
Prediction. Coastal Engineering, Vol. 54, pp. 630638.
Pilar, P., Guedes Soares, C. & Carretero, J.C., 2008. 44 Year
Wave Hindcast For The North East Atlantic European
Coast. Coastal Engineering, Vol. 55(11):861871.
Potter, K.W. & Lettenmaier, D.P., 1990. A comparison
of regional flood frequency estimation methods using
a resampling method. Water Resources Research, 26,
415424.

Maximum entropy modelling of extreme significant wave heights

from the Portuguese coast
V. Petrov & C. Guedes Soares
H. Gotovac
Department of Civil and Architectural Engineering, University of Split, Croatia
ABSTRACT: This paper presents maximum entropy (ME) as a plausible method for prediction of extreme
values with high return periods. Using peak over threshold approach, estimates of return values of significant
wave heights were obtained with generalized Pareto distribution (GPD) model and ME model. The results
showed similarity when comparing return values obtained from distributions derived from the maximization of
Shannons entropy and GPD model. A case study of the Portuguese coast data set is presented and the results
are compared with those obtained with the Annual Maxima method.
1
INTRODUCTION
In order to construct reliable offshore structures

or develop coastal areas one must have very good
knowledge about extreme sea wave heights. These
extreme sea heights are primary design loads on mentioned structures. Importance is greater due to evident
migration of population towards coastal regions.
The majority of the early work in this field include
defining the appropriate parametric model, such as
Weibull, Log Normal or others, for certain data set, and
then calculating the return values with low probability, i.e. high return periods, through the extrapolation
of fitted model, as described in Ochi (1992), Van
Vledder et al. (1993), Ferreira and Guedes Soares
(1999), Guedes Soares and Scotto (2001).
The reliability of extrapolation outside the bounds
of the parametric model covered with measured data is
always questionable and depends on the tail behaviour
of the proposed model. It is becoming increasingly
clear that common goodness of fit tests and techniques
are basically incapable of distinguishing two or more
models that are very close to each other within the sample range, while on the other hand those similar models
can produce very different values when extrapolated
to high quantile levels, meaning that they could show
different behaviour outside the range of the data even
though they could be almost the same within sample
range. Therefore extrapolating to extreme high levels, especially using models with different upper tail
behaviours should be obtained with caution, as stated
in Ferreira and Guedes Soares (1999).
In order to obtain reliable models of sea height
extremes, it is clear that one should put more emphasis
on the values in the upper range of the sample instead
of taking into account the whole data set which produce less reliable estimates in tail region. In that sense
Extreme Value Theory (EVT) proves to be useful since
it implies that by using enough upper range values from
sample one can produce distribution which could be,
according to this theory, only one of three possible
distributions, namely Gumbel, Frechet and Weibull.
Justification for this could be found in Leadbetter et al.
(1983). There is a long list of papers that applied EVT
to environmental data such as for example river levels,
sea heights, precipitation quantities, wind speed as can
be seen in Bermudez (2010) and references therein.
The simplest application of EVT is the Annual
Maxima (AM) method, which consist of fitting only
maximum values taken from succeeding intervals of
time to Generalized Extreme Value distribution as
described in Coles (2001). This method has been used
by, among others, Guedes Soares and Scotto (2001).
More appealing approach is Peak Over Threshold
(POT) method, which take into account excesses or
peaks over high enough threshold and fit them to
Generalized Pareto distribution. Statistical properties
of this methodology have been discussed in detail by
Davison and Smith (1990) synthesizing their early
work in Davison (1984) and Smith (1986).
Since extremes are by default rare it is reasonable
to take into account as much as possible data and
therefore POT approach could be advised as more
appropriate than AM, within the EVT framework as
discussed in Ferreira and Guedes Soares (1998) and
Guedes Soares and Scotto (2004).
With ultimate goal of determining return values
of significant wave heights with low probability, at
the moment widely accepted and used model is EVT.
It is considered to be the most appealing model since
497
it has advantage over other statistical models due to its

asymptotic properties, even though due to relatively
small data sets samples violate necessary prerequisites
in order to provide valid results according to EVT.
The Maximum Entropy (ME) method is presented
here as a powerful tool which even though is relatively
old and known method has not been widely used in
the field such as prediction of extreme wave heights.
It is argued that the ME could be an equally suitable
model as favoured EVT in tasks such as predictions of
significant wave heights associated with high return
periods.
Maximum entropy (ME) is non-parametric method
that provides the least biased estimation of distribution among all possible distributions based upon
the available information, i.e. sample moments. Over
last few decades maximum entropy has been the
theme of many researchers and number of algorithms
have been developed and applied in many fields,
including mechanics, geophysics, and econometrics.
This paper employs the approach of maximizing
Shannonss entropy by means of Chebyshev polynomials, as described in Bandyopadhyay et al. (2005)
and Gotovac et al. (2009).
The paper is organized as follows. In the next section
the EVT methods and ME method are presented. In
Section 3, the aforementioned techniques are applied
to a data set from the Portuguese coast in order to
produce predictions of extreme values of significant
wave height. In Section 4 finally some concluding
remarks are given.
2
2.1
METHODS FOR DISTRIBUTION

FUNCTIONS
Methods of extreme value theory
This section provides an overview of the distribution functions used according to EVT methodology. The EVT focuses on the statistical behaviour
of Mn = max(X1 , . . . , Xn ), where (X1 , . . . , Xn ) is a
sequence of independent random variables having a
common distribution F. In this application, X represents significant wave height measured on a regular
time-scale, meaning Mn stands for maximum value
over n time units of observation.
If there is a sequence of constants {an > 0} and {bn }
such that:
for a non-degenerate distribution function G, then G

is a member of the Generalized Extreme Value (GEV)
family.
The model has three parameters: a location parameter ; a scale parameter ; and a shape parameter
. According to the convergence of (Mn bn )/an the
parent distribution F is said to belong to the domain
of attraction of one of the three classes of distribution, widely known as Gumbel (with = 0 interpreted
as the limit of Equation 2 when ), the Frechet
( > 0) or the Weibull ( < 0).
The AM method consists of fitting the GEV distribution in Equation 2 as an approximation to the
distribution of maxima over the sample of the k maximum values obtained by blocking data into sequences
of observations of length n for some large value of n,
i.e. generating series of block maxima.
Blocks usually correspond to the reference period
such as annual maxima or monthly maxima. This
method can still be used when the process remains
stationary and has a weak dependence structure.
Estimates of return values (extreme quantiles) associated with the return period 1/p are then obtained by
inverting Equation 2
where G(zp ) = 1 p. The value zp is expected to be

exceeded on average once every 1/p years or more
strictly speaking zp is expected to be exceeded by
the annual maximum in any particular year with
probability p.
For the asymptotic arguments to hold within AM
approach, the data taken from blocks of length
one year should be independent, which is a plausible assumption. There are many techniques that are
proposed for parameter estimation in EVT, but in this
paper the maximum likelihood methodology is followed. In assessing the accuracy of the maximum
likelihood estimators appropriate profile likelihood
functions were used as described in Coles (2001).
The POT method is the extension of the classical
block maxima approaches, where the key idea is to
select the data not from the series of time blocks but
from the upper segment of data set by using the threshold in order to extract exceedances or just peaks over
a threshold. POT method is based on the fact that if
block maxima have approximating GEV distribution
given by Equation 2, then excesses over high enough
threshold u have the corresponding approximate distribution within the Generalized Pareto distributions
(GPD) family, which has representation
There is consistency between the GEV and GPD

defined on {x:1 + (x )/ > 0} where < < , model, meaning that the parameters can be related by
= and = + (u).
> 0 and < < ,
498
In order to obtain return level first exceedances

rate of threshold should be specified as = P(X > u),
which is related to the GEV parameters by
It follows that the mean crossing rate of any level

(x > u) is
where constraints are known statistical moments of

arbitrary basis functions (hj (x); j = 0, . . . , m). Optimization problem can be solved by introducing the
Lagrangian function and corresponding multipliers j .
So the optimization problem is reduced to finding the

maximum of the Lagrangian function with the respect
to all possible functions f that satisfy the constraints
given by Equation 9b. The distribution obtained with
ME has following analytical expression
By setting the Equation 6 equal to 1/m we obtain the

m-year return level qm as
According to the EVT, the GPD model is an appropriate model for the distribution of significant wave
heights excesses over a suitable chosen high threshold value. The parameters of GPD are estimated using
maximum likelihood method as in the case of GEV
model.
2.2
Maximum entropy method
The Maximum Entropy principle (ME) is an efficient statistical tool which serves for characterization
of distributions in the case where there is a known
set of statistical moments that describes the data set.
Shannons information entropy can be defined as
following
where ln(f(x)) is a quantity of information and f is

distribution function. In that way the defined entropy
poses some desired properties such as: (i) higher
uncertainty causes higher entropy, (ii) decreasing
probability causes an increase of information and
(iii) total entropy of two independent events is equal
to the sum of the individual entropies.
According to the ME definition, the distribution
with the highest entropy is selected to give the most
information among all other possible distributions
that satisfy known constraints, as described by Jaynes
(1957). ME is defined as the following optimization
problem
Further, for a given constraints the ME problem

requires solving the Lagrangian multipliers from the
nonlinear system with (m + 1) equations, which is
formed by introducing the ME distribution in Equation
11 into constraints given by Equation 9b.
The basis functions used in this paper belong to
the family of Chebyshev orthogonal polynomials, as
described along with the corresponding numerical
algorithm in papers of Bandyopadhyay et al. (2005)
and Gotovac et al. (2009).
To conclude, the ME method ensures selection of
the least-biased estimate among all possible distributions that satisfy information provided in terms of
constraints as defined in Equation 9b.
3 APPLICATION TO A PORTUGUESE COAST
SITE DATA SET
This study has been performed on the data set obtained
by hindcasting technique from the wind data sets,
at three hourly intervals over the significantly large
period from January 1958 until December 2001.
Measurement points are located near the coast of
Portugal.
3.1 Results of annual maxima (AM) method
The initial representation of data related to the AM
method is given on Figure 1. By assuming that the
pattern of variation has remained constant over the
observation period, the necessary prerequisite of data
independency has been satisfied, so one could apply
the GEV distribution to data.
Table 1 presents the parameter estimates along with
the obtained standard error in parentheses by applying maximum likelihood method, as well as the point
estimates of the 25, 50 and 100 years return levels,
with 95% confidence interval obtained from the
profile log-likelihood.
For this model the log-likelihood is 72.4. Figure 1
shows diagnostic plots, as discussed by Coles (2001)
499
Table 1.
Results of fitting the GEV distribution.
Parameter estimates
Return values
25
50
100
8.254
(0.205)
1.205
(0.148)
0.219
(0.118)
11.02
(10.49,
12.35)
11.41
(10.81,
13.29)
11.75
11.05,
14.26)
Figure 2. Mean residual life plot for the data from Portuguese coast.
Figure 1. Diagnostic plots for GEV fit to the data from

Portuguese coast.
and produced by the package of Coles and Stephenson

(2006) run in R Development Core Team (2009). Such
plots are now used routinely, and so have not been
edited here; detailed explanation is provided in the
caption.
These diagnostic plots indicate that the fitted GEV
model is satisfactory. The probability plot seems to
be more convincing than the quantile plot, but nevertheless the return level plot with confidence intervals
along with histogram show little difference between
empirical and fitted values from the model, indicating
a good fit. The goodness of fit tests persuasively shows
that one should not doubt in validity of the presented
model.
Approximate 95% confidence intervals for the
parameters of the GEV model are [7.85, 8.66] for ,
[0.92, 1.49] for , [0.45, 0.01] for . If we focus our
attention to the confidence interval gap of the shape
parameter , it could be concluded that the hypothesis
that 0 is rejected by a test at any conventional level
of significance leading to conclusion that the Weibull
model could be reasonable choice to fit this data set.
Furthermore, for this distribution it is possible to
make an inference on the upper end-point. The maximum likelihood estimate is xm = / = 13.75
Figure 3. Parameter estimates against threshold for the data

from Portuguese coast.
3.2 Results of peak over threshold (POT) method

The initial step in the POT approach is to scrutinize
the data set and carefully select a reasonable threshold
value. The issue of threshold choice is analogous to
the choice of time frame for block maxima selection
in AM approach, where going too far toward the small
or big value can produce bias or variance.
Two different methods are applied: first the mean
residual life plot is analyzed and secondly the stability
of parameters estimates is assessed, based on fitting
models across the certain threshold range.
Since selection of an appropriate threshold value
is performed on a visual basis therefore can have a
range of associated errors. These visual procedures
require prior knowledge of the accurate interpretation of threshold choice plots to achieve a satisfactory
model fit.
Figure 2 shows the mean residual life plot with
approximate 95% confidence intervals. It seems that
selection of threshold range from u 6 until u 8.5
could be interpreted as optimal since there is some
evidence of linearity within that range.
Further support for this choice of threshold span
is provided with plots presenting variations of
parameters estimates, as shown on Figure 3.
500
Table 2. Extremal index, mean cluster size, number of

cluster peaks and number of clusters for various thresholds.
u
6.5
7.5
8.5
u
1
u
np
nc
0.149
6.70
256
1716
0.170
5.87
179
1050
0.167
6.00
107
642
0.190
5.27
75
395
0.233
4.29
58
249
0.220
4.55
33
150
Table 3. Results of fitting the GPD to the peaks of clustered

excesses.
u
25
50
100
1.155
(0.112)
0.011
(0.074)
6.5
1.086
(0.136)
0.012
(0.100)
1.434
(0.204)
0.189
(0.106)
7.5
1.463
(0.236)
0.256
(0.116)
1.117
(0.238)
0.143
(0.168)
8.5
1.516
(0.350)
0.419
(0.169)
11.59
(10.62,
13.62)
11.66
(10.61,
14.25)
11.10
(10.53,
12.59)
11.03
(10.55,
12.25)
11.07
(10.50,
12.99)
11.06
(10.72,
11.97)
12.35
(11.08,
15.16)
12.45
(11.06,
16.26)
11.52
(10.84,
13.61)
11.38
(10.85,
13.09)
11.52
(10.82,
14.59)
11.32
(10.94,
12.66)
13.10
(11.49,
15.11)
13.26
(11.45,
15.74)
11.90
(11.09,
14.68)
11.68
(11.08,
13.95)
11.93
(11.07,
14.38)
11.52
(11.30,
13.38)
Figure 4. Diagnostic plots for the 6 m threshold excess

model fitted to the data from Portuguese coast.
Tables 2 and 3 present the results of GPD fitting

across the range of thresholds. In Table 2 the number
of cluster peaks selected out of the number of data
declared as a cluster, estimates of the extremal index ,
which has an informal interpretation as the reciprocal
of the average size of clusters are showed.
Diagnostic plots for fitted GPD for range of threshold values are presented on Figures 4 to 9. As can be
noticed on Figures 49 the return level plots tend to
have more curved shape as threshold level rises, which
is a logical phenomenon since shape parameter of the
GPD is also rising, as can be seen in Table 3 and on
lower graph in Figure 3.
In order to decide which set of diagnostic plots
should be the most plausible one it can be only pointed
out that return values between u = 7 and u = 8.5 are
almost the same, which could lead to the conclusion
that the selection of one ideal threshold is irrelevant
within this range. Furthermore, it can be emphasised
that for low thresholds u = 6 and u = 6.5 the shape
parameter is close to 0 meaning that Weibull model is
not an adequate model, which is logically not likely
since natural phenomena such as wave heights should
have its upper bound and therefore results obtained
with higher threshold values could be more reasonable
estimates.
Figure 5. Diagnostic plots for the 6.5 m threshold excess


501



Figure 10. Probability density functions produced with

maximum entropy (ME) for range of thresholds on cluster
peaks data from Portuguese coast.
3.3 Results of maximum entropy (ME) method

The intention of using maximum entropy method in
prediction of high levels of significant wave heights,
related to high return periods, e.g. 100 years, was to
present ME model as a suitable tool for such purposes even though due to its comprehensive numerical
algorithm is computationally more expensive than the
procedures within framework of EVT.
The numerical algorithm of ME model used in this
paper is defined on finite domain, which means that
based on information given by constraints via statistical moments obtained directly from sample the
least-biased estimate of real pdf is generated on predefined domain. Therefore in order to gain as much
as possible accuracy it was decided to apply it only on
peaks over thresholds data sets, since they are smaller
domains than the domains for annual maxima sets.
According to the results of the EVT methods, the
ME algorithm were run on the data sets comprised
of cluster peaks, the same data sets that were used
for POT approach, with lower bound defined naturally
with threshold value and upper bound set to 15, which
corresponds with the largest upper limit of 95% confidence interval for 100 years return levels, obtained
within POT approach, as presented previously in
Table 3.
In order to assure validity of selected upper bound
we tested various upper bound values but everything
higher than proposed did not cause variation in 100
years return levels obtained with ME model, i.e. proved
that 15 is high enough upper bound.
For constraints 4th order moments were used, assuring that obtained 100 years return levels are produced
with good accuracy, i.e. differences between the results
produced with higher order moments are so small that
are irrelevant for practical engineering purposes.
As a final product of the ME algorithm, probability density functions (pdf) are produced for various
threshold values, as presented on Figure 10. By integration pdf values along the domain it is easy to
produce cumulative distribution function (cdf) which
is more suitable if one is interested in determining the
extreme values associated with low probability values.
502
Table 4. Return levels obtained with maximum entropy on

the data from Portuguese coast.
u
25
50
100
6
6.5
7
7.5
8
8.5
11.13
11.07
11.08
11.13
11.05
11.10
11.43
11.41
11.42
11.46
11.36
11.44
11.68
11.72
11.72
11.74
11.63
11.71
Figure 11. Cumulative distribution functions produced

with maximum entropy (ME) for range of thresholds on
cluster peaks data from Portuguese coast.
Figure 13. 100 years return level obtained with AM, ME

and POT method, along with 95% confidence interval, on
the data from Portuguese coast.
Figure 12. Return level plots produced with maximum

entropy (ME) for range of thresholds on cluster peaks data
from Portuguese coast.
varying less than competitive GPD model across the

given threshold range.
In Table 4 are presented quantiles corresponding
to values related to 25, 50 and 100 years return period
for various thresholds obtained with ME model.
If those values are compared with ones in Table 1
and 3 there is a reasonable agreement between the
obtained results with maximum entropy model and
models within extreme value theory.
4
Therefore on Figure 11 are presented cdf(s)
obtained with ME method for the same range of
threshold values used in previous analysis with EVT
models.
As can be noticed high quantiles derived from
all examined samples fell almost on the same line,
leading to conclusion that regarding the selected
threshold value quantiles that correspond with 100
year return levels falls within very narrow band,
That indicate that ME model is very robust in terms
of selection of threshold value for POT data sets.
More detailed and better perspective provides
Figure 12 where return levels are showed for range
of threshold values. Since there is significant similarity identified among presented return level plots it
can be concluded that ME model is very stabile and is
DISCUSSION
Figure 13 summarizes the results of prediction of

extreme wave heights related to the 100 years return
period from all methods. By comparing the values it
is clear that all methods produce very similar results,
if neglecting the results for threshold values of u = 6
and u = 6.5 in POT method, where 100 value vary
By reviewing the literature, there is general belief
that models used within EVT produce the best estimates in prediction of extreme wave heights, based
on its superiority in comparison to all other parametric models, but Figure 12 indicate that non parametric
ME model is not only capable of producing the same
high quantiles but it produce them in very stabile
manner. Thus, ME could be labelled as more suitable model for prediction of extreme wave heights.
503
In particular, ME model seems to be very robust to

the selection of threshold value, which is well known
weakness of POT method.
5
CONCLUSION
In this paper two extreme value methods (AM and

POT) and the non parametric maximum entropy
method were used for prediction of extreme significant wave heights related to high return periods,
i.e. 100 years.
It is argued that the maximum entropy method is
suitable to predict return values of low probability
levels because it produces the least biased pdf.
Furthermore it is argued that the ME method is
more adequate than EVT methods in those predictions,
since it is more robust to the variation of data set size,
i.e. to the variations of threshold values, as shown in
Figure 13.
These statements should be tested on several other
data sets from different regions in order to gain more
credibility but nevertheless results presented in this
paper indicate that such conclusion is plausible.
ACKNOWLEDGEMENTS
The work reported here has been performed within
the EXTREME SEAS project www.mar.ist.utl.pt/
extremeseas, Design for Ship Safety in Extreme
Seas, which has been partially financed by the
European Union thought its 7th Framework program
under contract SCP8-GA-2009-24175.
REFERENCES
Bandyopadhyay, K., Bhattacharya, A., Biswas, P. &
Drabold D. 2005. Maximum entropy and the problem of
moments. A stable algorithm. Physical Review E. 71 (5).
Bermudez, P. & Kotz, S. 2010. Parameter estimation of
the generalized Pareto distribution Part I, Journal of
Statistical Planning and Inference. 6 140: 13531373.
Bermudez, P. & Kotz, S. 2010. Parameter estimation of
the generalized Pareto distribution Part II, Journal of
Statistical Planning and Inference. 6 140: 13741388.
Coles, S.G. 2001. An Introduction to Statistical Modeling of

Extreme Values. New York: Springer.
Coles, S. & Stephenson, A. 2006. ismev package: An Introduction to Statistical Modelling of Extreme Values.
Davison, A. C. 1984. Modelling Excesses Over High
Thresholds, with an Application. Statistical Extremes and
Applications, 461482.
Davison, A. C. & Smith, R. L. 1990. Models for Exceedances
Over High Thresholds (with discussion). Journal of the
Royal Statistical Society, B 62: 191208.
Ferreira, J. A. & Guedes Soares, C. 1998. An Application of
the Peaks Over Threshold Method to Predict Extremes of
Significant Wave Height. Journal of Offshore Mechanics
and Arctic Engineering. 120(3):165176.
Ferreira, J.A. & Guedes Soares, C. 1999. Modelling the
LongTerm Distribution of Significant Wave Height with
the Beta and Gamma Models. Ocean Engineering. 26:
713725.
Guedes Soares, C. & Scotto, M. 2001. Modelling Uncertainty
of LongTerm Predictions of Significant Wave Height.
Ocean Engineering. 28: 329342.
Guedes Soares, C. & Scotto, M.G. 2004. Application of the
r-order statistics for long-term predictions of significant
wave heights. Coastal Engineering. 51: 387394.
Gotovac, H. & Gotovac, B. 2009. Maximum entropy algorithm with inexact upper entropy bound based on Fup
basis functions with compact support. Journal of Computational Physics 228: 90799091.
Jaynes, E.T. 1957. Information theory and statistical
mechanics, Physical Review 106: 620630.
Leadbetter, M.R., Lindgreen, G. & Rootzen, H. 1983.
Extremes and related properties of random sequences and
series. Springer Verlag, New York.
Ochi, M.K. 1992. New Approach in Estimating the Severest Sea State from Statistical Data. Proceedings Coastal
Engineering Conference. ASCE. 512523.
R Development Core Team, 2009. R: A language and
Environment for Statistical Computing. 3-900051-07-0.
http://www.R-project.org, Vienna, Austria.
Smith, R. L. 1986. Extreme Value Theory Based on the r
Largest Annual Events. Journal of Hydrology 86: 2743.
Van Vledder, G., Goda, Y., Hawkes, P., Mansard, E., Martin,
M.J., Mathiesen, M., Peltier, E. & Thompson, E. 1993.
Case Studies of Extreme Wave Analysis: A Comparative
Analysis. Proc. of the Second International Symposium on
Ocean Wave Measurement and Analysis. 978992.
Winterstein, S. R., Kleiven, G. & Hagen, . 2001, Comparing
Extreme Wave Estimates from Hourly and Annual Data.
Proceedings ISOPE Conference. 700708.
504

Boussinesq-type numerical model for wave propagation near shore

and wave penetration in harbors
L. Pinheiro, C.J. Fortes & J.A. Santos
LNEC, Lisbon, Portugal
L. Fernandes
IST, Lisbon, Portugal
M. Walkley
School of computing, Leeds University, Leeds, UK
ABSTRACT: Sea waves can pose significant problems to port infrastructures and activities. Hence it is important to correctly predict and characterize the wave field inside ports, more precisely on navigation routes and
berthing locations. This paper describes the numerical model BOUSS3W, a finite element model for nonlinear
wave propagation near shore and wave penetration in harbors. This model is also able to perform resonance
studies in harbors. It is based upon an extended version of Boussinesq equations. Terms were added to generate
regular or irregular waves inside the numerical domain, absorb outgoing waves, partially reflect waves at physical
boundaries, control numerical instabilities and reproduce energy dissipation due to bottom friction and wave
breaking. Implementation of partial reflection, bottom friction and wave breaking are described and applications
to real and experimental test cases are presented, namely, Vila do Porto Harbor. Results are compared with
physical model tests and other numerical models.
INTRODUCTION
The most important physical effects associated with

the nonlinear wave transformation of waves in near
shore regions can be described by Boussinesq-type
equations. One example of this class of equations
was introduced by Nwogu (1993). These equations
describe the nonlinear evolution of waves over a
sloping impermeable bottom. Their range of validity extends from shallow up to intermediate water
depths where the nonlinear and dispersive effects are
mild. Therefore, they seem adequate to describe the
wave field outside and inside ports, harbors and sheltered zones. In the last few decades several authors
have been working to extend the applicability domain
of these equations to deep as well as very shallow
waters and also to include other physical phenomena such as currents, wave breaking, bottom friction,
etc. . .
The numerical resolution of Boussinesq-type
equations have mostly used finite difference
methods (Kirby et al. 1998). But, although computationally more complex, the finite element method
deals directly with unstructured grids that correctly
represent the physical boundaries of the domain,
including the coastline, islands and other obstacles.
Moreover the finite element method allows minimizing the number of points in the grid using local
refinement techniques. Several authors have used this
method with success (Antunes do Carmo and Seabra
Santos 1996; Li et al. 1999; Walkley and Berzins
2002).
Developments on Walkeys model (Walkley and
Berzins 2002) led to BOUSS3W model, Pinheiro
et al. 2010, which includes internal wave generation (using the source function method with
which regular and irregular waves can be generated), artificial numerical viscosity (to control numerical instabilities), numerical sponge layers (placed
on radiation boundaries to absorb outgoing waves),
numerical porosity layers (placed whether on physical boundaries or inside the domain to simulate
the reflection, transmission and energy dissipation
effects of porous structures on the waves) and
energy dissipation due to bottom friction and wave
breaking.
In the following section, the governing equations
are summarized. The wave generation, boundary conditions and energy dissipation methods are discussed.
Three numerical applications are presented and results
are compared to physical model measurements and
other numerical models.
505
2
2.1
BOUSS3W NUMERICAL MODEL

Generic model description
The extended Boussinesq equations derived by

Nwogu (1993) are given by the following equations,
at depth Z = h.
equations and by using Greens theorem, an explicit

relation between the desired surface wave amplitude
and the source function amplitude is obtained. A
Gaussian function is used to distribute the generated
wave over several mesh points.
where D is the amplitude of the source function, is

a parameter corresponding to the width of the source
region and xs is the centre of the source region.
where 1 = + 1/3, and I1 is given by:

where is the free surface elevation, u = u(x, y, t) =
(u, v) is the velocity vector, h is the water depth.
The original Nwogus equations were further
extended to take into account some important physical
processes (wave transmission through porous structures, bottom friction and wave breaking) as well as
other source/damping terms for numerical reasons.
The BOUSS3W model equations result as follows:
2.3 Boundary conditions

The boundary conditions can be of three types: full
absorption, full reflection or partial reflection.
Full reflection represents a solid impermeable vertical wall. Non permeability and mass conservation
conditions lead to the following boundary conditions:
where n is perpendicular to the boundary.

Full absorption is obtained with use of viscous
damping layers (sponge layers) This viscous term
grows quadratically in the part of the domain corresponding to the sponge layer.
Where the added terms stand for:
Sf source function for wave generation;

(t + s ) 2 viscous damping (numerical stability control and sponge layers);
n fl u + n ft u|u| laminar and turbulent friction
(porous structures);
1
h+
f u|u| wave induced bottom friction;
w
1
h+
e (h + )u wave breaking.
These additional terms are detailed in the following

sections.
2.2 Wave generation
The wave generation is made by an internal generation condition, which is added to the model, using
a source function following the procedure described
by Wei et al. (1999). In this method, the source function is derived by a linearized form of the Boussinesq
where, v1 is the viscosity used, throughout the domain,

to control the numerical instabilities of the model, and
v2 is the maximum viscosity at the sponge layer.
To partially absorb wave energy at a given boundary the method presented by Nwogu and Dermirbilek
(2001) is used. It simulates partial wave reflection and
also transmission through porous structures such as
breakwaters. The modified equations for the porous
region are obtained by replacing u with u/n, where n
is the porosity, and including a term to account for
energy dissipation inside the structure.
Laminar and turbulent friction factors are obtained
using the empirical relationships recommended by
Engelund (1953):
where is the kinematic viscosity of water, d is the

characteristic stone size, and 0 and 0 are empirical
506
constants that range from 780 to 1500, and 1.8 to 3.6

respectively. The characteristic stone size is given by:
where Ws is the stone weight in kN and s is the mass

density of armor material (2.65 kg/m3 for quarry stone
and 2.3 kg/m3 for concrete blocks).
2.4
Energy dissipation
2.4.1 Bottom friction

The bottom boundary layer of flow associated with
the passage of waves is normally restricted to a
small region above the sea floor. There is therefore
a very small amount of energy dissipation due to
bottom friction in typical wave propagation distances
of the order of 1 km used in Boussinesq-type models.
The energy dissipation due to bottom friction however
plays an important role in the wave transformations
near shore, in very shallow waters.
The effect of energy dissipation due to a turbulent bottom boundary layer is simulated using a term
of bottom shear stress, Fb , to the momentum equation, following the procedure adopted by Nwogu and
Demirbilek (2001).
where fw is the wave friction factor. This equation is

expressed in terms of U instead of the bottom velocity in order to minimize the computational effort to
determine it.
The wave friction factor estimates the bottom shear
stress induced by the passage of the wave. To estimate
the wave friction factor a method presented by Leroux
(2003) is used. This author proposes a rigorous form of
expressing fw using solely two variables, the equivalent
diameter of the particles, D, and the wave period, T.
with wcr = 0.027Wds0.6757 , where Wds is the nondimensional sedimentation velocity and can be
evaluated according to the empirical formulation of
Dietrich (1982):
2.4.2 Wave breaking

Wave breaking is a very complex turbulent phenomenon that constitutes an important form of energy
dissipation and cannot be neglected in near shore areas.
Several empirical formulations have been adopted
by different authors to model wave breaking.
To model the turbulent mixing and dissipation
caused by breaking, an eddy viscosity approach is
used, Kennedy et al. (2000). It consists in adding an
ad-hoc dissipative and momentum conservative term
Rb to the momentum equation. This term contains the
eddy viscosity, which is defined on agreement with
experimental data.
and v is the eddy viscosity and h + represents the

total water depth. The eddy viscosity is defined as:
where:
The use of Ha instead of h + provides more

stability to the eddy viscosity term.
The parameter b is the mixing length coefficient.
The purpose of the parameter B is to avoid an impulsive start of the wave breaking and consequently the
instability of the solution.
where and are the densities of the submerged

particles and of water, respectively. The Shields
parameter, , is given by:
The critical orbital velocity, Uwcr is given by:
where /t determines the onset and the cessation

of wave breaking. The use of /t as an initiation
parameter ensures that the dissipation is concentrated
in the front face of the wave as in nature.
A breaking event begins when /t exceeds some
initial threshold value and it will continue even if
/t drops bellow that value. The magnitude of the
threshold value will decrease in time from the initial
507
value (I ) /t to a final one (F) /t. A simple linear

relation is used to model the evolution of /t:
where T is the transition time, t0 is the instant where

breaking was initiated and so t t0 is the age of the
wave breaking event. The expressions for (I ) /t,
(F) /t and T are:
Figure 1. Port of Vila do Porto.
The terms ini, fin and tcst are parameters that define
the initiation, termination and duration of the breaking
events.
3 APPLICATIONS
3.1
Figure 2. Measured points and groups of points (zones)

inside the harbor.
Port of Vila do Porto evaluation of partial

reflection
The port of Vila do Porto is located in the southwest

coast of Santa Maria Island. This is the southernmost
and easternmost island in the Azores archipelago. It
has 18 km long, about 10 km wide and an approximate
area of 97 km2 . It is characterized by a predominantly
abrupt coast cut by deep bays and inlets.
The harbour is protected from south by a rubble
mound breakwater. The inner part of the breakwater
bears a quay for cargo boats and on the opposite side
there is another quay for ferry-boats.
The construction of a new marina for recreation
boats was studied in scale model tests at LNEC and
several different layouts were considered, Fortes et al.
(2005).
Physical model tests were performed for six
different harbour configurations. Wave measurements
were taken at 15 points inside the harbour,
Figure 2. Another point in front of the wave generator was also considered.
For each incident wave condition, the values of the
wave height indexes (ratio of the wave height at each
point of the domain to the wave height at generation
were calculated. Those tests permitted the characterization of the wave conditions inside Vila do Porto
harbour, and especially inside the marina, for different
wave incident conditions.
The numerical calculations presented here were performed for the configuration without the marina. The
Figure 3. Bathymetry of the harbor. Domain boundaries

with different characteristics.
finite element mesh had 44 167 nodes and 87 196

elements. The average number of points per wave
length was 23, considering a period of 11s.
Regular waves (: South; T: 11 s H: 0.4 m) were
simulated for three different test cases in order to
assess the partial reflection boundary conditions
influence in overall results:
Case A full absorption;

Case B full reflection;
Case C partial reflection.
Boundary A is always full absorption and nonphysical boundaries B and M are set to the radiation
condition. See table 2 for coefficients. The calculation
of the coefficients was made using empirical formulae
proposed by Seeling and Ahrens (1995). In Table 1 the
508
Table 1.
Structure parameters.
Table 3.
Statistics of results bias.

bias
Parameter
Description
ds
N
Ho
m
cot()
Dn50
Tp
P
Water depth at structures toe

Number of armor layers
Wave height
Bottom slope
Slope of the structure
Armor unit diameter
Peak wave period
Structural porosity
Zones
Case A
Case B
Zone A
Zone B
Zone C
Zone D
Zone E
Total
0.013
0.098
0.043
0.128
0.033
0.001
0.017
0.113
0.076
0.033
0.025
0.090
Table 4.
Case C
0.059
0.047
0.009
0.030
0.051
0.016
Statistics of results rmse.

rmse
Table 2.
Reflection coefficients.
Zones
Case A
Case B
Case C
Zone A
Zone B
Zone C
Zone D
Zone E
Total
0.180
0.018
0.209
0.116
0.133
0.102
0.140
0.059
0.147
0.089
0.034
0.025
0.103
0.079
0.102
0.039
0.030
0.054
Reflection coefficients
Section
Type
Case A
Case B
Case C
A
B
C
D
E
F
G
H
I
J
K
L
M
sea
sea
rocks
sandy beach
vertical quay
vertical quay
ruble-mound
vertical quay
ruble-mound
vertical quay
antifer cubes
tetrapodes
sea
0
1
0
0
0
0
0
0
0
0
0
1
1
0
1
1
1
0
0
1
1
1
1
1
1
1
0
1
0.39
0.34
0.9
0.9
0.68
0.9
0.68
0.9
0.6
0.6
1
Table 5.
Statistics of results ia.

ia
Zones
Case A
Case B
Case C
Total
63.60%
81.79%
90.21%
the indexes become 0.2 to 0.5. However, due to reflections between the two vertical quays, there are places
inside the port with the higher indexes.
To compare numerical results with measurements
three statistical parameters were calculated, bias, rmse
and ia index of agreement given by:
structure parameters used to define the breakwaters

are presented.
A time step of 0.01 s was used and parameter
was set equal to 5.49 106 m2 /s in order to maintain
numerical stability. The total time of the simulation
was 200 s.
The wave height is reduced as it propagates into
the harbour. This reduction is due to wave diffraction around the breakwater. However, there are areas
within the port, with higher wave height indexes due to
reflections. This is even more obvious from Case A to
Case B, where the border in Case A is totally absorbing
while for Case B is totally reflective.
The biggest wave height indexes occur in front of
the rocky area (border C) and they can reach 3.0. The
area of the port entrance is characterized by wave
height index values of 1.0. As the wave enters the
sheltered part of the port, there is a reduction and
where xi are the experimental values (the reference),

yi are the values of the numerical model, x is the
average of the experimental values and n is the
number of points. The bias measures the relative
mean error and can be positive or negative whether
the model overestimates or underestimates the results.
The rmse root mean square error gives an absolute
mean error. The ia index of agreement quantifies
the overall agreement between numerical and
experimental results, it varies from zero to one and
the higher it is better is the agreement.
In most points in the study areas A, C and E, experimental and numerical results are similar. However,
509
in some points, there are significant differences.

The major differences occur in zones B and D.
On average, the numerical results are slightly higher
than the experimental ones on A, D and E zones where
at B and C zones the opposite occur.
The mean square error is around 14% for A, 18%
for B and 10% for C, and it is clear that the case C
(with the boundary conditions of partial reflection) is
the closest to the results experiments. This is also true
if we look to B, C and D zones.
However, for the areas A (in front of the ferry
pier) and E (future marine), the same does not apply.
In zone A, test B (total reflection) is the one that
conduces to values more similar to the experimental
ones while for Zone E, is the test A (total absorption).
This indicates that the coefficients of Case C are
underestimated in the area inside the port, while they
are overestimate in the area over the beach.
The bias parameter indicates the same, but in an
absolute value. As expected, the case A always underestimates the values while the case B overestimates,
except in zone B. This is due to the fact, that it was
applied a total absorption area in borders E, F and D
to absorb some energy of the waves.
The overall mean agreement index shows that the
new boundary condition of partial reflection improves
the numerical model results. From between 60%
and 80% the agreement is now around 90%. A more
careful adjustment of the coefficients applied to each
section can further improve this value.
It should be borne in mind that there are always
differences between the numerical and physical model
that are related to several factors, including:
The model has an artificial diffusion coefficient to

control the numerical instabilities, which absorbs
some of the wave energy, reducing the wave height
a bit;
The bathymetry of the numerical and physical
models are not exactly the same due to limitations of space of the physical model or constructive
restraints;
In the physical model, the measurement devices
(resistive gauges) also have limitations especially in areas where the wave heights are small
because the sensitivity of the measuring gauges is
reduced.
3.2
Faro beach evaluation of bottom friction
Faro beach is located in the Anco peninsula which

delimitates the Ria Formosa lagoon to the west. Faro
beach is a sandy beach which extends for several
kilometers on the South coast of mainland Portugal.
The area is characterized by a more or less regular
bathymetry parallel to the shore. Two topographic surveys, obtained in the scope of the BRISA project, were
used to characterize the bathymetry of the study area.
For this test case results are compared with another
Boussinesq-type model, COULWAVE, Lynett (2002).
This model is based upon the extended Boussinesq
Figure 4. Faro Beach.
Figure 5. Location of generation line, sponge layers and

points P1 to P6.
equations derived by Wei et al., 1995. It also includes

bottom friction and has been previously tested successfully, making it a good comparison tool.
Regular waves of 0.3 m of amplitude, 8 s period and
wave direction of S37W (217 ) were generated. The
tide level was of +2.0 m (Z.H.).
The numerical domain was discretized with a
finite element mesh containing 110 828 nodes and
220 470 elements. In average, the mesh contains 22
points per wave length considering a period of 8 s.
In Figure 5 the numerical domain is depicted including the generation line and the sponge layers two wave
lengths (132 m) wide each.
The time step was of 0.1 s. A viscous damping
factor of 7.0 106 m2 /s was used. The simulation
time was of 200 s. The wave friction factor was of
fw = 0.0023.
Six points were defined for results analysis, P1
to P6.
In Figure 6 the results of the free surface elevation at 200 s and the wave height indexes (H/H0) are
presented. This shows the wave transformations and
interactions with the bottom along the beach slope as
the wave propagates. Figure 7 presents the free surface elevation at the six points with the two models
BOUSS3W and COULWAVE.
In general, both models reproduce well the wave
transformations. Both reproduce the shoaling of the
wave due to decrease of water depth. Both reproduce
the nonlinear wave interactions and the harmonics
510
Figure 6. Free surface elevation and wave height indexes

at time instant t = 200 s.
Figure 9. Bathymetry and wave heights of the BOUSS3W
and FUNWAVE models and experimental data.
Table 6.
Statistics of results.
BOUSS3W
FUNWAVE
Figure 7. Free surface elevation at points P2 and P4.

BOUSS3W (blue), COULWAVE (orange).
Figure 8. Wave breaking test of a bared beach: Shape of the

bathymetry.
generation. However, there are some differences in the

points closer to the shore, where the second harmonic
appears stronger in BOUSS3W than in COULWAVE.
Overall it is considered that BOUSS3W behaved quite
well considering that it is weakly nonlinear while
COULWAVE is fully nonlinear, and so it is expected
that differences occur in very shallow waters for these
two models.
3.3
Bared beach evaluation of wave breaking
A bared beach case has been done to check the behavior of the model when several wave breakings occur.
This corresponds to an experimental study performed
by Sancho et al. (2001).
The bathymetry is 119.6 long, Figure 8.
The initial amplitude was 0.056 m and a small viscosity of 2.0 103 was put in order to avoid all
numerical instabilities. The period was 2.5s and the
source function of was at x = 26.0 m.
bias
rmse
ia
0.00419
0.000164
0.0344
0.0452
0.953
0.922
The domain was discretized by linear finite elements with two nodes and the spacing between two
nodes was x = 0.05 m. The corresponding domain
has 2393 nodes. The simulation lasted 70 s. Along the
slope, 43 gauges were considered.
Figure 9 presents the comparison between the wave
height experimental data and numerical results along
the bathymetry obtained with BOUSS3W model and
the FUNWAVE model. FUNWAVE (Kirby et al. 1998)
is a phase-resolving, time-stepping Boussinesq model
for ocean surface wave propagation in the near shore
and the comparison of both results on the bared beach
case is interesting. Both of models give results at the
same gauges as the ones of experimental work.
BOUSS3W model reproduced very well the behavior of waves up to the wave breaking location, which
has occurred slightly before the experimental one.
Little differences can be noticed around 90.0 m, but
the last wave breaking is very well re-produced.
Although the BOUSS3W model seemed to be more
accurate then FUNWAVE, computation of the bias,
the root mean square error (rmse) and the index of
agreement (ia) has been done in order to strictly
compare the two numerical models.
A negative bias for BOUSS3W means that the
model tends to underestimate the wave height, while
the FUNWAVEs bias signifies that it overestimates
it. Both of the values are very small. The root mean
square error and the index of agreement show very
good results from both models.
CONCLUSIONS
This paper describes the implementation of partial

reflection, bottom friction and wave breaking physical
processes in the BOUSS3W model.
511
Previous applications of the model confirmed that

the model was able to simulate quite well the main
characteristics of the wave field outside and inside harbor configurations. However, important physical processes were not simulated, namely partial reflection,
bottom friction and wave breaking. These phenomena
constitute an important form of energy dissipation that
cannot be neglected in near shore areas.
All the enhancements made for the BOUSS3W
model improve its capacity to reproduce in a more
realistic way phenomenon involved near coastal zone.
It permits now to manage with full absorption and
full or partial reflection boundary conditions, bottom
friction and wave breaking events.
The results and comparisons with physical model
tests data and other well established numerical models
showed that:
The three physical processes introduced were

adequately implemented in the model;
The model was able to simulate correctly the wave
propagation and most of the wave transformations
present;
There is a very good agreement with measured data.
In sum it can be concluded that BOUSS3W model
is a powerful tool to characterize wave fields in near
shore areas and more importantly with complex harbor
geometries.
Future work involves the coupling of BOUSS3W
with a model for the behavior of moored ships.
ACKNOWLEDGEMENTS
Financial support from the Fundao para a
Cincia e Tecnologia (FCT) of the Ministry of
Science and High Education of Portugal, through
projects PTDC/ECM/73145/2006 and PTDC/ECM/
67411/2006 are gratefully acknowledged.
REFERENCES
Antunes Do Carmo J.S. & Seabra Santos F.J. On breaking waves and wave-current interaction in shallow
water: a 2DH finite element model International
Journal for Numerical Methods in Fluids, vol. 22,
429444.
Dietrich W. E. (1982). Settling Velocity of Natural Particles.

Water Resources Research, Vol. 18, No. 6, pp. 16151626.
December.
Engelund, F.A., (1953). On the laminar and turbulent flows
of ground water through homogeneous sand. Danish
Academy of Technical Sciences.
Fortes, C.J.E.M., Silva, L.G.P., & Sousa, I.A. (2005). Ensaios
em modelo reduzido do Porto de Vila do Porto Santa
Maria,Aores. Relatrio 361/05-NPE, LNEC, Novembro.
131 pp.
Kennedy A. B., Chen Q., Kirby J. T. & Dalrymple R. A.
(2000). Boussinesq Modeling of Wave Transformation,
Breaking, and Runup. I: 1D. J. Waterway, Port, Coastal,
and Ocean Eng. pp. 3947.
Kirby J.T., Wei G. & Chen Q. FUNWAVE 1.0 Fully nonlinear Boussinesq wave model. Documentation and users
manual. UD, Newark, Rel.CACR 98 06, September 1998.
Le Roux, J.P., (2001). A simple method to predict the
threshold of particle transport under oscillatory waves.
Sedimentary Geology 143 (2001): 5970Reply to
discussion.
Li Y.S., Liu S.-X., Yu Y.-X. & Lai G.-Z. (1999). Numerical modeling of Boussinesq equations by a finite element
method. Coastal Engineering, vol. 37, 97122.
Nwogu, O. (1993) Alternative form of Boussinesq equations for near-shore wave propagation. J. Waterway, Port,
Coastal, and Ocean Engineering. 119, 6, pp. 618638.
Nwogu, O. & Demirbilek, Z. (2001). BOUSS-2D: A Boussinesq Wave Model for Coastal Regions and Harbors.
Report 1 Theoretical Background and User.s Manual,
ERDC/CHL TR-01-25, U.S. Army Corps of Engineers.
Pinheiro L.V., Virginie Rodriguez, Fortes C. J., Teixeira
P.R., Walkley M.A. BOUSS3W Nonlinear Wave Propagation Model. Bottom Friction Implementation. 7th
European Conference on Computational Fluid Dynamics
ECCOMAS CFD 2010 Lisbon, Portugal, 1417 June
2010.
Sancho F.E., Mendes P.A., Carmo J.A., Neves M.G.,
Tomasicchio G.R., Archetti R., Damiani L., Mossa M.,
Rinaldi A., Gironella X. And Sanchez-Arcilla A. 2001.
Wave hydrodynamics over a barred beach. Proceedings of
the fourth International Symposium on Ocean Wave Measurement and Analysis Waves, San Francisco, ASCE,
11701179.
Seeling, W.N. & J.P. Ahrens, (1995). Wave forces on inclined
and vertical structures. ASCE, pp. 2855.
Walkley M. & Berzins M. A. (2002). Finite element method
for the two-dimensional extended Boussinesq equations
International journal for numerical methods in fluids, vol.
39, no. 2, 865885.
Wei, G., Kirby, J. T. & Sinha, A. (1999). Generations of waves
in Boussinesq models using a source function method.
Coastal Engineering, 36(4), 271299.
512

Influence of wave data bases on the long term prediction

of wave induced loads in ships
J. Queirs, M. Bernardino1 & C. Guedes Soares
1 Instituto de Meteorologia, Lisbon, Portugal
ABSTRACT: The North Atlantic routes and the wave climate to which the ships are subjected are studied using
global wave data samples for each route obtained from global wave climate data bases. A comparison of the
results obtained for three wave data bases is made for a grid covering the North Atlantic Ocean. Two transatlantic
routes are considered. The significant wave height is compared for HIPOCAS, ERA-40 and COADS data sets,
by evaluating the average values obtained for several periods and performing error calculations between the
different data sets. Similar patterns were found for HIPOCAS and ERA-40, being the first the one that shows
higher maxima. COADS have smaller seasonal amplitude between maxima. The same procedure is applied to the
average period, only for HIPOCAS and ERA-40 data. Higher maxima were found for the second and different
contour patterns. Data sets (HIPOCAS e ERA-40) of significant wave height and peak period associated with
the two transatlantic routes are obtained. A study on the suitability of the Lonowe model for the description of
the sea state by determining the joint probability distribution function for both routes is performed. A long-term
formulation of wave induced vertical bending moments is applied to the several sea state descriptions obtained
from different data, in order to achieve the value of wave induced vertical bending moment response, considering
an exceeding long term probability of 108 . For this probability level HIPOCAS data shows higher values than
ERA-40.
INTRODUCTION
The knowledge of global and local climate over the

Ocean promotes the development of better and more
economical ship structures. Knowing the variability
of sea state is the key to assure safety standards when
exploring sea resources. Because maritime transport,
offshore platforms and onshore facilities rely on these
standards, it is vital to know climatic characteristics of
routes and maritime operation zones.
Different sea states result in different wave severity
levels, which imply distinct structural load amplitudes
on ships or other structures. For floating structures
design it is important to know the sea characteristics
which they will be subjected to during their life, so
that a correct structural dimensioning can be made to
support the wave induced loads. However, over dimensioning should be avoided to prevent the increase of
building and operating costs.
The commercial sector is interested in sea state
forecast services, requiring ship routing aids and
appropriate communication systems. These needs can
be fulfilled with ocean forecast methods using high
technology wave modelling, where corrections can be
made from real time and historical data. For a reliable
simulation, wind fields obtained from atmospheric

models should be included.
The design of marine structures is performed having as reference the sea state characterizations derived
from the North Atlantic wave data. The high wave
severity indices registered in this area are used as a
model of the extreme operation conditions that these
structures can be subjected.
The main objective of this paper is to contribute to
the understanding of the wave data bases role on the
short and long term characterization of the sea state.
Different data bases are used and the data sets for two
transatlantic routes are obtained. Using these data sets
a long term study is performed obtaining probability distributions of sea state parameters. Long term
descriptions are the accumulation of a large number
of short term ones. During its life time, a ship is subjected to a succession of short term sea states during
which it is subjected to the corresponding wave loads.
These loads are considered here for ship design purpose through long term statistical analysis in a given
route.
Two of the databases are made of hindcast data
produced by numerical wave models that have been
forced by historical wind fields and they represent
513
state of the art estimation of the wave climate of

the North Atlantic. The other database results from
the collection of observations made in oceangoing
ships, which has inbuilt the effect of bad weather
avoidance.
The methodologies adopted to achieve the objectives follow those used by several authors. The long
term sea state characterization by a probability distribution model was performed based on the model
proposed by Haver (1985) for the Norwegian Sea.
Calculations of the structural ship responses to a
long term sea state characterization obtained from
wave data bases, were based on the methodology
proposed and adopted by Guedes Soares & Viana
(1988), Guedes Soares & Trovo (1990 & 1991) and
Guedes Soares (1996), among other studies.
In section 2 emphases is given to theoretical formulations. Short term descriptions of sea state by
frequency spectra are referred. A probabilistic distribution fit model is used to obtain long term descriptions of the sea state. Theory for ship response to
periodic excitation and transition to irregular waves
is also referred. Formulation to obtain long term ship
response is also exposed and rules for design vertical
bending moment are referred. In section 3 references to
data bases used for this study are given. The methodology applied to develop the study is explained and grid
results are presented in section 4. In section 5, results
for the routes and the fit to probabilistic distributions
are shown. Finally in section 6, long term response
results are shown. Concluding remarks on the results
obtained are made in section 7.
2
The form for the spectral density function used here

is the Pierson-Moskowitz form, which is used for fully
developed seas:
with P = 2/TP , HS the significant wave height,

the frequency, P the peak frequency and TP the peak
period. This wave spectrum form is used in this study
for long term calculations. The uncertainty associated
to the spectral form in the wave induced ship response
was studied by Guedes Soares (1990), concluding that
this spectral form can be used for long term calculation
purposes.
The ship response to regular wave excitation was
obtained with the strip theory of Salvesen et al. (1970).
When modelling the sea state for short periods in
which this is considered constant, it can be assumed
that the sea surface elevation is a zero mean stochastic process, Gaussian and stationary. Then, the wave
elevation has a normal or Gaussian distribution:
Additionally, if the spectrum is narrow banded the

maxima probability is approximated by a Rayleigh distribution. The short term probability QS of exceeding
the amplitude a in a sea state of variance R is given by:
SHIP BEHAVIOUR AT SEA
The generic sea surface elevation, considering all

wave directions is given by:
This can be reduced considering a single direction and

a coordinate point coincident with the origin to:
After extensive derivation using the wave profile variance approach, which can be found for example in
Young (1999), or using the Wiener-Khintchine theorem approach, which can be found in Ochi (1990), the
last equation becomes:
meaning that the variance of the wave elevation process equals the spectral density function (S()) integrated along the frequency domain () and the sum of
the squares of the individual wave amplitudes (ai ).
Being the ship response to wave excitation linear, these

probabilistic properties are applicable to them also.
Ship motions or load amplitudes obtained in the frequency domain are transfer functions. In this case the
vertical bending moment ones are considered. The
response to a sea state can be determined in the frequency domain by the product of the square of the
transfer function by the wave spectrum. As shown
before, the variance of this response for a given wave
direction R(i ) is achieved by the integration along the
frequency domain:
where S() is the Pierson-Moskowitz wave spectrum

and H (, i ) is the vertical bending moment transfer
function. It is then possible to find as many variances
as wave directions considered in each sea state, so
that the final variance may be found. Applying this
method to several wave spectra (sea states) a long term
description can be found.
A long term description of wave climate is achieved
using the Lonowe model as referred in Haver (1985).
This model intends to describe the Hs distribution by
two distinct probability distributions Lognormal and
514
Weibull, so the Lonowe probability density function is

composed by two different density functions:
To extend this short term formulation to long term

case one must realize that the short term statistical
results are conditional to the occurrence of the sea
states. Thus, the long term probability of exceeding
a limit value (QL ) should consider the probability of
occurrence of each sea state and its weight according
to its duration ((TP )):
where and are the mean and standard deviation

of nHS and and are the Weibull scale and shape
factors. The high probability region is modelled by a
lognormal distribution and the low probability region
by a Weibull. The transition point is where the distribution stops being modelled by a lognormal and starts
being modelled by a Weibull. To assure continuity,
both lognormal and Weibull density and cumulative
function must be equal at the transition point.
The Lognormal parameters are estimated as if a lognormal model was to be applied, . The Weibull
parameters are obtained varying until a reasonable
fit to the tail region is obtained. This is a reasonable
procedure only as far as is sufficiently large.
This procedure does not assure the achievement of
the best possible fit. An iterative procedure based on
the Chi-square criterion is used to determine the best
Weibull distribution found with the last procedure.
Assuming that a short-term sea state is approximately characterized by HS and Tp , the long term
variation in the wave climate can be described by the
joint probability distribution of HS and Tp . From a
scatter diagram, a sample density function may be
considered as a first estimate for the unknown probability density function fHS Tp (HS , Tp ), but its validity
is restricted to the range where sufficient observations are found. An extrapolation for more rare sea
states is then necessary. The joint probability density
function is:
The function fR (r) represents the response variance

probability density function. Complex forms for this
function exists if considering the speed, load conditions, wave directionality, voluntary speed variations,
and heading angle under heavy seas. The extensive
form of this function as written in Guedes Soares
(1995) is:
where fHS (HS ) is the Lonowe probability density function and fHS Tp (HS |Tp ) is the conditional probability
density function for Tp given HS . The last can be
modelled by the lognormal distribution as:
where is the mean value and the standard deviation

of nTp . This may be a reasonable choice regarding its
adoption in previous studies.
For each class of HS including a sufficient number of observations these parameters can be estimated.
In order to obtain reasonable parameter estimates for
large values of HS , continuous functions must be fitted to the estimated parameters and extrapolated for
the range of classes covering extreme sea states.
The speed and heading angle conditional to HS may be

simplified to constant functions. The cargo condition
can also be simplified due to small variations in the full
load condition. Considering that the wave directionality was included in the determination of ship response
variance spectrum, this function may be reduced to
one single term the joint probability density function of HS and Tp . Considering all these assumptions,
the equation (11) becomes in the discrete form as:
where h and t match the classes width in the scatter diagram. The long term probability value must
be set to find the unknown vertical bending moment
response amplitude inside the short term probability
of exceeding.
When designing floating structures which must support extreme load conditions, usually a high enough
probability level is used. This level represents the
probability of extreme wave conditions to be verified
once in a structure lifetime. For ships, a probability
level of 108 is used.
Ship design and construction quality is assured
by classification societies through minimum imposed
requirements. One of those is the establishment of
the vertical bending moment maximum value along
the ship length. The regulations used here are related
to longitudinal strength, with requirements to bending moment and shear forces, which have been made
uniform among all classification societies. Vertical
bending moment values are given as design values
obtained for an exceeding probability level of 108 .
3 WAVE DATA BASES
Wave data bases contain climatologic data for a
spatial grid covering a specific area. A sea state
515
characterization is made through the analysis of this

data, which can be obtained from simulation models,
measurements or real observations. The latest are the
oldest data bases, but also the unreliable ones.
Numerical models solve the spectral energy equation using it to describe the wind wave spectrum
evolution. Phase averaging models predict the wave
field average or integral properties. They predict the
directional wave spectrum temporal and spatial evolution, solving a transport equation which considers
that a source term is equal to the sum of temporal variation of directional spectrum with the product of its
derivatives and the wave group velocity. The source
term is the sum of the wind input, nonlinear interactions between spectral components and white capping
dissipations.
The WAM (wave prediction model) is a third
generation model that is forced by the wind input.
The HIPOCAS project (Hindcast of Dynamic Processes of the Ocean and Coastal Areas of Europe)
provided a simulation of 44 years of wind, waves, sea
level data and climatology (Guedes Soares et al. 2002,
Guedes Soares 2008). The wind field used as input to
the wave model was obtained by a limited area atmospheric model forced with global data from NCEP and
NCAR global reanalysis. For the HIPOCAS project,
the WAM model cycle 4.0 was used to provide the
hindcast (Pilar et al. 2008).
The ECMWF (European Centre for Medium-Range
Weather Forecast) atmospheric data comes from the
ERA-40 re-analysis (Uppala et al. 2005). To model the
atmospheric effects on the wave generation, the WAM
model was also used, obtaining a full description of
the sea state through the energy spectrum.
The COADS (Comprehensive Ocean-Atmosphere
Data Set) is a data base resulting from the Voluntary
Observing Ships (VOS) scheme, which is an international scheme by which sailing ships are recruited
by National Meteorological Services for taking and
transmitting meteorological observations.
The COADS data base has records of visually
observed height, period and direction for wind wave
and swell. However, only 4060% of the total number of reports includes wave information (Gulev and
Hasse, 1998). This type of date has some uncertainty (e.g. Guedes Soares, 1986a, b) and Gulev
et al. (1998) among others, have concluded that wave
height is always overestimated in VOS observations
in comparison with remote sensed data and numerical
models in the areas with small significant wave height.
VOS data is irreplaceable and is the only source of data
along the ship routes and the ones that have the bad
weather avoidance inbuilt in them.
4
The domain covers mostly the North Atlantic, and is

limited to [64:18] Longitude and [14:70] Latitude.
The grid (Figure 1) is limited by the same coordinates
in a resolution of 2-by-2 degrees.
Each data base presents the data in different formats, and organized by distinct files. In order to obtain
data for the defined grid, data treatment was necessary.
Data from HIPOCAS is grouped in 2-by-2
degrees annual matrices covering [66:18] Longitude, [14:70] Latitude. Data is available for every
three hours, but a 6 hour sampling is used in order
to compare it with the ERA-40 data.
Data from ERA-40 covers [65:25] Longitude,
[75:10] Latitude in a 2.5-by-2.5 degrees resolution
matrix. Data is available every six hours and stored in
a single file. An interpolation to 2-by-2 resolution was
made.
The COADS data is stored monthly for 40 years with
average values. The initial resolution is 2-by-2 degrees
for a grid covering [0:360] Longitude, [80:80]
Latitude.
Except for the COADS data, after the preprocessing, the final files consist in tridimensional
monthly matrices containing all the available data for
that period. Seasonal or averaged matrices can be
obtained by simple operations of these monthly matrices. For COADS data, due to its nature, the final file
is a bi-dimensional monthly matrix.
The results shown here are syntheses of the general results obtained. Results for the severest month
(January) are presented for the North Atlantic region.
The HIPOCAS data shows the highest values of all
the considered data. Annually it is possible to define
the area of the maxima where the averaged values of
HS oscillate between 3.4 and 3.6 meters. This area is
limited by [40;15] Longitude, [45;60] Latitude,
and remains as the zone of maxima for most of the
considered periods. The contours for this parameter
develop concentrically from this location outwards,
uniformly. The critical month is January (Figure 2)
with significant wave heights exceeding 5.2 meters.
NORTH ATLANTIC DATA COMPARISON
The data used on this study comes from the three

data bases referred previously. For HIPOCAS and
ERA-40 data, 10 years (19581967) are considered,
and for COADS, due to the nature of the data, 40 years
(19581997) are considered.
Figure 1. Grid points, GB-USA route and Norway-USA

route.
516
Figure 2. HS January contour (HIPOCAS).
Figure 3. Tm January contour (HIPOCAS).
As winter ends and spring begins, two distinct

maxima zones can be identified, developing at Northeast and Southwest of the first zone referred. The
considered routes intersect the maxima zone in all
periods.
The Tm in HIPOCAS data oscillates between 5.0 and
7.5 seconds for the annual average. Similarly to HS ,
the severest periods are found during winter. Except
for the summer months, the maxima zone is defined by
[40;15] Longitude, [45;60] Latitude. In January
(Figure 3), maximum values exceed 9.0 seconds in this
area.
Similarly to the HIPOCAS data, also in the HS
ERA-40 data is it possible to define a grid zone where
high values are dominant for most of the considered
periods. Still, annually, the oscillation of the maxima
for the same zone is higher for ERA-40. The maximum annual average value is lower than the one for
HIPOCAS data, registering a value of 3.2 meters. The
severest month is January (Figure 4), where values
higher than 4.0 meters are found for an expanded area
covering the central North Atlantic.
The average annual Tm values registered for ERA40 oscillate between 7.0 and 9.2 seconds (higher than
HIPOCAS and with lower amplitude). Winter months
are the severest ones. In January (Figure 5) values
higher than 10.5 seconds are registered. The pattern is substantially different from that observed in
HIPOCAS, with maxima over the shore and contours
developing outwards in direction to North America.
For the annual average of HS , the contours obtained
for COADS data differ from those of HIPOCAS and
ERA-40. More maxima areas are identifiable, and
shifts on the limits of the great zone are visible. The
contours evolution is not similar to the previous data
sets showing an area of large values close to the coasts
of Canada and USA, which are in fact the areas that
have been identified by Guedes Soares et al. (2001) as
Figure 4. HS January contour (ERA-40).
the ones with more frequent ship accidents due to bad

weather.
Annual maxima can exceed 3.5 meters. January is
similarly the severest month, with average HS values exceeding 4.5 meters (Figure 6), which are values
between the ones of HIPOCAS and ERA-40. Taking
in consideration that the COADS data, have inbuilt
the bad weather avoidance effect, the real wave climate should have higher values as does the HIPOCAS
database show.
Caires and Sterl, (2005), have validated the ERA-40
wave data with buoy and altimeter data and concluded
that it was predicting lower extremes, which is in
agreement with the present results in which ERA-40
data has the lowest maxima and HIPOCAS has higher
values than COADS, as it should be expected.
517
Figure 5. Tm January contour (ERA-40).
Figure 7. HS Bias in January (HIPOCAS/ERA-40).
Figure 6. HS January contour (COADS).
Figure 8. Tm Bias in January (HIPOCAS/ERA-40).
The summer months in COADS data are the severest

of all data bases, which agrees with the observation of
Gulev et al (1998), who concluded that the COADS
observations overestimate the low sea states.
Calculations of the error between the data sets
were made and results of the bias between HIPOCAS
and ERA-40 are shown in Figures 7 and 8.
Observing Figure 7, it can be seen that the Hs bias
is typically positive with values close to 1.0 meters.
The analysis made to all periods considering also
the mean absolute error and the root mean squared
error calculations shows the existence of a constant
difference between data in all periods, and small
variation of the contour pattern.
The Tm results are slightly different due to the variation of contours between data bases. Although high
differences in the near shore can be found, in general
smaller and constant differences can be seen for all

periods. This can also be seen for January (Figure 8).
Despite these differences, analysing other statistics,
consistence is found when constant differences along
the periods are present.
The bias shows significant differences around the
Azores islands, which results from the significant differences in the patterns between figures 23 and 45.
The differences result from the fact that the HIPOCAS
grid had nested a fine grid of 0.25 0.25 around
the Azores islands in order to obtain realistic results
around the islands, while ERA-40 had only a coarse
grid of 2 2 . The fine grid allowed the resolution
of the islands and modelled the blocking effect that
they have on wave propagation, which was studied in
detail by Ponce de Len and Guedes Soares (2005).
518
The figures 34 and 78 show that the shadow effect

of the islands is felt at a significant distance.
5
NORTH ATLANTIC ROUTES LONG TERM

MODELLING
Two transatlantic routes connecting Great Britain and

Norway to USA (Figure 9) are considered. The HS
results obtained over these routes are shown here.
The presented results refer to January and July for
both routes (Figure 10 & Figure 11), but global results
were also obtained and are commented.
Figure 9 shows how the Norway-USA route points
were numbered for analysis. The numbering for
GB-USA route (not shown) is similar. The following result analysis is based on the global results
obtained.
Annually, the route connecting Norway to USA
shows higher HS maxima in the HIPOCAS data,
although near the American shore higher values are
found in COADS. Analysing all the data bases, a
critical area limited by [32;22] Longitude. January is the severest month, with maxima exceeding
5.3 meters (HIPOCAS), 4.3 meters (ERA-40), and
4.6 meters (COADS). In the summer, the highest
maxima are found in COADS, being the data set
with smallest seasonal amplitude. These results are
in agreement with the overall conclusions obtained in
section 4.
Figure 9. Norway-USA route points numbering.
Figure 10. HS evolution in January and August (Norway-USA).
Figure 11. HS evolution in January and August (GB-USA).
519
Table 1.
Norway-USA route scatter diagram (HIPOCAS data)
For the GB USA route, similar results are found.

For all the considered periods, HIPOCAS shows the
highest maxima, except for the summer months and for
a small area near American shore where maxima are
found for COADS.Again January is the severest month
with maximums exceeding 5.3 meters (HIPOCAS),
4.3 meters (ERA-40) and 4.7 meters (COADS) (higher
than the other route).
Long term results on induced loads can only be
obtained through the use of ship transfer functions.
A tanker with 264 meters length and beam of 45.1
meters with a displacement of 173,165 tonnes was
chosen for these calculations. A strip theory program
based on the method of Salvesen et al. (1970) was
used to determine the vertical bending moment transfer functions in zero speed conditions for a range of
[0:180] of wave direction (gap of 30 ).
It is expected that the maximum bending moment
occur at or near the mid ship section. The used
transfer functions correspond to a section located 25.3

meters before the mid-section, which correspond to
the section with higher values of response.
Calculations on the requirements imposed by DNV
were also performed for sagging and hogging in stillwater and for wave conditions. The maximum bending
moment in the considered section and for a probability of exceedance of 108 , should not be higher than
8 864 272 [kNm].
The methodology and results are presented only
for the Norway USA route using HIPOCAS data
as example. Final results are commented.
The scatter diagram for HS and TP was obtained
for this route by collecting the data from all grid
points of the HIPOCAS data, as in Guedes Soares
and Moan (1971) and the result is shown in Table 1.
In this table all the observations obtained during
10 years in all points constituting the route are
included.
520
Figure 12. Lonowe p.d.f. fitted to HS (Norway-USA).
Figure 13. Best Lonowe fit found (Norway-USA).
Then the Lonowe model is used to fit the HS data.

A fit by a lognormal distribution is made. Once the
parameters are defined, for each value of HS (), a value
for the Lognormal density and cumulative functions is
well defined. The parameters of Weibull distribution
for that point of HS can then be determined by solving
Because density and cumulative functions are dependent, a large amount of computational time is required
to find a solution. Therefore the system was reduced
to a single parameter using
Figure 14. TP
(Norway-USA).
An evaluation of the best fit obtained from the Lonowe

model is made by a Chi-square criterion. The best fit
is obtained when the Lonowe distribution chi-square
error is lower than the error obtained for the Lognormal. The HS histogram and the fitted distribution by
Lonowe model are shown in Figure 12.
However, this method does not ensure that the best
Lonowe is chosen to fit the distribution. An evaluation
of the obtained consecutive Lonowe distributions is
therefore necessary. Again, the Chi-square criterion is
used to quantify the error of the functions. The best
fit is shown in Figure 13, where the fitted lognormal
distribution is included for comparison.
A complete sea state modelling requires the HS and
TP joint probability density function for each sea state.
As shown in the scatter diagram for each range of
empirical
cumulative
distributions
HS , there are series of TP observations and the way

they spread along the ranges of TP should follow a
lognormal distribution (Guedes Soares et al. 1988).
The parameters of these distributions were determined.
In Figure 14 these distributions are presented in normal scale. The approximation towards a straight line
ensures the approximation to a lognormal distribution.
As the range of HS increases, the number of observations becomes scarce, meaning a reduction of quality
in the parameters. A fit to the mean and standard
deviation is made based on the initial ranges of HS .
These fits are shown in Figures 15 and 16.
6
LONG TERM SHIP RESPONSE
For each route and for each data base (HIPOCAS and
ERA-40), the joint probability distribution function
was determined.
521
Figure 15. Fit to TP mean parameter (Norway-USA).
Figure 17. Vertical bending moment ship response comparison for both routes and data bases.
Figure 16. Fit to TP standard deviation parameter

(Norway-USA).
As referred in section 2, the vertical bending

moment amplitude of response for an exceeding probability of 108 is given by a weighting function that
considers all the sea states. The final solution can be
found by
The results show similarities between data bases. The

magnitude of the response for this probability level
is substantially higher considering HIPOCAS data
(Figure 17), exceeding the reference values from Classification Societies. This type of differences from the
Rule values had already been noted before, using
databases from visual observations (Guedes Soares
1996).
This result is in agreement with the ranges of HS
evident on the scatter diagrams. Both data bases agree
in the severity of the routes, giving higher responses
for the route connecting GB to the USA.
7
CONCLUSIONS
Three North Atlantic wave data bases were evaluated

concerning the significant wave height (HIPOCAS,
ERA-40 and COADS) and the average period
(HIPOCAS and ERA-40). Different distributions and
magnitudes for these parameters were found when different modelling methods and different input data are
used.
Data from HIPOCAS shows the higher significant

wave height values for all the considered periods, with
exception for the summer months (higher values for
COADS data). With the higher minima in all periods,
COADS data show the smallest amplitude between
minima and maxima.
ERA-40 has the smallest maxima (even smaller than
COADS), which agrees with earlier result that showed
that the database was underestimating the extremes.
COADS data show lower maxima than HIPOCAS
as one would expect from having inbuilt the effect of
bad weather avoidance.
HIPOCAS and ERA-40 show similar contour patterns although with different values. COADS shows a
similar pattern, although slightly displaced and with
larger maxima areas.
The critical ocean area is that within [40;18]
Longitude, [46;56] Latitude, which the two routes
intersect. The registered maxima in January for the
two routes are similar for the same data base. Only
COADS shows higher maxima for the GB-USA route
then for the Norway-USA.
ERA-40 data shows higher average period values
for all the considered periods than HIPOCAS data. The
distribution pattern is different, although similarities
are found for the summer months. The bias remains
constant for all periods, showing constant differences
between data.
The Lonowe model was applied to the sea state
descriptions obtained from HIPOCAS and ERA-40
data for both routes, showing better fits then those
obtained by Lognormal or Weibull models.
Different long-term vertical bending moment
response amplitudes were found from HIPOCAS and
ERA-40 data for a probability level of 108 .The higher
significant wave height extreme values registered for
HIPOCAS result in a response that exceeds the one
included in Classification Society rules obtained with
the same level of probability, which confirms earlier
results.
522
ACKNOWLEDGEMENTS
REFERENCES
Caires, S. and A. Sterl, (2005), 100-year return value estimates for ocean wind speed and significant wave height
from the ERA-40 data. J. Climate, 18, 10321048.
Guedes Soares C. (1986a) Assessment of the uncertainty in
visual observations of wave height Ocean Engineering,
Vol. 13, pp. 3756.
Guedes Soares, C. (1986b) Calibration of Visual Observations of Wave Period, Ocean Engineering. 13, No. 6,
pp. 539547
Guedes Soares, C. (1990), Effect of spectral shape uncertainty in the short term wave induced ship responses,
Applied Ocean Research, Vol. 34, no. 2, pp. 5469.
Guedes Soares, C. (1995), Effect of wave directionality on
long-term wave-induced load effects in ships, Journal of
Ship Research, Vol. 39, No. 2, June 1995, pp. 150159.
Guedes Soares, C. (1996), On the definition of rule requirements for wave induced vertical bending moments,
Marine Structures, 9, pp. 409425.
Guedes Soares, C. (2008), Hindcast of Dynamic Processes
Engineering; 55, pp. 825826.
Guedes Soares, C. and Moan, T. (1991), Model uncertainty
in the long-term distribution of wave-induced bending
moments for fatigue design of ship structures, Marine
Structures 4 (1991), pp. 295315.
Guedes Soares, C. and Trovo M.F.S. (1990), Sensitivity of
ship motions predictions to wave climate descriptions, Int.
Shipbuild. Progr., 39, no. 418, (1992), pp. 135155.
Guedes Soares, C. and Trovo M.F.S. (1991), Influence
of wave modeling climate on the long-term prediction of
wave induced responses of ship structures, Dynamics of
Marine Vehicles and Structures in Waves, Elsevier Science
Publishers B.V.
Guedes Soares, C. and Viana, P.C. (1988), Sensitivity of

the response of marine structures to wave climatology,
Computer Modeling in Ocean Engineering, Eds. B.A.
Schreffler and O.C. Zienkiewicz, (A.A. Balkema Pub.,
Rotterdam, 1988), pp. 487492.
Guedes Soares, C., Bitner-Gregersen, E., & Antao, P. (2001)
Analysis of the Frequency of Ship Accidents under Severe
North Atlantic Weather Conditions. Proceedings of the
Conference on Design and Operation for Abnormal Conditions II, RINA, London, UK, 221230.
Guedes Soares, C., Lopes, L.C. and Costa, M.D.S. (1988),
Wave climate modelling for engineering purposes,
Computer Modelling in Ocean Engineering, Balkema,
Rotterdam.
Guedes Soares, C., Weisse, R., Carretero, J.C. and Alvarez, E.
(2002), A 40-years hindcast of wind, sea level and waves
in European waters, Proc. 21st International Conf. on Offshore Mechanics and Arctic Engineering,(OMAE 2002),
ASME, New York, USA, Paper OMAE2002-28604.
Gulev S.K., Hasse L. (1998) North Atlantic wind waves
and wind stress fields from Voluntary Observing Ship
data Journal of Physical Oceanography, Vol. 29,
pp. 11071130.
Gulev S.K., Cotton D., Sterl A. (1998) Intercomparison
of the North Atlantic wave climatology from Voluntary
Observing Ships, satellite data and modelling Physics
and Chemistry of the Earth, Vol. 23, pp. 587592.
Haver, S. (1985), Wave climate off Northern Norway,
Applied Ocean Research, Vol. 7, No. 2.
Ochi, M.K. (1990), Applied probability and stochastic processes in engineering and physical sciences, Wiley
Pilar P., Guedes Soares, C., Carretero J.C. (2008), 44-year
wave hindcast for the North East Atlantic European coast,
Coastal Engineering, Vol. 55(11), pp. 86171.
Ponce de Len S, Guedes Soares C. (2005), On the sheltering effect of islands in ocean wave modelling. Journal of
Geophysical Research 110:C09020.
Salvesen N., Tuck E.O., Faltinsen O.M. (1970), Ship motions
and sea loads. Trans. SNAME 1970, Vol. 78, pp. 25087.
Uppala, S.M. et al. (2005), The ERA-40 re-analysis, Quart.
J. Roy. Meteor. Soc., 131, 29613012.
Young, I.R. (1999), Wind generated ocean waves, Elsevier.
523

Study of the wave transformation in the central part

of the Portuguese nearshore with high resolution models
Eugen Rusu, Marta Gonalves & C. Guedes Soares
ABSTRACT: An evaluation of the two state of the art high resolution coastal wave models FUNWAVE and
REFDIF1, is carried out in the present work. The target area is a coastal sector located in the central part of
the Portuguese continental nearshore, south of the city Figueira da Foz. The wave input for the two models
is provided by an ADCP and transferred via a coarse REFDIF driver to three high resolution computational
domains. Simulations with the models FUNWAVE and REFDIF1 were performed in parallel in these three areas
for various wave and tide conditions. The results show in general a good agreement between the two models
and the field data outside the surf zone. Inside the surf zone a calibration process is carried out for FUNWAVE
and afterwards the results provided by the two models become significantly closer. The present study gives
information concerning the performance of the two wave models evaluated in different forcing conditions and
also about some specific features concerning the process of wave breaking in the Portuguese nearshore.
1
INTRODUCTION
In the last decades the analysis of surface gravity

waves by means of Boussinesq models has proven to
be a successful tool in simulating waves propagating in
shallow water. These models have been shown to give
reliable predictions in comparisons with either field or
laboratory data. In fact, Boussinesq equations coincide
with shallow water equations for weakly dispersive
and weakly nonlinear wave propagation (Peregrine,
1967). From this reason, in their initial form these
equations are invalid in intermediate and deep water.
Moreover, standard Boussinesq equations are also
invalid in the surf zone if dissipation due to wave
breaking is not included.
More recent studies have been carried out to overcome the shallow water restrictions and the extended
Boussinesq equations have been developed by adding
terms to varying degrees of accuracy in representing
non linearity and frequency dispersion. Thus Nwogu
(1993) used velocity at a reference depth as the dependent variable in the governing equations developing
in this way a new set of equations that are able to
simulate wave propagation from relatively deep to
shallow water. Based also on these equations, Wei
et al. (1995) implemented a high order numerical
model but they have made no assumption on the small
nonlinear effects, thus deriving Boussinesq equations
which include additional nonlinear dispersive terms.
In this case, the equations can be applied to simulate
wave propagation with strong nonlinear interaction
and to model wave-current interactions.
Kennedy et al. (2000) set out from the fully non linear Boussinesq equations of Wei et al. (1995) and also
included surf zone phenomena such as breaking and

run up in their modeling (Chen et al., 2000). All these
modifications were implemented in the FUNWAVE
model (Kirby et al., 1998, Chen et al., 2000), which
can refer either to the standard equations deduced by
Peregrine (1967) or to the non linear shallow water
equations or extended equations by Nwogu (1993) by
setting some basic parameters.
The main objective of the present work is to evaluate the wave transformation in the Portuguese coastal
environment using FUNWAVE model and to compare
the results provided by this model with those coming
from another high resolution wave model REFDIF1
(based on the mild slope equation) and also with
some field data. The target area of the present study
is a coastal sector located in the central part of
the Portuguese continental nearshore south of the
Portuguese city Figueira da Foz.
In order to provide realistic estimates of the waves
and littoral drift current in the central part of the
Portuguese continental coastal environment, and to
contribute to the calibration of the numerical models for prediction of waves and nearshore circulation,
a field experiment was performed (Silva et al., 2004).
Some results coming from this field validation will be
used as a reference in the present study.
2 THEORETICAL BACKGROUND
FUNWAVE is a wave propagation model based upon
the extended Boussinesq equations derived by Wei
et al. (1995). The model solves the mass conservation
equation and the momentum conservation equation.
525
These equations describe the frictionless evolution

of non-breaking waves over a smooth impermeable
bottom. In order to develop a model for practical application several effects were incorporated into the model
scheme including physical effects of frictional damping and wave breaking as well as extensions needed
to perform some numerical tasks as wave generation,
boundary absorption and moving shoreline (Kennedy
et al., 2000).
In FUNWAVE waves are generated using an internal source mechanism. Sponge layers are placed in
the ends of the domain to effectively damp the energy
of outgoing waves with different frequencies and
directions. The source function f (x, y, t) is split into
two parts. In one hand, there is a Gaussian shape
function and in other hand the input time series of
the magnitude of source function:
slope equation. This equation, which originally was

introduced by Berchoff (1972), is an elliptical partial differential equation deduced for the case of the
monochromatic waves and has the expression:
in which (x, y) is the surface elevation, k the wave

number related to the relative frequency through the
dispersion relationship, c the phase velocity (c = /k)
and cg the group velocity (cg = /k).
A simpler approach in the form of a parabolic equation as developed initially by Mei and Tuck (1980)
is used in the REFDIF1 model. An energy loss term,
due to Booij (1981) and expanded by Dalrymple et al.
(1984), permits the model to treat bottom frictional
losses due to rough, porous or viscous bottoms, surface
films, and wave breaking.
where g(x) is the Gaussian shape function and s(y, t)

the input time series of the magnitude of source
function. These functions are defined as:
where is the shape coefficient for the source function and xs is the central location of the source in the
x direction for a source oriented parallel to the y-axis.
The model also provides for the presence of sources
along the lateral boundaries parallel to the x-axis. D is
the magnitude of the source function, = k sin() the
wave number in the y direction and k is the linear
wave number. Due to nonlinear interactions in the
model are generated higher order harmonics as the
wave propagates in the field. To eliminate the growth
of harmonics of very short wavelength, which can
cause non-convergence of the model (blow-up), a filter
is applied to pre defined time intervals.
Two kinds of boundary conditions are used in the
model, i.e. total reflected wall and sponge layer. A
sponge layer boundary condition is used here since
it is able to damp wave energy for a wide range of
frequencies and directions.
A high order finite difference scheme was implemented in FUNWAVE. Thus, a composite 4th-order
Adams-Bashforth-Moulton scheme (utilizing a 3rdorder Adams-Bashforth predictor step and 4th-order
Adams-Moulton corrector step) is used to step the
model forward in time. All errors involved in solving the underlying nonlinear shallow water equations
are thus reduced to 4th order in grid spacing and time
step size. Spatial and temporal differencing of the
higher-order dispersion terms is done to second-order
accuracy, which again reduces the truncation errors
to a size smaller than those terms themselves. No
further back-substitution of apparent truncation error
terms is performed.
On the other hand, REFDIF1 (Kirby and Dalrymple,
1994) uses an approach based on the so called mild
where A is the complex amplitude related to the

water surface displacement by = Aei(kxt) , the dissipation factor w, is given by a number of different
forms depending on the nature of the energy dissipation. The variable depth features of the mild-slope
equation (along with extensions to include effects of
wave-current interaction) are retained, but the REFDIF
model is developed in parabolic form and in terms
of the complex amplitude A. From numerical point
of view the model is solved in finite difference form
using a Crank-Nicolson Technique.
3 THE FIELD EXPERIMENT
The target area of the present study is a sector located
in the central part of the Portuguese continental coast
(south of the Portuguese city Figueira da Foz). This
is a 9-km long uniform coastal sector with a straight
shoreline parallel to the depth contours approximately
oriented N21 E. It is marked by the absence of coastal
structures, either natural (like rocky outcrops) or artificial (normal to the shoreline, like groins, or parallel,
like breakwaters), and free of discontinuities (like
river mouths). It is a sandy ocean bottom zone without sediment sources or sinks. The site bathymetry is
illustrated in Figure 1 (in the upper side) interpolated
from a 1994 hydrographical chart of scale 1:2 000. A
continuous submarine bar rises above the 10 m depth
contour, reaching minimum soundings of 4 m. The bar
dimensions are 3.56 m height and 350450 m width.
The bar crest is 500650 m from the mean sea level
coastline.
In order to provide a realistic estimate of the waves
and of the nearshore wave induced currents a field
experiment was performed in the target area (Silva
et al., 2004) by releasing and tracking, through georeferenced sequenced aerial photographs, of a substantial number of surface drifters at two previously
526
Table 1.
Main settings of the two wave models.
Model
Parameter
x (m)
y (m)
nx
ny
t (s)
nt
ibc
iftr/wc
break
ck_bt/FCW
Figure 1. Location of the target area, the bathymetry,

the computational domains and the nine reference points
considered.
selected points just off the surf zone. The drifters

were plastic covered colored A3 paper sheets. Markers
reaching the beach were collected and positioned by
a shore-team. The release sites were decided based
on numerical and analytical experiments performed
during the planning phase, and the inspection of
geo-referenced historical photographs of surf zone
patterns. Three pairs of release operations, separated
by 3560 minutes, were performed at each of the two
days of the experiment (6 and 7 October 2003). Sets of
250 markers of one specific color were released from
a rubber boat in each operation, a total of 3000. Six
different colors were chosen for the markers so as to
guarantee that at each day a particular color referred
to one single release point at one single release time.
A flight level of 2500 ft was adopted, corresponding
to a survey scale of 1:5 000. Six over flights of the
experiment area took place on 6 October, separated
by 30 minutes, and nine over flights on 7 October
separated by 11 to 33 minutes. During each flight
vertical color slides were taken routinely at intervals
of 11 seconds, corresponding to 20% overlap, with
additional frames shot whenever convenient.
Sea and swell conditions approaching the nearshore
area, as well as tide and background current over
the inner shelf, were monitored at the outer limit
of the experiment area with an Acoustic Doppler
Current Profiler (ADCP) equipped with a wave module moored at the 13-m depth contour (Figure 1).
The predominant waves in this region are from NW,
with significant wave heights commonly between 1
and 2 m, and mean periods between 5 and 7 s. The wind
conditions were observed in a coastal station 60 km
south of the experiment area.
The field experiment concluded with 120 photographs that were scanned and geo-referenced with
known points in land. The central coordinates of the
photographs were compared with those provided by
FUNWAVE
REFDIF1 +
SHORECIRC
2
2
301
301
0.1
9001
2
100
cbkv = 0.35
cbrk = 1.2
0.001
2
2
301
301
0.045
12001
2; 6; 5; 5
1000
k = 0.5
= 0.3
0.012
the in-flight navigation system, giving maximum differences of 10 m in N-S direction and 2 m in E-W
direction. Envelopes of markers groups were drawn,
and vector positioning of individual markers was
used for determining the group mass centre position
(Gomes et al., 2004). Approximate speed estimates
were obtained from the shortest distance between
consecutive centre positions.
For the observed wave conditions corresponding
to typical values at that site, the significant longshore
current was greater than 20 cm/s up to 1 m/s, for the
markers caught within the surf zone, with 30 cm/s
average and 21 cm/s standard deviation.
Simulations with the numerical models REFDIF1
as wave driver and SHORECIRC (Svendsen et al.,
2002) for assessing the coastal circulation were performed every 15 minutes for a time interval of 2 hours
and 45 minutes (between 9am and 11.45 am) corresponding to the conditions that occurred on 7 October
2003. At each time step, the input conditions and the
tide level were varied. In general, a good agreement
concerning the waves and nearshore currents between
the model results and the field data was obtained
(Rusu and Guedes Soares, 2010). Nevertheless, it has
to be also mentioned that in order to fit the field data
the breaking parameters corresponding to the modeling system had to be modified from their standard
values to the values indicated in Table 1. A more
detailed discussion concerning these modifications
will be made in the next sections.
4 COMPUTATIONAL STRATEGIES
AND SETTINGS
Since the results provided by the combined modeling
system REFDIF1-SHORECIRC were already validated in the target area, the objective of the present
work is to make a comparison between the results provided by the FUNWAVE model with those coming
from REFDIF1 for the same time frames considered
above (in 7 October 2003).
The computational strategy adopted is suggested
in Figure 1. The wave parameters provided by the
527
ADCP are used as input for a wave driver based on

the model REFDIF1 in a coarse computational domain
(with 10 m spatial resolution in both cross shore and
longshore directions). Three high resolution computational domains are considered inside this coarse area
as illustrated in Figure 1. These are square areas all
of them with length of 600 m and 2 m spatial resolutions in both directions. The input for the two models is
provided by the corresponding wave parameters from
the coarse domain simulations.
The same input wave parameters are considered for
both models. These are: wave amplitude (defined as
Hrms /2), peak period (Tp), wave direction (Dir) and
tide level (Tl).
The root-mean-square wave height Hrms is computed as:
in which Hj is the individual wave height. The main

settings of the two models are presented in Table 1.
In the above table x, y, nx and ny represent the
grid sizes and the number of grid points for x and y
directions, respectively. t and nt represent the time
step and the number of time steps, respectively while
ibc the type of boundary conditions activated.
While in FUNWAVE the time step is given as direct
input parameter in SHORECIRC this is controlled by
the Courant number, defined as:
where hm represents the maximum water depth in

the computational domain, x the spatial step in the
cross-shore direction, and g the acceleration of gravity.
The Courant number was 0.1, and time step was 0.045
seconds. To reach a steady state flow, at least 10,000
time steps were required. For both models the cold
start was considered.
Two kinds of boundary conditions are used in
the FUNWAVE model i.e.: (1) total reflected wall
and (2) sponge layer. Waves are generated inside the
domain by a source function technique as described
in section 2 of the present work. For the modeling
system REFDIF1-SHORECIRC five types of boundary conditions can be used, these are: (1) specified
flux, (2) absorbing/generating, (4) wall, (5) periodic,
(6) no-flux.
The time intervals for which a numerical smoothing filter is applied in FUNWAVE to eliminate the
growth of harmonics of very short wavelength, which
can cause non-convergence of the model (blow-up),
are indicated by the parameter iftr. For REFDIF1
the parameter wc represent the number of time
steps between short-wave updates, that is the number
of time steps in SHORECIRC after the waves are
recalculated by running again the wave model in the
updated field of nearshore currents.
As regards the wave breaking, in FUNWAVE two
parameters control this process, they are: cbkv
coefficient allowing the variation of parameter for
the breaking scheme and cbrk coefficient for wave

breaking. For REFDIF1 the parameter k represents the
wave height to water depth ratio (H /h) at the initiation of breaking, when H /h exceeds k the waves start
breaking. Similar when H /h decreases below the
wave breaking stops.
Finally in FUNWAVE ck_bt is the coefficient for
bottom friction formulated by the quadratic law while
in SHORECIRC FCW is the friction factor.
Another parameter (ibe) is specific only for the
FUNWAVE model. This controls diferent types of
Boussinesq equations: (1) ibe = 1 is for Nwogus
(1993) extended Boussinesq equations, (2) ibe = 2
for the fully nonlinear Boussinesq equations of Wei
et al. (1995), (3) ibe = 3 for Peregrines (1967)
Boussinesq equations, (4) ibe = 4 for nonlinear shallow water equations and (5) ibe = 0 is for linearized
Nwogus equations. For the present study option (1)
was activated in all simulations. As regards the others
parameters, the standard options were used in the two
modeling systems.
As an example, the results provided by the two
modeling systems over the computational domain
denoted as A2 are illustrated in Figure 2.
The wave conditions correspond to time frame
2003/10/07/h11.42 and are Hrms = 1.03 m, Tp = 9.5 s,
Dir = 28.7 , Tl = 2.6 m (high tide) and the incoming
waves are considered regular.
Thus Figure 2a presents the coherent wave field
resulting from the simulations with FUNWAVE while
Figure 2b the same wave field photographed from an
aircraft of the Portuguese Air Force. The nearshore
currents resulted from the field measurements are
represented as well in the figure. Figure 2c illustrates
the Hrms fields, wave and current vectors and
the breaking line as resulting from the simulation
with REFDIF1-SHORECIRC modeling system (area
rotated with 22 counter clockwise). The results illustrated in Figure 2 show in general a good concordance
between the two models and the field data.
5 ANALYSIS OF THREE CASE STUDIES

For three relevant case studies the results provided by
the two wave models under evaluation (FUNWAVE
and REFDIF1) were analysed and compared. These
are: case study 1 (CS1) that corresponds to the time
frame 2003/10/07/h09.03, case study 2 (CS2) corresponding to the time frame 2003/10/07/h09.51, and
case study 3 (CS3) for the 2003/10/07/h11.42. For
each computational domain three reference points
that define a cross shore direction were selected. The
positions of these nine reference points are indicated
in Figure 1.
Figure 3 illustrates the wave records provided by
FUNWAVE at the reference points from the first
computational domain (P1, P2 and P3) for CS1.
The distances of the reference points from the
offshore boundary and the resulting Hrms are also
indicated in the figure. Since a cold start condition
528
Figure 3. FUNWAVE results, wave records at the reference points a) P1, b) P2, c) P3. CS1 2003/10/07/h09.03. The
distances of the reference points from the offshore boundary
and the resulting Hrms are also indicated.
Figure 4. REFDIF1 (a) against FUNWAVE (b); results

along the line defined by the reference points P1, P2 and
P3. CS1 2003/10/07/h09.03. The bathymetric configuration
of the cross shore profile is illustrated as well.
Figure 2. Wave conditions corresponding to time frame

2003/10/07/h11.42 (CS 3) for the computational domain
denoted as A2. a) Wave field provided by simulation with
FUNWAVE, b) Aerial photography of the waves, c) Hrms
fields, wave and current vectors and the breaking line
resulting from simulation with REFDIF1-SHORECIRC
(area rotated with 22 counter clockwise).
was used, it can be noticed from Figure 3 that for P1

less than 500 time steps are necessary to achieve the
relative steady state while for P3 about 1500 time steps
are required.
The wave transformation due to the nearshore
bathymetry can be noticed also following the variation of the wave parameter Hrms that was computed
from the wave record using the definition given by

equation (5).
Some other results concerning the wave transformation along the cross shore profile from the first computational domain corresponding to CS1 are illustrated
in Figure 4. Thus Figure 4a presents the REFDIF1
outputs in terms of Hrms and Figure 4b the surface
elevation corresponding to the end of the FUNWAVE
simulation (after 9001 time steps). The bathymetric
configuration of the cross shore profile is illustrated
in Figure 4b as well.
A comparison of Hrms as provided by the two
models is illustrated in Figure 5. Besides Hrms
that is a direct output in the case of the simulations
with REFDIF1 and it was computed with the formula given in equation (5) for FUNWAVE results
(denoted wit FUNWAVE 1 for the standard breaking
configuration).
529
Figure 5. REFDIF1 against FUNWAVE, comparison for

Hrms at the nine reference points considered; a) CS1
2003/10/07/h09.03, b) CS2 2003/10/07/h09.51, c) CS3
2003/10/07/h11.42. FUNWAVE 1 represents the results
with the standard breaking parameterization and FUNWAVE
2 the results after modifying the breaking parameters.
The results illustrated in Figure 5 (REFDIF1 against

FUNWAVE 1) show in general a good agreement
between the two models in terms of Hrms outside
the breaking zone but considerable differences inside
the breaking zone. Some reasons concerning these differences are related to the breaking mechanisms which
are rather different between the two models. Moreover in the case of the simulations performed with
REFDIF1 the effect of the wave-induced currents
on the waves that had generated them was also
accounted while in the simulations performed with the
FUNWAVE model this effect was ignored.
Nevertheless, the main reason for such significant
differences inside the surf zone is related to the fact
that while in FUNWAVE the standard parameterization for wave breaking was considered in the simulations with REFDIF1 the breaking configuration was
set in accordance with the field data. Thus taking
into account the reality in the surf zone the breaking
parameters were tuned in REFDIF1 during the field
experiment and the parameter k that represents the
wave height to water depth ratio (H /h) at the initiation of breaking was set 0.5 from (0.73 which is the
standard value) and the parameter which represents
the inferior limit of the ratio H /h (below this limit the
wave breaking stops) was set to 0.3 (from 0.4).
In Boussinesq models, wave breaking has been
modeled by considering a range of techniques.
FUNWAVE uses an eddy viscosity formulation
similar to that of Zelt, (1991) but with extension to
provide a more realistic description of the initiation
and cessation of wave breaking.
The tunable parameters defined in FUNWAVE are
cbrk, which is a mixing length coefficient that enters
in the expression of the eddy viscosity and has an
empirical value of 1.2 and cbkv which is a coefficient
allowing the variation of the parameter for the breaking scheme. 0.35 is the recommended value for this
parameter and this is related with the ratio between the
surface elevation and the wave celerity when breaking
initially occurs (Chen et al., 2000). Keeping the same
proportion as in REFDIF1, the parameter cbkv was
set to the value of 0.22. After several tests the value
of 1.8 was found more appropriate for the parameter
cbrk.
The results are illustrated also in Figure 5 where
they are denoted as FUNWAVE 2. Only the situations when relevant changes in relationship with the
previous FUNWAVE results occur are represented in
the figure. As reflected by the results illustrated in
Figure 5 the results in the breaking zone in terms of
Hrms are much closer than those provided by REFDIF1
and validated through the field experiment.
The maximum wave heights provided by the
FUNWAVE model in each case study were also
evaluated and the global maximum (with the value of
2.93 m) occurred in the third computational domain
(A3) in the second case study (CS2). On the other
hand, REFDIF1 provided the maximum Hrms for CS1
and A2.
7
CONCLUSIONS
The work developed herewith presents the results of

a FUNWAVE model implementation in the central
part of the Portuguese continental nearshore (in the
coastal environment south of the Portuguese city
Figueira da Foz). The model is based on the Boussinesq equations extended to be valid also in deep and
intermediate water. The main advantage brought by
the application of such phase resolving model would
be that it provides coherent wave fields, but also
these models may account better for some coastal
processes as wave shoaling or combined refractiondiffraction than the models based on the spectrum
concept especially as regards the wave transformation
in the presence of high bathymetric gradients. Some
statistical characteristics of the wave field and information concerning the maximum wave heights can be
also obtained in a more reliable way than when using
the standard Rayleigh distribution.
A field experiment was used as a reference. This
was carried out in the target area in October 2003 in
order to validate the results of a nearshore modeling
system based on the models REFDIF1 as wave driver
and SHORECIRC for nearshore circulation. In this
context the FUNWAVE model simulations were performed for the same time frames as REFDIF1 and
the results of the two models were compared between
them and with the field data. In terms of Hrms , a good
concordance was encountered between the results provided by the two models and the reality in the field
outside the surf zone.
As regards the surf zone area, the results make clear
the fact that in their standard parameterizations both
models do not estimate properly the processes associated with the depth induced wave breaking from
530
the target area and significant modifications of the

tunable coefficients related to this process are usually
required in the Portuguese nearshore.
The work is still ongoing and FUNWAVE seems
to be a very promising model for estimating the
nearshore wave transformation in the Portuguese
coastal environment.
ACKNOWLEDGEMENTS
The work presented has been made in the scope
of the project NEARPORT (Development of a
real-time nearshore wave prediction system for
the Portuguese ports) supported by the Portuguese
Foundation for Science and Technology under contract
PTDC/ECM/64373/2006.
REFERENCES
Berkhoff, J.C.W., 1972. Computation of combined
refractiondiffraction. Proceedings of the 13th International Conference on Coastal Engineering, ASCE,
pp. 471490.
Booij, N., 1981. Gravity waves on water with non uniform
depth and currents, PhD thesis, Technical University of
Delft, The Netherlands, 131 pp.
Chen, Q., Kirby, J.T., Dalrymple, R.A., Kennedy, B.A.,
Chawla, A., 2000. Boussinesq modelling of wave
transformation, breaking, and runup. II: 2D. Journal of
Waterway, Port, Coastal, and Ocean Engineering, ASCE
126 (1), 5762.
Dalrymple, R.A., Kirby, J.T. and Hwang, P.A., 1984. Wave
diffraction due to areas of energy dissipation, Journal of
Waterway, Port, Coastal, and Ocean Engineering, ASCE,
110, 6779.
Gomes, F., Pacheco, L.B., Silva, J., Silva, R., Rusu, E.,
2004. Using GIS in the Evaluation of the Wave Induced
Circulation in the Portuguese Nearshore, EGS General
Assembly, Nice, France.
Kennedy, B.A., Chen, Q., Kirby, J.T., Dalrymple, R.A., 2000.

Boussinesq modelling of wave transformation, breaking
and runup. I: 1D. Journal of Waterway, Port, Coastal, and
Ocean Engineering, ASCE 119 (6), 618638.
Kirby, J.T. and Dalrymple, R.A., 1994. Combined Refraction/Diffraction Model REF/DIF1 version 2.5,
Documentation and Users Manual, Centre for Applied
Coastal Research, University of Delaware, Newark, DE
19716 CACR Report No. 9422, 171p.
Kirby, J.T., Wei, G., Chen, Q., Kennedy, A., Dalrymple, R.A.,
1998. FUNWAVE 1.0: Fully Nonlinear Boussinesq Wave
Model Documentation and Users Manual, Research
Report CACR-98-06, Center for Applied Coastal
Research, University of Delaware.
Mei, C.C., Tuck, E.O., 1980. Forward scattering by thin
bodies, SIAM, Journal of Applied Mathematics, 39,
178199.
Nwogu O, 1993. An alternative form of the Boussinesq
equations for nearshore wave propagation, Journal of
Vol 119, Issue 6, pp. 618638.
Peregrine, D.H., 1967. Long waves on a beach. J. Fluid Mech.
27, 815827.
Rusu, E. and Guedes Soares, C., 2010. Validation of
Two Wave and Nearshore Current Models, Journal of
Vol. 136, Issue 1, January/February 2010, pp. 2745.
Silva, R., Jorge da Silva, A., Rusu, E., Oliveira, F.,
Larangeiro, S., Taborda R., 2004. Evaluation of the
Longshore Current for a Sector of the Portuguese West
Coast: Application of Different Methodologies, Coastal
Engineering 2004, World Scientific Pub Co Inc 2005,
Vol. II, pp. 14551467.
Svendsen I.A., Haas, K. and Zhao, Q., 2002. Quasi-3D
Nearshore Circulation Model SHORECIRC, version 2.0,
Center for Applied Coastal Research, University of
Delaware, Newark, DE 19716 U.S.A.
Wei, G., Kirby, J.T., Grilli, S.T., Subramanya, R., 1995.
A fully nonlinear Boussinesq model for surface waves.
I: highly nonlinear, unsteady waves. Journal of Fluid
Mechanics 294, 7172.
Zelt J.A., 1991. The run_up of nonbreaking and breaking
solitary waves, Coast. Engrg. 15, 205246.
531

Modelling the wave conditions in the Archipelago of Azores

L. Rusu, P. Pilar & C. Guedes Soares
ABSTRACT: The objective of the present work is to evaluate the wave conditions in the Azores Archipelago
using numerical models. Taking into account the geographical position of the nine islands that form the
Archipelago of Azores, the wave conditions are usually affected by various wave systems propagating across
the North Atlantic Ocean. As a general feature, the archipelago is subjected to both the direct approach of the
swell coming from distant storms as well as sea waves generated by local winds that create a local wave system
and modify the incoming swell fields. A wave prediction system based on the two state of-the-art spectral phase
averaging wave models, WAM applied at ocean scale and SWAN applied at regional and local scales was implemented in the Archipelago of Azores. Hindcast simulations were carried out for high energetic situations in the
North Atlantic Ocean. The validation of the results was made with altimeter data that provide good coverage for
the wave conditions.
INTRODUCTION
The Azores Archipelago is a group of nine islands

located in the North Atlantic Ocean approximately
1200 km west of Portugal; the entire archipelago is oriented west-northwest to east-southeast (Figure 1). The
islands are volcanic in origin and are the largest group
of peaks of the Mid-Atlantic Ridge to form islands.
These nine islands are divided in three groups: the Oriental group (Santa Maria and So Miguel), the Central
group (Terceira, Graciosa, So Jorge, Pico e Faial) and
the Occidental group (Flores and Corvo). The islands
of the Occidental group are relatively distant from the
others (about 234 kilometers of Faial) and they are
the most occidental territory of Portugal and Europe.
The bathymetry map of the Azores Archipelago is
Being bounded by physical boundaries that prevent
the exchange of resources and livelihoods with other
regions, the islands are dependent on a set of resources,
not only as vital nature, but also all those who support
the economy (transport, fisheries, tourism) as well as
those who determine its safety. All these factors are
dependent on weather climate, which include also the
sea state conditions.
From geographical reasons the Atlantic islands
are exposed to various waves systems that propagate
across the North Atlantic Ocean. When the waves
come from the west, southwest, or northwest, a sheltering effect occurs and an area protected from the high
waves towards Portugal continental appears (see Ponce
de Len & Guedes Soares 2005).
The archipelago is subjected to both the direct
approach of the swell coming from distant storms as
well as sea waves generated by local winds that create
a local wave system. When the swell system crosses

the archipelago the wave directions are significantly
modified and this induces the occurrence of various
wave systems with different directions in the coastal
environment of the archipelago.
In the last years, several vessels were lost due to
the extreme wave conditions that are often present in
the vicinity of the Azores area. Severe crossing sea
is particularly dangerous for small vessels because
the captain loses the ability to use the vessel heading to protect against beam seas (Niclasen et al.
2010).
In this region, characterized by an important
maritime activity, few in-situ measurements were performed to allow relevant studies concerning the wave
conditions. The most recent measurements become
available with the implementation of the project
CLIMAAT (Azevedo & Gonalo 2005).
Taking also into account the fact that the HIPOCAS
project Hindcast of Dynamic Processes of the Ocean
and Coastal Areas of Europe (Guedes Soares 2008)
has produced a database of 44 years of wind and
wave data along European waters, these data can be
used to force the wave modelling system and perform high resolution simulations around the islands,
in order to provide better knowledge of the sea states
in this region. The simulations carried on in the Iberian
nearshore demonstrated the high quality of the wind
and wave data used (see for example Rusu et al.
2008c; Rusu & Guedes Soares 2009).
To quantify the prediction skill, the wave
model performance was evaluated against altimeter
measurements over the entire region.
In this perspective, the goal of the present work
is to carry on hindcast studies based on numerical
533
Table 1.
Bathymetric grid details used in WAM model.

Resolution
Limits
Grids
Lat
Long
South
North
West
East
Coarse 1
Coarse 2
Medium
2
1
0.5
2
1
0.5
14
20
24
70
68
60
64
50
33
20
4
0
Table 2.
Computational grids for the SWAN simulations.
Grids
X axis ( )
Y axis ( )
x y ( )
Large area
Central area
Oriental area
8.5
2.5
2.0
5.0
1.5
2.0
0.05 0.05
0.01 0.01
0.01 0.01
Figure 1. Bathymetric map of the Azores Archipelago and

geographical spaces of the SWAN simulations.
models for assessing the wave conditions in the

Azores Archipelago.
The methodology proposed herewith is based on
the two state of-the-art spectral phase averaging wave
models. These are WAM (WAve prediction Model)
as wave generation model (WAMDI Group, 1988)
applied at ocean scale, and SWAN (Simulating Waves
Nearshore) applied at regional and local scales (Booij
et al. 1999). Ocean-scale simulations, covering almost
the entire North Atlantic basin with WAM, provide
the boundary conditions for SWAN implemented in a
large area that covers the Archipelago of Azores. The
same computational scheme was applied with good
results in Madeira Archipelago as presented in Rusu
et al. (2008a,b) and in Rusu (2011).
2 THE WAVE PREDICTION SYSTEM FOR

THE AZORES ARCHIPELAGO
The present WAM implementation covers the entire
North Atlantic basin and some details concerning the
bathymetric grids are presented in Table 1. The number
of frequencies used to describe the wave spectrum is
25 (the lowest frequency is equal to 0.0418 Hz) and
the number of directions is 24. The model time steps
used are 300 s for the propagation time step and for
the source time step are 900 s.
The SWAN simulations are made first in a large
area that covers the entire Archipelago of Azores, and
after that two areas with higher spatial resolution are
nested inside. The bathymetries of the higher resolution areas cover the Central and Oriental groups.
Details about the computational grids used for the
SWAN simulations are presented in Table 2 and
Figure 1. The implementation of the SWAN model
was made for 36 directions and 30 frequencies logarithmically spaced from 0.04 Hz to 0.6 Hz at intervals
of f /f = 0.1, and the simulations were performed in
the non-stationary mode.
Figure 2. Significant wave height fields and wave direction

vectors in Azore area energetic peak 2001/02/06-h06.
The reanalyzed wind fields from the North Atlantic

basin and the eastern European coasts, determined
hourly with a resolution of 0.5 in the HIPOCAS
project were used as input for the two wave models.
More details as regards the wind data fields are
presented in Pilar et al. (2008).
It was assumed that there are no currents, which
implies that refraction is due only to spatial variations
of water depths. Model outputs were produced and
comparisons with remotely sensed data are carried out
in the next sections.
FIELD DATA ANALYSES
The simulations were performed in January and

February 2001. This time interval is characterized
by both highly and normal energetic conditions. For
example, at the beginning of January wave groups with
significant wave heights of about 13m approached
the northern part of the archipelago, and in 6 of
February the North-Northwest part of the archipelago
was affected by a strong storm when the wind velocity was sometimes greater that 25 m/s. Figure 2 shows
that also at this date (6 of February 2001) wave fields
with significant wave heights of about 13m approach
again the islands Corvo and Flores from the Occidental
534

vectors in Central area 2001/01/03-h21.

vectors in Oriental area 2001/01/03-h21.
group, but also the islands from the western part of the
central group.
In the study of Ponce de Len & Guedes Soares
(2005) it was pointed out that the Azores islands
may contribute to the dissipation of the spectral wave
energy under certain weather conditions. Thus, for
the energetic situation presented in Figure 2 when the
wave direction is from North-Northwest, a strong
sheltering effect to the South-Southeast of the islands
was observed associated with the passage of the storm.
In general, in the shadowed area occur wave systems
with various directions that might be very dangerous for the small vessels navigating in the area. Such
examples are illustrated in Figure 3 and Figure 4 that
present the significant wave height fields resulted from
the simulations carried out in the Central and Oriental
areas. The representation of the wave direction vectors shows also quite clearly the refraction effects in
the vicinity of the islands.
4
4.1
EVALUATION OF THE RESULTS

Comparisons with altimeter data
For the period analyzed global altimeter significant

wave height (SWH) data set provided from the four
Figure 5. ERS-2, TOPEX, Poseidon and GEOSAT FollowOn (GFO) satellite tracks over the Azores Archipelago area
in January 2001.
altimeter missions ERS-2, TOPEX, Poseidon and

GEOSAT Follow-On (GFO) were used. Taking also
into account that the wave model results depend
directly on the quality of the wind data, the accuracy
of these fields will be verified with satellite data.
The altimeter footprints along the satellite tracks
provide a large spatial coverage that cannot be accomplished by in-situ observations at fixed stations. The
ground tracks for January 2001 over the geographical space of the Azores Archipelago are shown in
Figure 5. As the above figure shows, the combined
coverage from tracks of the all satellites is extensive
and relatively uniform over the area.
The results of the wave modelling system were
interpolated in space with a bilinear interpolation, and
in time to collocate with the altimeter data. To do this,
hourly model SWH within 60 min of the altimeter collection time were spatially interpolated from the grid
points to the locations of the altimeter measurements
along the satellite tracks, and in time to fit the time of
the satellite pass, which is very similar to the method
used by Cavaleri & Sclavo (2006). In all cases, collocation files were used to compute the statistics and
produce scatter diagrams that present the measured
against the simulated values. The same procedure was
applied for the wind data.
The comparisons with buoy data carried out by
Queffeulou (2004) showed that the altimeter SWH is
in general in agreement with the in-situ data, with
standard deviations of differences of the order of
0.30 m, but tends to slightly overestimate low SWH
and to underestimate high SWH. He has established corrections to SWH that are in general linear and correspond to a few percents of SWH.
These corrections were considered in relationship
with the altimeter data used in this study.
Figure 6 shows the spatial variation of SWH measured by GFO altimeter along the track that passed
in 24 of January 2001 at 12 hours over the Azores
Archipelago and simulated by the interpolated model
results. The GFO altimeter data provide point estimates at about seven kilometer intervals, along the
ground track of the satellite.
It can be observed that the satellite data present
a greater variability in comparison with the SWAN
535

SWH values and corresponding altimeter data in the large
SWAN area.
Figure 6. Comparison of GFO altimeter SWH data along

track (cycle 59, in 2001/01/24-h12), against interpolated
SWAN SWH. The x-axis represents the latitude of the
altimeter footprints.
results, despite the fact the spatial resolution of the

wave model and those of the satellite measurements
are sensible equal. This is probable due to the low resolution of the wind field used (0.5 , which means appreciatively 50 km), resolution that cannot represent with
accuracy the local phenomena. In general, the model
represents well the enhancements and decreases of
SWH along the track. In the southern part SWH is
underestimated by the model, effect that is accentuated in the points that correspond to the zone located
between the islands of the central group (about 38.5
latitude). As the satellite goes out the environment of
the islands a slight overestimation of SWH occurs.
4.2
Month
Bias
RMSE
SI
2001/01
2001/02
0.07
0.08
0.71
0.75
0.17
0.23
0.90
0.96
0.99
1.06
4433
3856

SWH and corresponding altimeter data (for each altimeter).
Month
Bias
RMSE
SI
ERS-2
TOPEX
Poseidon
GFO
0.04
0.07
0.04
0.11
0.69
0.80
0.68
0.69
0.23
0.21
0.21
0.17
0.81
0.96
0.93
0.95
0.99
1.04
1.04
1.00
2012
2806
473
2998
Statistical analysis
The validation statistics for SWH obtained from the

SWAN model (in large area) against satellite measurements are presented in Table 3. The statistical
parameters considered are the mean error computed as
the difference between the simulated and observed values, divided by the number of observations N (Bias),
the root mean square error (RMSE), the scatter index
(SI) defined as the ratio of standard deviation of error
to the mean observed SWH, the linear correlation
coefficient (R) and the symmetric slope (S), that is
the coefficient of the regression line through the origin (model against measured data). Model values are
overpredicted for S > 1 and underpredicted for S < 1.
The wave statistics presented in Table 3 shows that
the SWAN models results are good and in agreement
with the satellite measurements, with slopes close to
1, low values of the scatter indexes and high correlation coefficients. The fits are better in January that
is characterized by higher waves and better defined
meteorological situations.
Statistical analyses of the SWH corresponding to
each satellite that passed over the Azores Archipelago
in the two-month period were also carried out and the
results are presented in Table 4. These results show
that the best results were obtained against GEOSAT
Follow-On.
Figure 7 shows the scatter plots of SWH (altimeter
against wave model data) for January and February
2001. The blue lines denote perfect fit to model and
observed values. A linear regression (green line) was
adjusted to each data set as well, and it is observed
Figure 7. Significant wave height scatter plots for January

2001 (a) and February 2001 (b).
that in January the model slightly underestimates the

extreme significant wave heights, while in February
overestimates them.
It is expected by increasing the resolution of the
SWAN simulations in the geographical space by
nesting new areas inside the ones initially defined
will improved the model results. From this reason two
new SWAN computational domains with increased
spatial resolution were defined (see Table 2). These
areas cover the Central and Oriental groups of islands.
The simulations in the above two areas well carried out
only for January 2001, when the statistical results from
the large area presented in Table 3, are better.
The evaluation of the quality of the results from
these two areas was made by means also of the statistical parameters, the collocations files being obtained
by applying the methodology above mentioned.
In order to compare the results obtained in the large
SWAN area that covers the entire archipelago with
those from the Central and Oriental computational
domains, only the data covering these last two areas
(both simulations and measurements) were extracted
from the initial collocations files.
Thus, for each area either Central or Oriental, the
statistical analyses will be performed both considering
the results with higher and lower spatial resolution.
Table 5 presents the statistical results for the two areas,
were LR (low resolution) presents the results from the
536

SWH values and corresponding altimeter data in the Central
and Oriental SWAN areas for January 2001.
Month
Central area
LR
HR
Oriental area
LR
HR
Bias
RMSE
SI
0.12
0.08
0.67
0.65
0.17
0.16
0.90
0.91
0.99
0.99
411
411
0.26
0.28
0.67
0.67
0.19
0.19
0.84
0.84
1.09
1.09
468
468
For the evaluation of the results provided by the

SWAN model simulations in the Archipelago of
Azores multi-satellite significant wave heights were
used. The analysis of the statistical results show that
in general the model system represented reasonable
accurate the wave conditions from the zone analyzed.
It was also noticed that by increasing the bathymetric
resolution an enhancement of the accuracy occurs in
the Central area, while in the Oriental area this seems
that not affects significantly the system reliability.
Due to the difference that occurred in the quality of
the wave model results, when comparing the month
of January with February, an evaluation of the quality
of the wind field was also performed. This allows a
quantification of the degree in which the wind field
influences the quality of the wave results. Thus, from
this analysis resulted that the quality of the wind data
is considerably lower in February in relationship with
January.
Nevertheless, it can be also observed that as
regards the statistical parameters, the differences for
SWH are lower than those for the wind speed for the
two time intervals considered.
Although the results obtained in the framework of
the present study can be considered as reasonable,
extended periods analyses will be required in order
to provide a more complete picture. This will include
also a better evaluation of the influence of the REMO
wind field accuracy on the results provided in the
Archipelago of Azores by the modeling system based
on spectral phase averaged wave models.
Table 6. Statistics of the comparison between REMO wind

speed values and corresponding altimeter data over the large
SWAN area.
Month
Bias
RMSE
SI
2001/01
2001/02
0.04
0.94
1.74
2.67
0.17
0.25
0.89
0.81
1.01
0.93
4433
3856
CONCLUSIONS
Figure 8. Wind speed scatter plots for January 2001 (a) and
February 2001 (b).
ACKNOWLEDGEMENTS
larger area while HR (high resolution) presents the

statistical results obtained in the Central and Oriental
areas for the simulations corresponding with higher
resolution in the geographical space.
The results presented in Table 5 indicate the fact that
in the Central area the increasing of the bathymetric
resolution induces also a slight improvement of the
statistical results, especially in terms of Bias and
RMSE. In the Oriental zone no enhancement of the
accuracy can be observed.
An analysis of the quality of the wind data will
be performed next. Thus Table 6 presents the statistical results corresponding to the wind speeds. It can
be observed that in February al the statistical parameters indicate lower quality of the results and this
influenced the quality of the results provided by the
wave model. Figure 8 shows the scatter plots of wind
speeds (altimeter against simulated data) for January
and February 2001. Probably due to the lower resolution of the wind fields (0.5 ), the local phenomena
could not be properly represented and from this occurs
the large dispesion in the results from February.
The first and second authors have been funded by

Fundao para a Cincia e Tecnologia (Portuguese
Foundation for Science and Technology) under posdoctoral grant SFRH/BPD/65553/2009 and doctoral
grant SFRH/BPD/41063/2007 respectively.
REFERENCES
Azevedo, E.B. & Gonalo, V. 2005. The CLIMAAT project
and its contribution to the monitorization and characterization of waves in the Azores archipelago, (in
Portuguese), Proc. 4as Jornadas de Engenharia Costeira e
Porturia, Angra do Herosmo, Portugal, 2021 October,
CD Ed., 16p.
Booij, N., Ris, R.C. & Holthuijsen, L.H. 1999. A thirdgeneration wave model for coastal regions, 1, Model
description and validation, J. Geophys. Res. 104: 7649
7666.
Cavaleri, L. & Sclavo, M. 2006. The calibration of wind and
wave model data in the Mediterranean Sea. Coastal Eng.
53: 613627.
Guedes Soares, C. 2008. Hindcast of Dynamic Processes
Engineering, 55(11): 825826.
537
Niclasen, B.A., Simonsen, K. & Magnusson, A.K. 2010.

Wave forecasts and small-vessel safety: A review
of operational warning parameters, Marine Structures
23(1): 121.
Pilar P, Guedes Soares C. & Carretero J.C. 2008. 44-Year
hindcast for the north east Atlantic European coast,
Coastal Enginering 55(11): 86171.
Ponce de Len, S. & Guedes Soares, C. 2005. On the
sheltering effect of islands in ocean wave modelling.
Journal of Geophysical Research 110, C09020.
Queffeulou P. 2004. Long term validation of wave height
measurements from altimeters, Marine Geodesy 27:
495510.
Queffeulou P. & CroizFillon, D. 2007. Investigation of
large scale and regional features of wave height using
multisatellite altimeter measurements. Proceedings of
the EUMETSAT Meteorological Satellite Conference,
Amsterdam, The Netherlands, 2428, September 2007,
EUMETSAT P.50.
Rusu, E. 2011. Strategies in using numerical wave models
in ocean/coastal applications. Journal of Marine Science
and Technology Taiwan 19(1): 5875.
Rusu, E. & Guedes Soares C., 2009. Numerical modeling to estimate the spatial distribution of the wave energy
in the Portuguese nearshore, Renewable Energy 34(6):
15011516.
Rusu, E., Guedes Soares C. & Pilar, P. 2008a. Evaluation
of the Wave Conditions in Madeira Archipelago with
Spectral Models, Ocean Engineering 35(13): 13571371.
Rusu, E., Pilar, P. & Guedes Soares, C. 2008b. Development of a Wave Prediction System for the Madeira
Archipelago, Maritime Industry, Ocean Engineering and
Coastal Resources, P.Kolev, C. Guedes Soares (Eds.),
Francis & Taylor, London, pp. 787799.
Rusu, L., Pilar, P. & Guedes Soares, C. 2008c. Hindcast of
the wave conditions along the west Iberian coast. Coastal
Engineering 55(11): 906919.
WAMDI Group: Hasselmann, S., Hasselmann, K., Bauer, E.,
Janssen, P.A.E.M., Komen, G. J., Bertotti, L., Lionello,
P., Guillaume, A., Cardone, V.C., Greenwood, J. A.,
Reistad, M., Zambresky, L. & Ewing, J. A. 1988. The
WAM Model AThird Generation Ocean Wave Prediction
Model, J. Phys. Ocean. 18: 17751810.
538

The global wave age climate

Alvaro Semedo
Uppsala University, Uppsala, Sweden
CINAV Escola Naval, Lisbon, Portugal
Anna Rutgersson
Uppsala University, Uppsala, Sweden
Andreas Sterl
Royal Netherlands Meteorological Institute (KNMI), De Bilt, Netherlands
Kay Suelj
Jet Propulsion Laboratory, California Institute of Technology, Pasadena-California, USA
ABSTRACT: A global climatology of the wave age parameter, based on the ERA-40 wave reanalysis, is
presented in the present study. The annual and seasonal climatological patterns of the wave age are shown,
confirming, in line with previous studies, the global predominance of swell waves in the World Ocean. The
interannual variability of the wave age fields are studied by means of an EOF (empirical orthogonal function)
analysis. The leading modes of variability of the wave age demonstrate that the areas of higher explained
variability (the centers of action) occur mostly along the equator, coinciding with the so called swell pools,
where long waves coincide with light winds.
1
INTRODUCTION
Of all the types of waves at the ocean surface wind

waves account for more then half of the energy, surpassing the contribution of tides, tsunamis, coastal
surges, etc. (Kinsman 1965). Wind waves (henceforth
simply called waves) are generated by the wind and
propagate under the restoring force of gravity. At the
ocean surface waves that are generated locally by the
prevailing winds (wind sea waves) co-exist with waves
that were generated elsewhere and propagated into the
area (swell waves). Swell waves can propagate very
long distances across the World Ocean with little attenuation (Ardhuin et al. 2009). In the open ocean wind
sea and swell waves co-exist more often than not as
the result of contributions from waves systems with
different origins and ages.
Winds generate waves, but at the same time the
airflow in the lower atmosphere (1 m) is modified
due to the loss of energy in the form of momentum
to the waves (Janssen 1991). On the other hand as
swell waves propagate into light wind areas, energy
and momentum are also transferred to the lower atmosphere in a feedback process related to the swell
attenuation (Smedman et al. 1999, 2009; Semedo et al.
2009; Ardhuin et al. 2009; Sullivan and McWilliams
2010). Therefore the air-sea interaction depends in a
sensitive way of the state of the wave field, i.e., of
the state of development of the waves, which can be

expressed by the wave age parameter.
The wave age parameter, developed along with the
sheltering hypotheses of Jeffreys (1924, 1925) to designate the characteristic wave phase speed normalized
by a measure of the wind speed at a certain height,
expresses the relative speed between the peak wave
and the wind:
where cp = g/2fp is the wave phase speed of the peak

of the spectrum (g is the gravitational acceleration,
and fp is the peak frequency in deep waters), and U10 is
the wind speed at 10 m height. Asymptotic values for
fp for a fully developed wave spectrum estimated by
Pierson and Moskowitz (1964), and further corrected
for 10 m height by Alves et al. (2003), allow a characterization of the sea state as dominated by wind sea
(or young growing waves) if 10 < 1.2, or dominated
by swell if 10 > 1.2.
The wave age gives a statistically meaningful qualitative description of the sea surface, and has been
largely used in different aspects of the air-sea interaction theory (Csanady 2001). The wave age is also
used, for example, for practical modeling solutions of
wave-atmosphere coupled model systems, where the
539
feedback of the waves to the winds is modeled via a

wave age dependent Charnock parameter (Charnock
1955, 1958).
In the present study we present an insight on the
10 climate and interannual variability, based on the
ERA-40 reanalysis. The characterization of the wave
field using 10 can only be achieved from wave spectra, available globally only from wave model results
like the European Centre for Medium-Range Weather
Forecasts (ECMWF) ERA-40 reanalysis. The ERA-40
is the longest and most complete global wave data set
available, comparing better with observations in terms
of rmse (root mean square error) index than any other
reanalysis (Caires et al. 2004).
This remainder of the paper is organized as follows.
The ERA-40 reanalysis is described on section 2, as
well as the method used to compute 10 . The climatology and interannual variability of 10 are presented in
section 3. The summary and concluding remarks are
presented in section 4.
DATA AND METHODS OF ANALYSIS
The ERA-40 is a reanalysis of meteorological observations, including wave parameters, produced by

ECMWF (Uppala et al. 2005), covering 45 years, from
September 1957 to August 2002. The data set consists
of 6-hourly global fields of wave parameters with a
1.5 1.5 grid spacing resolution. The ERA-40 was
produced using a two-way coupled wave-atmosphere
model system. The wave model used in the coupled
system is the state of the art third generation Wave
Model (WAM; WAMDI Group 1988; Komen et al.
1994). Due to changes in the assimilated data, the
ERA-40 wave data is not completely homogeneous.
From December 1991 to May 1993 erroneous remote
sensing data were assimilated giving rise to corrupted
wave model output. Data from this period is therefore not used in the present study. Additional details
about the ERA-40 wave reanalysis are given by Caires
et al. (2004), Sterl and Caires (2005), Caires and Sterl
(2005), and Caires et al. (2005). Additional details
about the computation of the ERA-40 integrated wave
parameters can be found on Bidlot (2001) and Semedo
et al. (2011).
The most important caveat of the ERA-40 wave
data is the underestimation of the Hs for tall waves.
Caires and Sterl (2003) have also showed that high
wind speeds in ERA-40 are also slightly underestimated. We have no means to show how these caveats
can influence the overall global climatological of 10 ,
but we expect them to have a marginal effect.
A factor of cos (where is the relative angle
between the prevailing or mean wave and wind directions) for the wind speed (U10 cos ) is sometimes used
in the computation of the wave age, justified with the
need to take into account the component of the wind
aligned with the wave direction, with the argument that
is most of the times small making cos close to unity.
While this might be the case in certain areas of the
globe, mainly along extratropical storm tracks, it is far

from being a general characteristic of the World Ocean
wave field (Semedo 2010). Hgstrm et al. (2011), following Smedman et al. (2009), showed that the effect
of swell on the lower MABL is independent of the
relative angle between the waves and the wind propagating direction and therefore the cos should not be
used. In the present study the wave age will be computed using equation (1), since we are interested in the
relative speed between cp and U10 , regardless of the
angle between the waves and the wind directions.
CLIMATOLOGY OF THE GLOBAL

WAVE AGE
The wave age 10 was computed for each 6-hourly

gridded global fields and then then averaged. The four
seasons are organized as December to January and
February (DJF), March, April and May (MAM), June,
July and August (JJA), and September, October and
November (SON). Here we focus on annual and DJF
and JJA values only.
Figure 1 displays the global climatological annual
mean field of 10 . The highest climatological mean
values are located in the equatorial west continental
coasts, coinciding with the swell pools mentioned by
Chen et al. (2002), and Semedo et al. (2011). Some
high mean wave age values can also be found in
the tropical areas along the Horse Latitudes in both
hemispheres, where the mean U10 values are low, and
long swell waves generated in the extratropical latitudes travel predominantly equatorward (Semedo et al.
2011). The highest values (exceeding 8) are located
in the in the Pacific and Indian Oceans. In the east
equatorial Pacific Ocean, northeast of New Guinea,
there is also an area of high mean 10 values.
Along the extratropical storm tracks and the Trade
Winds, in both hemispheres, the mean 10 is much
lower, ranging from 2 to 3.
The global annual mean field of 10 is, on the average, always consistently higher than 1.2 everywhere
in the open ocean, supporting the swell dominance of
the ocean previously mentioned by Chen et al. (2002)
and Semedo et al. (2011). The spatial variability of
the wave age is a composite of the zonal variation of
cp (Young, 1999) and the mainly meridional variation
Figure 1. Annual averages of 10 (dimensionless).
540
of U10 (Semedo et al. 2011). Outside the equatorial

regions the meridional variation of 10 dominates,
with well define tripoles. Along the equator the zonal
variation of 10 is more predominant.
Figure 2 shows the global climatological means of
10 for DJF and JJA. In the winter (summer) hemisphere the mean 10 along the extratropical storm
tracks decreases (increases), mostly due to the increase
(decrease) of the mean U10 . Along the Trade Winds,
mostly during the hemisphere summer, there is a small
increase of 10 . In the swell pools the seasonality is
more pronounced, with the wave field being less swell
dominated in JJA in those areas in all ocean basins,
especially in the Atlantic Ocean.
Table 1 presents the seasonal variations of the wave
age (geographical, hemispherical, and global, obtained
by averaging 10 ). The mean is always consistently
higher than 1.2, in accordance with the swell dominance mentioned before. The swell dominance in the
Pacific Ocean is always higher than in the Atlantic

Ocean in all seasons in both North and South subbasins. The Indian Ocean is the most swell dominated
Ocean, and the Southern Hemisphere is more swell
dominated than the Northern Hemisphere.
An empirical orthogonal function (EOF) analysis
was performed to the DJF and JJA 10 fields to investigate the main patterns of variability. The EOF analysis
(e.g. von Storch and Zwiers 1999) decomposes multivariate data into sets of orthogonal spatial patterns
(the empirical orthogonal functions), and corresponding time series (the principal components PCs) that
maximize the spatially averaged explained variance
of the original data. The PCs are normalized to have
unit variance, so that the corresponding EOFs represent the typical variability of the data in their original
units. The three major ocean basins are analyzed separately to avoid the masking of inter-ocean variability
patterns. Here the Pacific, Atlantic and Indian Oceans
include their respective Southern Ocean sectors. The
percentages of the explained variances by the first
the EOF1s are shown in Table 2.
Figure 3 shows the DJF and JJA 10 EOF1 spatial
patterns. The areas of highest explained variance occur
around the equator. Long waves generated by intermittent extratropical storms propagate equatorward along
great circle paths into areas of light winds, contributing
to episodes of high cp /U10 that are replaced by a
Table 2. Empirical Orthogonal Functions explained variance of EOF1 (in %).
EOF1
Pacific
Atlantic
Indian
DJF
JJA
13.8
16.3
18.2
11.5
18.8
15.9
Figure 2. Seasonal averages of 10 (dimensionless) for DJF

(A) and JJA (B).
Table 1. Seasonal variations of the mean wave age. The
Pacific Ocean, Atlantic Ocean, and Indian Ocean include
their respective SO sectors.
Mean wave age
Hemisphere
Basin
N.Hemisphere N. Atlantic
N. Pacific
N. Indian
Total
S.Hemisphere S. Atlantic
S. Pacific
S. Indian
Total
Global
DJF MAM JJA SON Tot

2.84
2.86
4.50
2.89
3.26
3.54
3.67
3.40
3.18
2.99
3.30
5.42
3.08
3.12
3.37
3.57
3.26
3.27
3.02
3.51
3.18
3.30
2.69
2.84
2.94
2.79
3.04
3.01
3.15
5.23
3.12
2.73
3.21
3.26
2.98
3.13
2.98
3.30
4.59
3.09
3.01
3.27
3.41
3.16
3.18
Figure 3. First EOFs of the detrended 10 for the Pacific

Ocean in DJF and JJA (A and D), Atlantic Ocean in DJF and
JJA (B and E), and Indian Ocean in DJF and JJA (C and F).
The color scales vary between the seasons.
541
decrease in the wave age values after the wave front has
passed. In DJF, in the Pacific and Atlantic Oceans the
spatial patterns of the leading modes of 10 coincide
with the swell pools. The swell propagation patterns
in the Pacific Ocean are complicated, particularly in
DJF, since they are affected by waves generated in both
hemispheres. The area of high explained variability
in the centre of action is most certainly the result of
the arrival of wave fronts generated in the extratropical latitudes in both hemispheres (Young 1999). The
situation in the Atlantic Ocean although similar is different. The wind speeds along the extratropical storm
area in the South Atlantic, in DJF, are rather low. in
fact the lowest of all the extratropical areas (Semedo
et al. 2011), and the northward penetration of swell
waves generated in the South Atlantic is marginal compared to the predominant south bound swell generated
by the North Atlantic DJF extratropical storms. Therefore the maximum of explained variance (the centre of
action) located in the central equatorial Atlantic Ocean
between the Gulf of Guinea and Brazil is mostly due
to intermittent swell fronts generated in the North
Atlantic storm tracks. A similar situation occurs in JJA,
but now due to swell waves generated along Southern
Atlantic storm tracks and propagating northward.
The situation in the Indian Ocean in DJF (summer
in the Southern Hemisphere) shows a well defined
dipole along the equator, and a secondary dipole along
the extratropical latitudes. In JJA, also in the Indian
Ocean, is now similar to DJF in the Pacific and
Atlantic Oceans, with a well defined pattern maximum of explained variance west of Sumatra. In JJA, in
the Pacific and Atlantic Oceans, well defined dipoles
arte now present. In the Pacific ocean the area of
maximum of explained variance Northeast of New
Guinea.
4
CONCLUSIONS
This study presented a climatological insight of the

wave age parameter 10 , based on the ECMWF ERA40 wave reanalysis. The annual and the seasonal (DJF
and JJA only) global fields of 10 were presented,
revealing the prevalence of one wave regime (swell)
over the other. The first conclusion from this study
is that the mean wave age values are always higher
than 1.2, the fully developed stage value based on
the Pierson and Moskowitz (1964) spectral theory.
The highest climatological mean values of 10 (higher
then 7) are located along the equatorial west continental coasts coinciding with the areas of highest swell
dominance probability of occurrence mentioned by
Chen et al. (2002) and Semedo et al. (2011) the
swell pools. Long swell waves converge systematically
to these areas, where the wind speeds are usually low,
corresponding to high 10 values. Not surprisingly the
low mean 10 values are found along the extratropical storm tracks, mainly in the respective hemisphere
winter. The average values of 10 revealed that the
swell dominance in the Pacific Ocean is higher that in
the Atlantic Ocean. It also revealed that Indian Ocean
is the most swell dominated of the three main oceans,

which is justified by the fact that is has only one
extratropical storm area, where waves are generated,
propagating northward.
The interannual variability of 10 was studied separately for the Pacific, Atlantic, and Indian Oceans, for
DJF and JJA, by means of an EOF analysis. The areas
of highest explained variability (the centers of action),
roughly coinciding with the areas of highest 10 values,
occur around the equator. The swell fronts crossing the light wind areas along the equator contribute
to the highest explained variance in those areas.
ACKNOWLEDGEMENTS
We greatly appreciated the help of Jean Bidlot from
ECMWF (Reading, United Kingdom) with the many
details of the ERA-40 wave data.
REFERENCES
Alves, J.H., M.L. Banner, and I.R.Young, 2003: Revisiting the
PiersonMoskowitz asymptotic limits for fully developed
wind waves. J. Phys. Oceanogr., 33, 13011323.
Ardhuin, F., B. Chapron, and F. Collard, 2009: Observation of
swell dissipation across oceans. Geophys. Res. Lett., 36,
L06607, doi:10.1029/2008GL037030.
Bidlot, J.-R., 2001: ECMWF wave model products. ECMWF
Newsletter, 91, 915.
Caires, S., and A. Sterl, 2003: Validation of ocean wind and
wave data using tripole collocation. J. of Geophysical
Research, 108(C3): 3098, doi:10.1029/2002JC001491.
Caires, S., A. Sterl, J.-R. Bidlot, N. Graham, and V. Swail,
2004: Intercomparison of different wind wave re-analyses.
J. Climate, 17(10), 18931913.
Caires, S., and A. Sterl, 2005: 100-year return value estimates
for ocean wind speed and significant wave height from
the ERA-40 data. J. Climate, 18(7), 10321048.
Caires, S., and A. Sterl, and C.P. Gommenginger,
2005: Global ocean mean wave period data: Validation and description. J. Geophys. Res., 110, C02003,
doi:10.1029/2004JC002631.
Charnock, H., 1955: Wind stress on a water surface. Q. J. R.
Meteorol. Soc., 81, 639640.
Charnock, H., 1958: A note on empirical wind-wave
formulae. Q. J. R. Meteorol. Soc., 84, 443447.
Chen, G., B. Chapron, R. Ezraty, and D. Vandemark, 2002:
A global view of swell and wind sea climate in the ocean
by satellite altimeter and scatterometer. J. Atmos. Oceanic
Technol., 19, 18491859.
Csanady G.T., 2001. Air-Sea Interaction: Laws and Mechanisms. Cambridge, UK: Cambridge Univ. Press, 237 pp.
Janssen, P.A.E.M., 1991: Quasi-linear theory of windwave generation applied to wave forecasting. J. Phys.
Oceanogr., 21, 16311642.
Jeffreys, H., 1924: On the formation of waves by wind.
Proceedings of the Royal Society, A107, 189206.
Jeffreys, H., 1925: On the formation of waves by wind. II.
Proceedings of the Royal Society, A110, 341347.
Kinsman, B., 1965: Wind Waves. Prentice-Hall, 676 pp.
Komen, G.J., L. Cavaleri, M. Donelan, K. Hasslemann, S.
Hasslemann, and P.A.E.M. Janssen, 1994: Dynamics and
Modelling of Ocean Waves. Cambridge University Press.
520 pp.
542
Pierson, W. J., and L. Moskowitz, 1964: A proposed spectral

form for fully develop wind seas based on the similarity
theory of S. A. Kitaigorodskii. J. Geophys. Res., 69, 5181
5190.
Semedo, A., . Stra, A. Rutgersson, K.K. Kahma.,
H. Pettersson, 2009: Wave induced wind in the marine
boundary layer. J. Atmos. Sci., 66, 22562271.
Semedo, A., 2010: Atmosphere-ocean Interactions in Swell
Dominated Wave Fields. PhD thesis, Univ. Uppsala.
Semedo, A., K. Sueli, A. Rutgersson, A. Sterl, 2010: A
Global View on Wind Sea and Swell Waves Climate and
Variability from ERA-40. Journal of Climate, Vol. 24,
14611479.
Smedman, A., U. Hgstrm, H. Bergstrom, A. Rutgersson,
K.K. Kahma, and H. Pettersson, 1999: A case study of
air-sea interaction during swell conditions. J. Geophys.
Res., 104, 25, 833851.
Smedman, A., U. Hgstrm, E. Sahle, W.M. Drennan,
K. K. Kahma, H. Pettersson, and F. Zhang, 2009: Observational Study of Marine Atmospheric Boundary Layer
Characteristics during Swell. J. Atmos. Sci., 66, 2747

2763.
Sterl, A., and S. Caires, 2005: Climatology, Variability
and Extrema of Ocean Waves The Web-based KNMI/
ERA-40 Wave Atlas. Int. J. Climatology, 25(7), 963-997,
doi:10.1029/joc.1175.
Sullivan, P.P., and J.C. McWilliams, 2010: Dynamics of
winds and currents coupled to surface waves. Annual Rev.
Fluid Mec., Vol. 42, 1942.
Uppala, S. M., and Coauthors, 2005:The ERA-40 re-analysis.
Quart. J. Roy. Meteor. Soc., 131, 29613012.
von Storch, H. and F. Zwiers, 1999: Statistical Analysis in
Climate Research. Cambridge Univ. Press, New York,
494 pp.
Young, I. R., 1999: Seasonal variability of the global ocean
wind and wave climate. Int. J. Climatol., 19, 931950.
WAMDI Group, 1988: The WAM Model A Third Generation Ocean Wave Prediction Model. J. Phys. Oceanogr.
18, 17751810.
543
Renewable energy

Effect of the environment on the design loads on monopile

offshore wind turbine
ABSTRACT: The influence of the environmental conditions on wind turbine design loads for a monopile
foundation is studied by analyzing the bending moment at the tower base and blade root for various values of
water depth, tower height, pile diameter and wind turbulence model. The main design requirements are related
to the limitation of maximum deflection at the tower base which influences the vibration modes of the complete
structure. The analysis is done using FAST code for wind turbine, which operates in the time domain. The detailed
effects of pile diameter, height of tower and water depth are presented. The study is carried out by using a 5 MW
wind turbine with a monopile foundation.
INTRODUCTION
The offshore wind energy is one of the most important

marine renewable energy resources that can contribute
to worldwide energy demands. The placement of
wind turbines in harsh offshore environments is an
engineering challenge, which requires development
of suitable foundation to support the turbines.
In the past two decades it has been observed that
most of the wind energy development took place for
land base wind power. These developments took place
in Europe and in the United States (see Watson et al.
2005, Musial and Butterfield 2004). Going offshore
for the wind energy has the advantage of obtaining
in general more stable and stronger wind regimes and
thus developments have been observed for offshore
wind energy (Bagbanci et al. 2012). More recently,
research on floating offshore wind turbines concepts
is growing due to the even better prospects of offshore
wind resource offshore.
In Europe more than 900 MW of offshore wind
energy capacity has been installed in and around the
North and Baltic seas (Musial et al. 2006). The main
advantages of the installing offshore wind turbine is
due to the following reasons (Henderson et al. 2003,
Musial et al. 2004):
The wind blows strongly and consistently with less

turbulence intensity and smaller shear near the
shore.
The visual and noise annoyances of wind turbines can be avoided if the turbines are installed
a sufficient distance from shore.
Vast areas of uninterrupted open sea are available
and the installations will not occupy land, interfering
with other land uses.
Most of the shallow water offshore wind turbines are

installed on fixed bottom substructures mostly in water
of not more than 20 m by driving monopiles into the
seabed or by relying on conventional concrete gravity
bases. In the shallow water region, monopile offshore
wind turbines are mostly found to be used for the
generation of electricity. So, the detailed study on the
performance of monopile offshore wind turbine is very
necessary. A detailed study on the modal dynamics of
large wind turbine with different support structures
can be found in Bir and Jonkman (2008). Agarwal and
Manuel (2009) simulated the response for long-term
extreme load prediction of offshore wind turbine.
Recently, many countries are involved in the design
and installation of offshore wind turbines in deep
waters. A lot of research is going on the performances of offshore floating wind turbines. Bulder et al.
(2002) used linear frequency-domain hydrodynamics
techniques to find the response amplitude operators
(RAOs) and amplitude standard deviations of the six
rigid-body modes of motion for the support platform
of a tri-floater design for a 5 MW wind turbine. Lee
(2005) used linear frequency domain hydrodynamics
techniques to analyze a TLP design and a taut leg sparbuoy for a 1.5 MW wind turbine. Wayman et al. (2006)
and Wayman (2006) also used a similar process to analyze multiple TLP designs and a shallow drafted barge
(SDB) design for a 5 MW wind turbine. More recently,
through frequency domain analysis, Vijfhuizen (2006)
designed a barge for a 5 MW wind turbine, which was
also a platform for an oscillating water column (OWC)
wave energy device.
Since, most of the support platforms that have
been proposed for floating wind turbines are more or
less axisymmetric, and because there is no hydrodynamic mechanism that will induce yaw moments on
547
such floating bodies, one might question whether the

support platform yaw rotation degree of freedom is
necessary. The wind turbine, however, induces yaw
moments that are primarily the result of (i) the aerodynamic loads on the rotor when a yaw error exists
between the rotor axis and nominal wind direction;
and (ii) the spinning inertia of the rotor combined
with pitching motion, which induces a gyroscopic yaw
moment.
In both cases of offshore wind turbine concept for
shallow water with fixed foundation and for deep water
with floating foundation, the effect of the wind turbine
is dominant on the design load for the supporting structure. So an integrated load analyses need to be carried
out with comprehensive simulation tools. For offshore
based wind turbine these design codes are aero-servohydro-elastic tools, as they incorporate aerodynamic
models (aero), control system (servo) models, hydrodynamic loads (hydro) and structural dynamic (elastic)
models in a fully coupled simulation environment.
More precisely, these simulation tools incorporate
sophisticated models of both turbulent and deterministic wind inflow; aerodynamic, gravitational, and
inertial loading of the rotor, nacelle, and tower; elastic effects within and between components and in the
foundation; and mechanical actuation and electrical
responses of the generator and of the control and protection systems. The analysis of offshore wind turbines
must also account for the dynamic coupling between
the motions of the support platform and the wind turbine, as well as for the dynamic characterization of the
mooring system for compliant floating platforms.
In order to study the design loads of floating
wind turbine, the FAST code has been developed by
Jonkman (2009) to model the effect of wind turbine.

This code employs a combined modal and multibody dynamics formulation. The code can model most
common wind turbine configurations and control scenarios, including three-bladed turbines with a rigid
hub, two-bladed turbines with a rigid or teetering hub,
turbines with gearboxes or direct drives, turbines with
induction generators or variable-speed controllers, turbines with active blade-pitch regulation or passive stall
regulation, turbines with active or passive nacelleyaw control, and turbines with passive rotor or tail
furling (Jonkman and Bhul 2004a,b, Jonkman 2009).
Jonkman and Sclavounos (2006) developed a fully
coupled aeroelastic and hydrodynamic models for
offshore wind turbines.
In the present study the FAST code will be used to
study the effect of different environmental parameters
in the design loads for 5 MW offshore monopile wind
turbine. The objective is to access the relative importance of various parameters on the design loads on the
support structure.
2
STRUCTURAL DESIGN OF MONOPILE

OFFSHORE WIND TURBINE
The monopile has historically been the most commonly selected foundation type due to its lower cost,
simplicity, and appropriateness for shallow water (less
than 20 m).
The design is a long hollow steel pole that extends
from below the seabed to the base of the turbine. The
monopile generally does not require any preparation
of the seabed and is installed by drilling or driving the
structure into the ocean floor to depths of up to 40
meters. The monopile is relatively simple to manufacture, keeping its cost down despite reaching weights
of over 500 tons and diameters of up to 5.1 m, which
can be heavier than some more complex foundation
designs.
While the monopile is an appropriate foundation
choice for many projects, it can be unsuitable in some
applications. These foundations are not well suited
for soil strata with large boulders. Additionally the
required size of an acceptable monopile increases
disproportionately as turbine size increases and site
conditions become more challenging. Therefore, sites
with deeper water, harsh waves and currents, and larger
turbines may require the implementation of more complex and sturdier designs, such as the jacket, the tripod
and the tripile.
3
LOADS ON MONOPILE FOUNDATION
3.1 Wind loads
Figure 1. Monopile offshore wind turbine.
The wind load applied on the turbine tower comprises of the effects of the direct wind pressure on the
tower and the wind turbine. Therefore in designing the
tower, calculations are made to reflect each characteristic load and suitable safety factor. Loads that occur
548
simultaneously are combined whenever necessary.The

wind turbine loads have two components namely stationary and cyclic. The cyclic loads are aerodynamic
loads from a uniform, steady wind speed and the stationary loads arise from the centrifugal forces. A stationary but spatially uneven flow field over the swept
areas causes cyclic load changes on the turning rotor.
Further, the inertia forces that result from the rotating rotor blade masses cause periodic, non-stationary
loads.
In addition to the stationary and cyclic loads, the
rotor is exposed to non-periodic and random loads
caused by wind turbulence. The variables to be considered are direct wind pressure, gust factor and force
coefficient. The static lateral wind load along the tower
height is calculated by the direct wind pressure on
the projected area that varies with the diameter. The
wind shear force, the overturning moment along the
tower height and the tower deflection along the height
are computed using design formulae. If site specific
wind loads and directions are used, the design of the
tower becomes more economical. Also the optimum
hub height for the location will result in increased
production.
3.2 Wave and current loads
When the waves impinge on structures, their energy is
transferred as loads on the structure. The wave forces
are calculated for slender structures using simplified
linear theory and based on the Morison equation. The
wave loads comprise the inertia component and the
drag component and they depend on the wave height,
wave period and water depth at the location. Since
the wave loads depend on the water depth the shallow water and deep water structures are influenced
by the wave loads. The shape of the structure influences the drag and the inertia coefficients. The wave
loads decrease exponentially towards the sea bottom.
The loads due to the water current are dependent on
the square of the velocity of the current and similar to
wind loads and calculated using drag coefficients of
members.
3.3 Loads from wind rotor
The static and dynamic reaction components from
the rotor on to the wind tower have to be properly
accounted for, as they will produce axial force, shear
force, overturning moment and twisting moment on
the foundation. It is worthwhile to mention here that
special wind rotors that transfer minimum reactions
on to the tower have been developed recently.
3.4 Dynamic behavior
A designer is required to study the dynamic characteristics of the tower with the help of simulation
or modeling to understand dynamic properties of the
tower. It is necessary to understand the extent to which
the flexibility in the foundation plays a role as a design
parameter in influencing the dynamic behavior of the
tower. The dynamic magnification effects can directly

influence the fatigue loads to be considered in the
tower design. It is necessary for the designer to control the tower frequency such that it avoids excitation
of the resonant oscillations that result from the rotor
thrust fluctuations at the blade passing frequency or at
the blade rotational frequency.
Larger and heavier turbines will inevitably experience longer periods of natural oscillation. Offshore
wind turbines are also bigger than onshore turbines
(to take advantage of the steadier offshore winds
and economies of scale). A typical onshore turbine
installed today has a tower height of about 60 to
80 m, and blades about 30 to 40 m long; most offshore
and onshore wind turbines are at the top end of this
range. Offshore turbines installed today are generally
between 2 and 4 MW, with tower heights greater than
60 m and rotor diameters of 75 to 105 m.
4
SIMULATION MODEL
In the present study, a 5 MW wind turbine model developed at NREL is used in the simulation studies. The
turbine design variables are hub height of 90 m above
the mean sea level, and a rotor diameter of 126 m.
The turbine is a variable-speed and collective pitchcontrolled machine, with a maximum rotor speed of
12.1 rpm. The rated wind speed is 11.5 m/s. According
to the NREL design variables the turbine is assumed
to be placed in 20 m of water depth with a monopile
support structure. The FAST code is used to study
the bending moment at the tower base and tower root
for various values of water depth, tower height, pile
diameter and turbulence model.
Bending moment at blade root and bending moment
at tower base are examined for significant wave heights
of 0.5 m, 4.2 m and 9.4 m. The mean wind speed, diameter of pile and turbulence model are fixed respectively
at 12.1 m/s, 6 m and smooth, for three different water
depth conditions and for same control units significant
wave heights of 0.5 m, 4.2 m and 9.4 m is examined
with mean wind speed, diameter of pile and turbulence model are fixed respectively at 12.1 m/s, 6 meters
and smooth, for three different water depth conditions
(10 m, 20 m, 30 m).
4.1 Water depth
Water depth does not affect the blade root as this
depends only on the wind load on the blade. Calculations with the code show indeed that the maximum,
mean and standard deviations in 10 minute simulations, are constant for each water depth and wave
height, which is a verification of the adequacy of the
code.
In Fig. 2, it is observed that the bending moment at
the tower base is maximum when water depth is 30 m
and minimum when water depth is 10 m while wave
height is increasing. The maximum value increases
differently during 10 minutes simulation for 30 m
549
Figure 4. BM at tower base versus wave height.
Figure 5. BM at tower base versus wind speed.

water depth but mean and standard deviation increases

differently.
4.2 Wind speed
The effect of wind speed is to increase the load until
a certain value of speed. For larger values the load
decreases because of the inbuilt control mechanism
that starts changing the pitch of the blades to avoid
that they capture too much wind load and fail.
No effect of water depth at the blade root is observed
and there is no difference for each water depth when
wind speed increases. On the other hand the bending
moment at the tower base is maximum when the water
depth is maximum while wind speed increases and is
minimum when water depth is minimum as in Fig. 3.
4.3
Pile diameter
Bending Moment at tower base is examined under the

same wave height and wind speed conditions for each
pile diameter which are 4 m, 6 m, 8 m.
The bending moment has maximum value at the
tower base for pile diameter of 8 m and minimum for
pile diameter of 4 m but mean and standard deviation
are almost same as shown in Fig. 4. There is no difference at the blade root for each pile diameter when
the wave height increases and the code reproduces this
situation.
In Fig. 5, the bending moment at the tower base
is observed higher for higher pile diameter with the
increase in wind speed. This suggests that for higher
values of pile diameter the monopile wind turbine is

more stable.
4.4 Turbulence model
The bending moment at the tower base and blade
root is examined for each turbulence model. The
Ris Smooth-Terrain model (SMOOTH), a Kaimal
power spectrum (IECKAI) and a Von Karman power
spectrum (IECVKM) are employed to describe the
turbulence random field over the rotor plane.
The bending moment maximum values are similar for IECKAI and IECVKM models but higher
than SMOOTH turbulence model at the tower base as
in Fig. 6. However the standard deviation and mean
values are observed to be almost same.
In Fig. 7(a), the bending moment at the blade root
is observed to be higher for SMOOTH terbulence
model as compared to IECKAI terbulence model and
IECVKM terbulence model while changing the wind
speed.
550
Figure 7a. BM at blade root versus wind speed.
Figure 7b. BM at tower base versus wind speed.


In Fig. 7(b), the bending moment at the tower base

is also observed to be higher for SMOOTH terbulence
model while changing the wind speed.
4.5
4.6 Wind speed
Pile height
The bending moment at tower base is examined under

same wave height and wind speed conditions for each
pile height which are 80 m, 90 m, 100 m. The bending moment is higher for higher pile heights, which
are 100 m, 90 m, 80 m when wave height increases as
in Fig. 8. No significant effect on the blade root is
observed for each pile height when the wave height
increases. It is observed that when the pile height is
higher, the bending moment at tower base is observed
to be higher as in Fig. 9.
The bending moment at tower bases increases
with wind speed until around 12 m/s and they starts
decreasing due to the control mechanism that changes
the pitch of the blades to avoid overloading of the
structure.
The bending moment at tower base is examined for

various values of wind speed. It is observed that at
the tower base the bending moment is higher for wind
speed 12.1 m/s but with the increase in the wind speed
the bending moment at the tower base decreases as
observed in Fig. 9.
In Fig. 10, the bending moment at the tower base
increases with the increase in wave height and the
bending moment is observed to be higher at wind
speed 12.1 m/s.
4.7 Wave height
The bending moment at the tower base is obtained for
various values of wave height. The bending moment at
the tower base changes with the change in wave height
as in Fig. 11. The bending moment at the tower base
is observed to be higher for higher values of the wave
height.
551
It is observed that with the increase in the wind

speed the bending moment at the tower base first
increases within 12.1 m/s but afterwards it decreases.
This is due to the change in the pitch angle of the blade
which reduces the bending moment with the increase
in the wind speed.
CONCLUSION
The design loads for a monopile foundation is studied by analyzing the bending moment at the tower
base and tower root for various values of water depth,
tower height, pile diameter and turbulence model.
FAST code is used to study the bending moment at
the tower base and tower root for a 5 MW offshore
wind turbine. It is observed that the maximum blade
bending moment increases with wind speed, up to the
rated wind speed of 11.5 m/s, and then decreases, as
is expected due to blade-pitch control actions. For
each water depth the bending moment at the tower
base is maximum when water depth is 30 m and minimum when water depth is 10 m. The blade loads are
seen to be relatively insensitive to wave height. It is
an expected result that bending moment at the blade
root is similar. These observations are of significant
importance in the design of monopile offshore wind
turbine.
ACKNOWLEDGEMENTS
This work was performed within the project
FLOATERS-Wave Interaction with Floating Structures in Shallow Water financed by the Portuguese Foundation for Science and Technology
PTDC/ECM/111242/2009. The second author has
been funded by the Portuguese Foundation for Science
and Technology through a post-doctoral fellowship
under the contract no. SFRH/BPD/81010/2011.
REFERENCES
Agarwal, P. and Manuel, L. 2009. Simulation of offshore wind
turbine response for long-term extreme load prediction,
J. of Eng. Struct. 31, pp. 22362246.
Bagbanci, H., Karmakar, D. and Guedes Soares, C. 2012.
Review of offshore floating wind turbines concepts,
Maritime Technology and Engineering, C. Guedes Soares
et al. (Eds.), Taylor & Francis Group, London.
Bir, G. and Jonkman, J. 2008. Modal Dynamics of
Large Wind Turbines with Different Support Structures,
Proceedings of the ASME 27th International Conferences

on Offshore Mechanics and Arctic Engineering, Portugal,
OMAE2008-57446.
Bulder, B. H., et al. 2002. Study to Feasibility of and Boundary Conditions for Floating Offshore Wind Turbines,
Novem 2002-CMC-R43, ECN, MARIN, Lagerway the
Windmaster, TNO, TUD, MSC.
Henderson, A. R., Morgan, C. S., Smith, B., Srensen, H. C.,
Barthelmie, R. J., and Boesmans, B. 2003. Offshore Wind
Energy in Europe, A Review of the State-of-the-Art, Wind
Energy, Vol. 6, No. 1, pp. 3552.
Jonkman, J. M. 2009. Dynamics of offshore floating wind
turbine model development and verification, Wind
Energy, 12, 459492.
Jonkman, J. M. and Buhl, M. L., Jr. 2004a. New Developments for the NWTCs FAST Aeroelastic HAWT Simulator, ASME Wind Energy Symposium, 42nd AIAA
Aerospace Sciences Meeting and Exhibit, Reno Nevada,
USA, New York.
Jonkman, J. M. and Buhl, M. L., Jr. 2004b. FAST Users Guide,
NREL/EL-500-29798, Golden, CO: National Renewable
Energy Laboratory.
Jonkman, J. M. and Sclavounos, P. D. 2006. Development of
Fully Coupled Aeroelastic and Hydrodynamic Models for
Offshore Wind Turbines, 44th AIAA Aerospace Sciences
Meeting and Exhibit, 912 January 2006, Reno, NV.
Lee, K. H., Responses of Floating Wind Turbines to Wind and
Wave Excitation, M.S. Dissertation, Department of Ocean
Engineering, Massachusetts Institute of Technology,
Cambridge, MA, USA, January 2005.
Musial, W. and Butterfield, S. 2004. Future for Offshore Wind
Energy in the United States, Energy Ocean Proceedings,
June 2004, Palm Beach Florida, USA, NREL/CP-50036313, National Renewable Energy Laboratory.
Musial, W., Butterfield, S., and Boone, A. 2004. Feasibility of Floating Platform Systems for Wind Turbines, A
Collection of the 2004 ASME Wind Energy Symposium
Technical Papers Presented at the 42nd AIAA Aerospace
Sciences Meeting and Exhibit, 57 January 2004, Reno
Nevada, USA, New York.
Musial, W., Butterfield, S., and Ram, B. 2006. Energy from
Offshore Wind, 2006 Offshore Technology Conference,
14 May 2006, Houston, TX.
Vijfhuizen, W. J. M. J. 2006. Design of a Wind and Wave
Power Barge, M.S. Dissertation, Department of Naval
Architecture and Mechanical Engineering, Universities of
Glasgow and Strathclyde, Glasgow, Scotland.
Watson, G., et al., 2005. A Framework for Offshore Wind
Energy Development in the United States, Massachusetts
Technology Collaborative (MTC).
Wayman, E. 2006. Coupled Dynamics and Economic Analysis of Floating Wind Turbine Systems, M.S. Dissertation,
Department of Mechanical Engineering, Massachusetts
Institute of Technology, Cambridge, MA, USA.
Wayman, E. N., Sclavounos, P. D., Butterfield, S., Jonkman,
J., and Musial, W. 2006. Coupled Dynamic Modeling
of Floating Wind Turbine Systems, Offshore Technology
Conference, 14 May 2006, Houston, TX.
552

Review of offshore floating wind turbines concepts

ABSTRACT: A literature survey of research and development on offshore floating wind turbines and its
applications is presented. The performance of various floating wind turbine models in different water depths are
outlined to understand the importance of the water depth for floater design. The working principle of various
floater concepts and the conceptual designs for floating platforms used for floating wind turbines are also
described. The detailed research work that has been undertaken in the study of floating wind turbine technology
by various researchers and research institutes are discussed. A brief description on the various components of
the wind turbines, the role of various techniques and loads for design, control systems and economics of wind
energy conversion system are also presented.
INTRODUCTION
In the recent decades, the demand for energy is increasing due to the increase in world population and also
due to industrialization. Most of the people are dependent on fossil fuel as energy source and as a result
there is a negative impact in the society due to the
depletion of fossil fuel. So, efforts are being made to
gradually adopt renewable energy resources that are
going to be exempt from any environmental consequences. The wind energy is observed as an essential
and powerful energy resource for the socio-economic
development and economic growth which helps in
reducing the dependency on fossil fuels and provides
clean energy. It has been estimated that about 10 million MW of energy are available in the earths wind and
it provides security at a time when decreasing global
reserves of fossil fuels threatens the long-term sustainability of global economy. Thus, wind energy emerged
as a promising technology for the utilization of offshore wind resources for the large scale generation of
electricity.
The generation of power from wind can be obtained
from the wind turbines that convert wind energy to
electrical energy. The wind turbines can produce large
amounts of electricity as compared to other energy
sources which are generally placed in (i) onshore and
(ii) offshore.
It has been observed that the relatively low surface
roughness of the ocean results in higher wind speeds
than in onshore. So the offshore turbines are appearing
as an attractive option for generating electricity.
In the year 1885, wind energy was first used for the
production of electrical energy in Askov, Denmark by
Poul la Cour. He converted an old wooden wind mill
into the first wind turbine, which covered the energy
demands of Askov high school. Thus, from 1885, the
use of wind energy for the production of electricity

has progressed but all the wind farms were located
onshore.
The concept of locating wind turbines offshore
came after 1930 and it was suggested that the wind turbines be placed on pylons, but the suggestion was never
used. It was approximately 42 years after the original
idea, the concept for large scale floating wind turbines for the production of electricity was introduced
by Dr. William E Heronemus, professor at University
of Massachusetts in 1972. It was in the year 1990,
a company called World Wind first constructed
and installed the offshore wind turbine at sea. Afterwards, many countries took part in the construction of
offshore floating wind turbines but among them Denmark, Netherlands, Germany, Spain and United States
are currently the world leaders in wind energy technology. These countries have mega wind farms which
comprise a few hundred wind turbines that are spread
over hundreds of square kilometres.
Offshore wind turbines are classified into three
major types depending upon the water depths as
Shallow water foundation

Transitional water foundation
Deep water wind turbine structure
The shallow water wind turbines are generally placed

in between 5 m30 m water depth and are in general
classified into three types as (i) Monopile structure
(ii) Gravity base structure and (iii) Suction bucket
structure.
The transitional offshore wind turbine are placed
between 30 m60 m water depth and are classified
as (i) Tripod tower, (ii) Guyed monopole, (iii) Fullheight jacket, (iv) Submerged jacket with transition
to tube tower and (v) Enhanced suction bucked or
gravity base.
553
wind turbine systems is required. Also, developing

offshore floating wind farms is important because it
can minimize the scenery disturbance, avoid the noise
problems generated by wind-driven blades, provide
high wind speed by low surface roughness, and make
use of extremely abundant deepwater wind resources.
In our present study we will give more emphasis on
the floating offshore wind turbine structures in deep
waters.
2
Figure 1. Offshore wind turbine in shallow water (Musial

et al. 2006).
DEVELOPMENT OF OFFSHORE FLOATING

WIND TURBINE
The interest in offshore wind energy is continuously

increasing and many countries are involved in the
construction of offshore wind parks for production of
large scale electricity. These offshore wind turbines are
environmental friendly and are one of the productive
ways to cover the electrical energy demands.
2.1 Wind power in different countries and
current wind project
Figure 2. Transitional offshore wind turbine (Musial et al.

2006).
The deep water offshore wind turbines are generally floating structures and are placed in more
than 60 m water depth. The floating wind turbines
in deep water fall into four main categories (i) Sparbuoy type, (ii) Tension Leg Platform (TLP) type, (iii)
Semi-submersible type and (iv) Pontoon type.
Recently, most of the offshore wind power projects
are proposed in deep water where the winds are of
sufficient velocities. Thus the wind turbines on the
floating support are the best solution to utilize the wind
resources in areas with deep water depths.
In order to improve the wind energy production and
to get large scale generation of electricity, wind turbine
technology needs offshore wind energy resources. So
far, most projects of offshore wind farms are located
in relatively shallow water using bottom-fixed type
wind turbines. To extend wind turbine systems to
deeper water, practical research of offshore floating
In the last few years, a number of offshore wind

farms have been put into operation in European countries such as Denmark, United Kingdom (UK) and
Netherlands. They are all situated in shallow waters,
having a water depth of less than 25 meters and are relatively close to shore. For these developments, it proved
economical to use either simple concrete gravity
structures or steel monopiles as substructures
(Musial and Butterfield, 2006).
A list of the offshore wind projects built in the
last few years can be found at the Wind Service Holland (2008). In terms of installed power, the main
projects were the following: in the UK, the Lynn and
Inner Dowsing (194 MW), the Kentish Flats project
(90 MW) and the Burbo Banks project (90 MW); in
the Netherlands, the Q7 project (120 MW); and in
Denmark, the Nysted offshore Windfarm (165 MW)
and the Horns Rev project (160 MW).
Projects are currently in development for deeper
sites. The cost of the support structure and foundation
will be a proportionally higher part of the total cost
than for turbines in shallow waters. This means that
finding an economically feasible design is vital for
overall project viability. This necessitates the development of other substructure and foundation types than
those previously used for shallow waters; an example can be found in Klose et al. (2007), where the
development of a jacket substructure in 45 meters of
water is described. The most recent advance in wind
farm development is the investigation of floating wind
turbines in very deep waters. Significant efforts have
been made to implement this modified turbine structure, especially in Norway (Skaare et al. 2006, 2007),
the United States of America (Wayman et al. 2006,
Jonkman et al. 2007) and Japan (Suzuki and Sato,
2006)). A floating wind turbine prototype has been
put into real sea test in Norway in 2009 (Nielsen et al.
554
Figure 3. The trend of the wind turbine sizes and capacity

(US Dept of Energy).
Figure 4. The structure of nacelle and rotor (US Department

of Energy).
(2009). A detailed literature survey on floating wind

turbine is presented in Wang et al. (2010).
Recently, In Portugal a 2 MW prototype offshore
wind turbine is being installed in the north of the
country placed on the floating platform WindFloat
developed by Principle Power. In the next phase a
additional 5 MW turbine will follow. The project is
undertaken by EDP, Portugal and it is proposed to
achieve a total capacity of 150 MW.
2.2 Wind turbine size and its development
In the early and mid-1980s, the typical wind turbine
size was less than 100 kW. By the late 1980s and early
1990s, turbine sizes had increased from 100 to 500 kW.
Further, in the mid-1990s, the typical size ranged from
750 to 1000 kW. By the late 1990s, the turbine size had
gone up to 2.5 MW. Now turbines are available with
capacities above 5 MW (as in Fig. 3) (Herbert et al.
2007).
So, nowadays, the research is on a full swing to
develop floating wind turbines of high power. In the
next section, the various components of floating wind
turbines will be discussed.
2.3
Components of offshore floating wind turbine
The wind turbine has seven major subsystems such

as blades, nacelle, controller, generator, rotor, tower,
and floating body. The detailed descriptions of each of
these components are as follows.
2.3.1 Blades
The generation of power increases with the increase
in the number of blades. Most of the wind turbines
have three blades, though there are some with two
blades. Blades are generally 30 m to 50 m long, with
the most common sizes around 40 m. Blade weights
vary, depending on the design and materials. A 40 m
LM glass fibre blade for a 1.5 MW turbine weighs 6.4
tons and one for a 2.0 MW turbine weighs 6.9 tons.
2.3.2 Nacelles
The nacelle houses the main components of the wind
turbine, such as the controller, gearbox, generator, and
shafts. This part protects the wind turbine equipment.
2.3.3 Controller
The controller monitors the condition of the turbine
and controls the turbine movement. The control system changes the blade pitch, nacelle yaw, and generator
loading of a wind turbine. The control system can also
change the pitch of the blades to alter the amount
of torque produced by the rotor. The purpose of the
control system is to maximize power output.
2.3.4 Gearbox
The gearbox present in the turbine helps in increasing
the rotational speed of the shaft. A low-speed shaft
feeds into the gearbox and a high-speed shaft feeds
from the gearbox into the generator. Some turbines use
direct drive generators that are capable of producing
electricity at a lower rotational speed.
2.3.5 Generators
Wind turbines typically have a single AC generator that converts the mechanical energy from the
wind turbines rotation into electrical energy. Clipper
Wind power uses a different design that features four
DC generators. Offshore wind turbines typically send
power through cables.
2.3.6 Rotor
The rotor includes both the blades and the hub. The
rotor consists of normally two or three blades attached
to a hub. The system performance of the wind turbine
is based on the selection of blade number, shape, and
length. The rotor can be either upwind or downwind
design. Most wind turbines are three bladed upwind
designs.
2.3.7 Towers
Towers are usually tubular steel towers 60 m to 80 m
high that consist of three sections of varying heights.
555
There are some towers with heights around 100 m. The

tower supports the wind turbine nacelle and rotor.
2.3.8 Floating body
The floating body supports all wind turbine elements
in the ocean. The floating body is kept in place by the
mooring systems and has enough buoyancy to support
the structure. In the next section, various types of
floating wind turbines and their importance in deep
water will be discussed.
FIXED OFFSHORE WIND TURBINES
3.1 Wind turbine in shallow water depth

Offshore wind energy began in shallow waters of the
North Sea where the abundance of sites and higher
wind resources are more favourable by comparison
with Europes land-based alternatives. The first installation was in Sweden with a single 300-kW turbine in
1990 and the industry has grown slowly over the past
15 years. There are now 18 operating projects with an
installed capacity of 804 MW. Fig. 3 shows a breakdown of where the installed capacity is located as a
percentage of the total capacity and by country. The
majority of the capacity is now located in Denmark
and the United Kingdom, using mostly Danish turbine
technology.
3.1.1 Monopile structure
The monopile foundation is a simple construction.
The foundation consists of a steel pile with a diameter of between 3.5 and 4.5 meters. The pile is driven
some 10 to 20 meters into the seabed depending on
the type of ground. The monopile foundation is effectively extending the turbine tower under water and into
the seabed. Benefits of this foundation are simplicity
and there are no problems with erosion. Monopiles
do not need any seabed preparations but on the other
hand they require heavy piling equipment. Monopiles
are easiest to transform from onshore to offshore with
minimal design and they leave a minimal footprint to
the seabed. However monopiles are somewhat flexible
and that is why they have depth limits where they can
be used. This foundation is not suitable for locations
with many large boulders in the seabed. Agarwal and
Manuel (2009) performed the simulation of offshore
wind turbine response for ling term extreme loads prediction. Recently, the detailed study on the effect of
environment on the design loads on monopole wind
turbine can be found in Bagbanci et al. (2012).
3.1.2 Gravity base
The gravity base is an alternative foundation to a
monopile. As can be seen from its name it relies on
gravity to keep the turbine in place. This is the foundation solution that most of the existing offshore wind
parks use. The gravity base foundations require well
prepared seabed to ensure horizontal smooth ground
for foundation. These gravity based foundations can be
manufactured with reinforced concrete or steel. Newer

technology is to use steel in similar ways as concrete.
Studies in Denmark have shown that steel is better foundation solution than concrete in large offshore
wind farms which base on gravity. Using steel in gravity based foundations a cylindrical steel tube placed on
a flat steel box on the sea bed. The steel foundations
are filled with olivine, which gives the foundations
sufficient weight to withstand waves and ice pressure.
A steel gravity foundation is considerably lighter than
concrete foundations but it gives as good rigidity and
stability as concrete.
Cases where reinforced concrete is used, pre-casted
concrete is floated out to sea to its final location
and then filled with sand to achieve the necessary
weight. This way to build foundation for wind turbine
reminds bridge building. The low weight allows barges
to transport and install many foundations rapidly. This
gravity base alternative has not flair for flexibility like
monopile but costs increase rabidly as well when water
depth grows. In both cases erosion protection is taking
care by placing rocks around the edge of the base.
3.1.3 Suction bucket
A suction bucket foundation consists of an upside
down cylinder that is pressed into the subsoil. The
bucket penetrates into the seabed partly by self-weight
and partly by applied suction. Suction buckets are
tubular steel foundations that are installed by sealing
the top and applying suction inside the bucket. The
hydrostatic pressure difference and the deadweight
cause the bucket to penetrate the soil. This benign
installation procedure allows the buckets to be connected to the rest of the structure before installation,
enabling a reduction in steps of the installation procedure. The system has been tried in practice in the
Norwegian oil and gas fields in the North Sea and in
Angola. Because of the large hydrostatic force that is
required for installation, suction buckets commonly
have a much lower aspect ratio than driven piles. A
diameter to length ratio of 10 is a practical maximum,
which depends on water depth and soil properties.
4
FLOATING WIND TURBINE CONCEPTS
Currently, there are a number of offshore wind turbine floating foundation concepts in various stages of
development. The main concern is to study the floating
wind turbine in deep water depth where the generation
of power of each unit can be increased.
4.1 Wind turbine in deep water depth
In general terms, the spar-buoy type have better heave
performance than semisubmersibles due to their deep
draft and reduced vertical wave exciting forces, but
have more pitch and roll motions, since the water plane
area contribution to stability is reduced. TLPs have
very good heave and angular motions, but the complexity and cost of the mooring installation, the change
556
Table 1.
Spar-type floating wind turbine properties.
Power rating
Turbine weight
Turbine height
Rotor diameter
Draft hull
Displacement
Diameter of water line
Diameter of submerged body
Water depths
Mooring
2.3 MW
138 tons
65 m
82.4 m
100 m
5300 m3
6m
8.3 m
120700 m
3 lines
Source: Hywind pilot project, Norway
Figure 5. Spar-buoy type floating wind turbine (Roddier

et al. 2009).
in tendon tension due to tidal variations, and the structural frequency coupling between the mast and the
mooring system, are three major hurdles for such systems. Semi-submersible concepts with a shallow draft
and good stability in operational and transit conditions
are significantly cheaper to tow out, install and commission than spar-buoy, due to their draft. The various
floating turbines will be discussed in some detail.
4.1.1 Spar-type floating wind turbine
The spar-type floating wind turbine comprises the
floating foundation which is referred as the floater,
the tower and the rotor-nacelle assembly (RNA). The
floater is towed in the horizontal position to the deployment site in calm waters. It is then upended, stabilised,
and the tower and the RNA mounted by a derrick crane
barge before finally being towed by escort tugs in the
vertical position to the deployment site for connection
to the mooring system.
The floating foundation consists of a steel and/or
concrete cylinder filled with a ballast of water and
gravel to keep the centre of gravity well below the centre of buoyancy which ensures the wind turbine floats
in the sea and stays upright since it creates a large righting moment arm and high inertial resistance to pitch
and roll motions. The floater is ballasted by permanent
solid iron ore ballast, concrete or gravel from a chute.
Alternatively, the ballast tanks may be injected with
grout. It should be remarked that the spar-buoy type
is difficult to capsize. The draft of the floating foundation is usually larger than or at least equal to the
hub height above the mean sea level for stability and
to minimise heave motion. Therefore, it is necessary
to have deep water for deployment of this spar-buoy
type floating wind turbine as adequate keel to seabed
vertical clearance is required for the mooring system
to be effective.
The spar-type floating wind turbine is usually kept
in position by a taut or a catenary spread mooring
system using anchor-chains, steel cables and/or synthetic fibre ropes. Alternatively, it may be moored by
a single vertical tendon held at the base by a swivel
connection that allows the wind turbine to revolve as
the wind changes direction (as proposed by the company SWAY). This free yawing effect is similar to the
swinging mechanism found in a Floating Production
Storage and Offloading vessel (FPSO) turret mooring
in the offshore oil and gas industry.
Although favourable because the wind turbine will
always face the direction of oncoming wind thus optimising power generation, the single vertical tendon
may not provide for a degree of redundancy in the
event of failure, resulting in unrestrained drifting of
the floater. The first full scale size spar floating turbine
has been deployed off the south-west coast of Karmoy
Island, Norway by Statoil in the Hywind demonstration project. The specification of a 2.3 MW spar-buoy
floating wind turbine is given in Table 1.
Tong (1998) analyzed the technical and economic
aspects of a wind farm. The conceptual design for
FLOAT, which is a spar-buoy type floating wind turbine was presented. Nielson et al. (2006) discussed the
integrated dynamic analysis of spar-buoy type floating wind turbines. They developed simulation models
for Hywind and compared their numerical results with
model scale test results.
Skaare et al. (2007) presented the importance
of control strategies on the fatigue life of floating
wind turbines. They considered various environmental conditions and wind turbine control schemes. They
showed the importance of the effect of pitch-angle
control of blades on the dynamic response of the floating wind turbine for wind speeds above the rated
wind speed. Suzuki and Sato (2007) investigated the
loads on turbine blades induced by the motion of the
floating platform and design requirement for the platform. They analyzed the effect of a stabilizing the
fin attached to the base of the floating foundation in
reducing the pitch motion of the floating spar wind
turbine.
Matsukuma and Utsunomiya (2008) performed a
motion analysis of a spar-buoy type floating wind turbine under steady wind considering rotor rotation. The
wind loads acting on the rotor blades are calculated
using the blade element momentum theory. As a result,
557
Table 2. TLP floating wind turbine properties.

Power rating
Water depth
Radius
Cylinder height
Concrete ballast height
Turbine mass
5 MW
108 m
11 m
21.5 m
4.5 m
698 ton
Source: Blue H, Netherlands (Wayman et al. 2006)
Figure 6. TLP type floating wind turbine (Roddier et al.

2009).
the motion of yaw, sway and roll are generated due to

the effect of the gyro moment for the rotor-rotation.
Utsunomiya et al. (2009) continued the experimental validation for motion of a spar-buoy type floating
offshore wind turbine. In this case the motion of a
prototype spar wind turbine was determined under regular and irregular waves and a steady horizontal force
that simulates the steady wind condition was analyzed.
Karimirad and Moan (2010) carried out structural
dynamic response analyses of a spar-type wind turbine
in the extreme survival condition. Numerical simulation for coupled wave and wind induced motion and
structural response in harsh conditions for a parked
floating wind turbine were undertaken. Recently,
Bagbanci et al. (2011a) studied the dynamic analysis of
spar type floating wind turbine and a detailed analysis
on the comparison of spar-type floating wind turbine
and barge-type floating wind turbine is performed in
Bagbanci et al. (2011b).
4.1.2 Tension leg platform (TLP) type floating
wind turbine
The TLP type comprises a floating platform structure
to carrying the wind turbine as in Fig. 6. In the offshore
oil and gas industry, the conventional TLP platform
comprises a square pontoon with columns on which
the topside deck rests. A smaller version of this conventional hull form is the mini-TLP which has been
adopted by the TLP-type floating wind turbine.
Unlike the spar-buoy type floating wind turbine,
which needs to be assembled in at sea, this TLP wind
turbine may be assembled and commissioned onshore
thereby avoiding the logistical difficulties of offshore
assembly. The fully fitted up platform is towed to the
deployment site thus precluding the need to charter
and mobilize expensive heavy-lift vessels or derrick
crane barges for offshore construction.
The floating platform is held in position by vertical tendons (also called tethers) which are anchored
either by a template foundation, suction caissons or by

pile driven anchors. The pre-tensioned tethers provide
the righting stability. This type of floating wind turbine
has a relatively small dynamic response to waves when
compared to the spar-buoy type, the semi-submersible
type or the pontoon type but it is subject to a phenomenon known as pull down which is an increase
in draft as the platform is offset from its equilibrium
position.
A TLP wind turbine has since been installed off the
coast of Puglia, southern Italy by Blue H Technologies.
This large scale prototype is used to test the assembly,
transportation and installation of the TLP type wind
energy converter as well as to serve as a metering platform with sensors to measure site specific data. The
specification of a 5 MW TLP floating wind turbine is
given in Table 2.
Withee and Sclavounos (2004) studied the fully
coupled dynamic analysis of a floating wind turbine
system. They performed fully coupled time domain
simulations of the system responses for a 1.5 MW
wind turbine mounted on a TLP floater under wind
and wave forces. They observed that the two damping
mechanisms were comparable in magnitude and that
the damping arising from the turbine rotor appears to
obey a distinctly linear law.
Lee (2004) analyzed the responses of floating wind
turbines to wind and wave excitation. He carried out
frequency domain response analysis of both TLP and
Spar-buoy type of floating wind turbines in order to
compare the performances of the two floater concepts.
Suzuki et al. (2009) developed a conceptual design of
a TLP-type floating structure for offshore wind farms.
The design is based on their past experience in the
design and fabrication of TLP which takes into consideration the stability of the structure, mooring forces
and ease in maintenance. Weinzettel et al. (2009) performed a life cycle assessment of a floating offshore
wind turbine based on the Sway concept TLP type
floater. The preliminary life cycle assessment considered the environmental impact of the floating wind
turbine and highlighted the importance of the decommissioning scenario whereby materials are recycled to
mitigate detrimental effects on the environment.
Bae et al. (2010) performed a rotor-floater-tether
coupled dynamic analysis on a mini TLP-type offshore
floating wind turbine. The dynamic coupling between
the rotating blades and the floater, in addition to the
mooring-floater dynamic coupling was considered and
assessed.
558
Table 3.
Semi-submersible wind turbine properties.
Power rating
Rotor diameter
Turbine hub height
Turbine nacelle weight
Tower weight
Hull weight
Total displacement
Hull draft
Operational water depth
Conventional mooring component
3.610 MW
120150 m
8090 m
225315 ton
180315 ton
10801630 ton
<5500 ton
<20 m
>50 m
46 lines
Source: Principal Power, USA
Figure 7. Semi-submersible type floating wind turbine

(Roddier et al. 2009).
Nihei and Fujioka (2010) investigated the motion

characteristics of a TLP type offshore wind turbine in
waves and wind Their experiments indicated that in
the case of applied waves and wind, the wind has the
beneficial effect of stabilizing the floater pitch motion
and decreasing the vibration of the mooring lines.
4.1.3 Semi-submersible type floating wind turbine
The semi-submersible type comprises a few large
column tubes connected to each other by tubular members. A wind turbine may sit on one of the column
tubes or there could be wind turbines sitting on all
the columns. Alternatively, the wind turbine may be
positioned at the geometric centre of the column tubes
and supported by lateral bracing members.The column
tubes provide the ballast and they are partially filled
with water. When in the afloat condition, the waterplane area of the columns primarily provides floatation
stability, hence the term column stabilized.
This design is good in providing stability to the wind
turbine and its relatively shallow draft allows for site
flexibility. The semi-submersible floating wind turbine
is kept in position by mooring lines. This type of floating wind turbine may be constructed onshore. Until
now, there is no semi-submersible floating wind turbine in operation. Principle Power Inc. is promoting
the semi-submersible type which consists of three column tubes with patented horizontal water entrapment
heave plates at the bases as in Fig. 7.
The heave plates primarily serve to reduce heave
and pitch motion without increasing floating platform
size. The specification of a semi-submersible-type
floating wind turbine is given in Table 3.
Henderson and Patel (1998) presented analytical and numerical design tools for evaluating the
performance of semi-submersible floating wind turbines. Zambrano et al. (2006) presented the dynamic
modelling of deep water offshore structure fitted with

three wind turbines in the Gulf of Mexico storm
conditions using Fourier spectrum based model. The
WAMIT program was used to develop the wave forces
on the platform.
Shimada et al. (2007) and Ishihara et al. (2007a,b,
2008) studied the dynamic response of a semisubmersible type under resonance condition. In the
numerical computations, the nonlinear damping effect
due to hydrodynamic drag forces (modelled by the
Morisons formula) and inertia forces (by using Srinivasans model) are considered in the equation of
motion, where the linear hydrodynamic forces are
obtained from the Greens function model. Ishihara
et al. (2009) analyzed the influence of heave plate on
the dynamic response of floating offshore wind turbines in water tank experiment. It is observed that
the heave plates help to increase the natural period
of heave, thereby resulting in a reduction of heave
response at rated and extreme sea states.
Roddier et al. (2009), Cermelli et al. (2009) and
Aubault et al. (2009) conducted a feasibility study for
the WINDFLOAT technology for a semi-submersible
floating wind turbine. Roddier et al. (2009) focused on
the design basis for wind turbine floating foundations.
The authors pointed out that the design of the hull
for a large wind turbine must draw on the synergies
with oil and gas offshore platform technology, while
accounting for the different design requirements and
functionality of the wind turbine. Then, Cermelli et al.
(2009) carried out hydrodynamic analysis of the hull of
the floating foundation.Aubault et al. (2009) discussed
the structural assessment of a column-stabilised floating foundation. They focused on the methodology for
estimating the strength and fatigue of the structural
components.
4.1.4 Pontoon type floating wind turbine
The pontoon type floating wind turbine has a very
large pontoon structure to carry a group of wind turbines. The large pontoon structure achieves stability
via distributed buoyancy and by taking advantage of
the weighted water plane area for righting moment.
The pontoon type may be moored by conventional
catenary anchor chains as in Fig. 8.
However, the setback of the pontoon-type wind turbine is that it is susceptible to the roll and pitch motions
559
Table 4.
Pontoon type floating wind turbine properties.
Power rating
Rotor orientation
Rotor diameter
Hub height
Cut-In, Rated, Cut-Out wind speed
Cut-In, Rated rotor speed
Rated Tip Speed
Rotor mass
Nacelle mass
Tower mass
5 MW
Upwind
126 m
90 m
3 m/s, 11.4 m/s, 25 m/s
6.9 rpm, 12.1 rpm
80 m/s
121.25 ton
264.55 ton
383 ton
Source: Jonkman and Bhul (2007)
Figure 8. Pontoon type floating wind turbine (Wang et al.

2010).
in waves experienced by ocean-going ship shaped vessels and may only be sited in calm seas, like in a
harbour, sheltered cove or lagoon. The National Maritime Research Institute (NMRI) in Tokyo has made
some studies on such pontoon-type floating wind turbines. It should be remarked that there are clearly
hybrid types of floating wind turbines, for example
a combination of spar floater and tension leg mooring system. Also, there is an interesting concept of
a sailing-type floating wind turbine that was studied
at the National Institute for Environmental Studies,
Japan.
The floating wind power plant has no mooring system but navigates with sails and azimuth thrusters.
The self-sailing and self propelled mobility allows the
wind farm to move to a location that maximizes the
generation of wind power as well as to weather route
from storms. The slender lower and upper hulls are
lightweight composites to minimize construction costs
and enhance the propulsive performance.
Based on the buoyancy stabilized concept, NREL
and MIT collaborated in a pontoon-type (or bargetype) floating wind turbine. The pontoon-type is
adopted because of its simplicity in design, fabrication and installation. Jonkman and Buhl (2007)
presented fully coupled aero hydro servo elastic simulation tools for a preliminary loads analysis of a 5 MW
pontoon-type wind turbine. The analysis aims to characterize the dynamic response and to identify potential
loads and instabilities that would be detrimental to
such a pontoon-type design. They found that coupling
between the turbine response and the pontoon pitch
motion produces larger extreme loads in the floating
turbine tower and blades. The pontoon was found to
be susceptible to excessive pitching in extreme wave
conditions. The compliance of the floating pontoon
wind turbine did, however, reduce a tower side-to-side
instability that occurred in land-based turbines. They
suggested some design modifications to reduce the
pontoon motions and to eliminate the instabilities.
Wayman et al. (2006) presented the coupled

dynamic modelling of floating wind turbine system
in the frequency domain. The floating wind turbine is
kept in place by mooring lines. Iijima et al. (2010)
described their numerical procedure for the fully
coupled aerodynamic and hydroelastic time-domain
analysis of an offshore floating wind turbine system including rotor blade dynamics, dynamic motions
and flexible deflections of the structural system. The
dynamic coupling between the rotating blades and the
structural system under wind and wave loads was taken
into account.
4.2 Method of analysis
In order to analyze the practical models of the floating wind turbine, WAMIT (Wave Analysis at MIT)
developed by WAMIT Inc. and Charm3D (Coupled
hull/mooring/riser analysis) developed at Texas A&M
University can be employed. For the prediction of loads
and responses of floating wind turbine, the primary
design code of wind turbines, FAST (Fatigue, Aerodynamics, Structures, and Turbulence) promoted by
National Renewable Energy Laboratory (NREL), can
also be used (Jonkman and Buhl (2004a,b), Jonkman
(2009, 2010)).
The wind turbine is modelled using the FAST wind
turbine code in conjunction with ADAMS. ADAMS
(Automatic Dynamic Analysis of Mechanical Systems) is commercially available software that is widely
used to model mechanical systems. FAST codes modelling the floater loading and mechanical properties
are coupled to the ADAMS wind turbine model to
determine the effect on the wind turbine components.
The FAST code is a fully coupled dynamic analysis
simulator that could be used for floating wind turbine
concepts. FAST code uses two different subroutines
such as AeroDyn and HydroDyn and they seek wind
data and hydro data for the analysis. Thus a hydrodynamic simulator and a wind simulator are required.
AeroDyn calculates wind load along blades using
blade element theory. TurbSim (Turbulence modeling scaling) generates wind data for FAST code and
HydroDyn calculates wave load on floating system.
To obtain the hydrodynamic loads on the floaters
WAMIT (Wave Analysis at MIT) or similar hydrodynamic simulator are also used for FAST code. The
560
The wind power projects that were installed in

various countries are discussed.
The performances of various floating wind turbine
models in different water depths are outlined to
understand the importance of the water depth for
power generation.
The working principle of various floater concepts,
the conceptual designs for floating platforms used
for floating wind turbines are described.
The energy production capacity and the importance
of offshore floating wind turbines are discussed in
detail.
ACKNOWLEDGEMENTS
Figure 9. Simulation of FAST code.
FAST code simulates stochastic time-domain turbine

response and computes hydrodynamic loads using
Morisons equation.
A floating wind turbine is split into two bodies such
as floater and wind turbine. Therefore, there is an interactive dynamic relation between floating platform and
wind turbine and thus it is called multi-body dynamics. FAST code uses Kanes Method for multi-body
problem between floating platform and wind turbine.
Multi-body motion has in total 22 degrees of freedom
(DOF) for a two-bladed horizontal axis floating wind
turbine model and 24 degrees of freedom (DOF) for
a three-bladed horizontal axis floating wind turbine
model. FAST code can simulate two or three bladed
floating wind turbine.
Multi body motions in degrees of freedom are platform translation and rotation (6 DOF), tower flexibility
(4 DOF), nacelle yaw (1 DOF), variable generator and
rotor speeds (2 DOF), rotor furl (1 DOF), and tail furl
(1 DOF) for both bladed kind horizontal axis floating
wind turbine. The main difference is blade flexibility (6 DOF) in two bladed turbine and (9 DOF) in
three bladed turbine. The blade teetering (1 DOF) in
two bladed wind turbine is considered whereas it is
ignored in three bladed turbine. The detailed FAST
code simulation diagram is shown in Fig. 9.
CONCLUSION
The floating wind turbine is one of the fastest growing technologies in the recent years and may become
a major generator of offshore renewable electricity
worldwide. Floating wind farms located far out in
the deep sea draw more wind power than land-based
wind farms. This paper has presented a brief literature
review on the work pursued on offshore floating wind
turbine. The following conclusions are drawn based on
the study carried out in this paper:
Various components associated with the floating

wind turbine and their importance is described.
This work was performed within the project

FLOATERS-Wave Interaction with Floating Structures in Shallow Water financed by the Portuguese
para a Cincia e Tecnologia) under contract
PTDC/ECM/111242/2009. The second author has
been funded by the Portuguese Foundation for Science
and Technology through a post-doctoral fellowship
under the contract no. SFRH/BPD/81010/2011.
REFERENCES
Agarwal, P. and Manuel, L. 2009. Simulation of offshore wind
turbine response for long-term extreme load prediction,
J. of Eng. Struct. 31, pp. 22362246.
Aubault, A., Cermelli, C., and Roddier, D., 2009. WIND
FLOAT: a floating foundation for offshore wind turbines
Part III: structural analysis. Proc. 28th Intl Conf. on
Ocean, Offshore and Arctic Eng., Honolulu, Hawaii, USA,
OMAE-79232.
Bae, Y.H., Kim, M.H., and Shin, Y.S., 2010. Rotor-floater
mooring coupled dynamic analysis of mini TLP-type offshore floating wind turbines. Proc. 29th Intl. Conf. on
Ocean, Offshore and Arctic Eng., OMAE-20555.
Bagbanci, H., Karmakar, D. and Guedes Soares, C. 2011a.
Dynamic analysis of spar-type floating offshore wind
turbine. 2nd Coastal and Maritime Mediterranean Conference CM2-2011, Tangier, Morocco, pp. 407412.
Bagbanci, H., Karmakar, D. and Guedes Soares, C. 2011b.
Comparative study on the coupled dynamic analysis
of spar-type and barge-type floating wind turbine. 1st
Intl. Conference on Naval Architecture and Marine and
Engineering, INT-NAM 2011, Turkey, Istanbul.
Bagbanci, H., Karmakar, D. and Guedes Soares, C. 2012.
Effect of environment on the design loads on monopole
offshore wind turbine, Maritime Technology and Engineering, C. Guedes Soares et al. (Eds.), Taylor & Francis
Group, London.
Cermelli, C., Roddier, D., and Aubault, A., 2009. WINDFLOAT: a floating foundation for offshore wind turbines
Part II: Hydrodynamics analysis. Proc. 28th Intl Conf. on
Ocean, Offshore and Arctic Eng., Honolulu, Hawaii, USA,
Paper OMAE2009-79231.
Herbert G.M.J. Joselin Herbert, Iniyan, S., Sreevalsan,E.,
Rajapandian, S. 2007. A review of wind energy Technologies, Renewable and Sustainable Energy Reviews, 11,
11171145.
Henderson, A.R. and Patel, M.H., 1998. Floating offshore
wind energy. Proc. of 17th Intl Conf. on Ocean, Offshore
and Arctic Eng., Lisbon, Portugal, July 59, 1998.
561
Iijima, K., Kim, J., and Fujikubo, M., 2010. Coupled aerodynamic and hydroelastic analysis of an offshore floating
wind turbine system under wind and wave loads. Proc.
29th Intl. Conf. on Ocean, Offshore and Arctic Eng., Paper
OMAE2010-20772.
Ishihara, T., Phuc, P.V., and Sukegawa, H., 2007a. Numerical study on the dynamic response of a floating offshore
wind turbine system due to resonance and nonlinear wave.
Proc. 2nd EOW, Berlin, Germany.
Ishihara, T., Phuc, P.V., and Sukegawa, H., 2007b. A study on
the dynamic response of a semi-submersible floating offshore wind turbine system Part 1: a water tank test. Proc.
12th Intl. Conf. on Wind Engineering, Cairns, Australia.
Ishihara, T., Waris, M.B., and Sukegawa, H., 2009. A study
on influence of heave plate on dynamic response of floating offshore wind turbine system. Proc. 3rd European
Offshore Wind Conference and Exhibition, Stockholm,
Sweden.
Jonkman, J.M. (2009). Dynamics of offshore floating
wind turbine model development and verification, Wind
Energy, 12, 459492.
Jonkman, J.M. (2010). Defination of the floating system for
phase IV of OC3, NREL technical report, 125.
Jonkman, J.M. and Buhl, M.L., Jr. (2004a). New Developments for the NWTCs FAST Aeroelastic HAWT
Simulator, ASME Wind Energy Symposium, 42nd AIAA
Aerospace Sciences Meeting and Exhibit, Reno Nevada,
USA, New York.
Jonkman, J.M. and Buhl, M.L.,Jr. (2004b). FAST Users
Guide, NREL/EL-500-29798, Golden, CO: National
Renewable Energy Laboratory.
Jonkman, J.M. and Buhl, M.L. Jr, (2007). Loads analysis
of a floating offshore wind turbine using fully coupled
simulation. Proc. of Wind Power 2007 Conference and
Exhibition, Los Angeles, California.
Jonkman, J. M. and Sclavounos, P. D. (2006). Development of
Fully Coupled Aeroelastic and Hydrodynamic Models for
Offshore Wind Turbines, 44th AIAA Aerospace Sciences
Meeting and Exhibit, 912 January 2006, Reno, NV.
Karimirad, M. and Moan, T., 2010. Extreme structural
response of a spar type wind turbine. Proc. 29th Intl.
Conf. on Ocean, Offshore and Arctic Engineering, OMAE20044.
Klose, M., Dalhoff, P. and Argyriadis, K. 2007. Integrated
load and strength analysis for offshore wind turbines
with jacket structures. Proc. 17th Intl. Offshore and Polar
Engineering Conf., Lisbon, Portugal.
Lee, K.H., 2004. Responses of floating wind turbines to
wind and wave excitation. Thesis (MSc). Massachusetts
Institute of Technology.
Matsukuma, H. and Utsunomiya, T., 2008. Motion analysis of
a floating offshore wind turbine considering rotor rotation
The IES Journal Part A: Civil and Structural Eng., 1 (4),
268279.
Musial, W., Butterfield, S. and Ram, B., 2006. Energy from
offshore wind. Offshore Technology Conf., OTC 18355
(OER), Houston.
Nielsen, F.G., Hanson, T.D., and Skaare, B., 2006. Integrated
dynamic analysis of floating offshore wind turbines. Proc.
25th Intl. Conf. Offshore Mechanics and Arctic Eng.
Hamburg, Germany, OMAE2006-92291.
Nielsen, F.G., Argyriadis, K., Fonseca, N., Le Boulluec,
M., Liu, P., Suzuki, H., Sirkar, J., Tarp-Johansen, N.J.,
Turnock, S.R., Waegter, J. and Zong, Z. 2009. Ocean, Wind

and wave energy utilization. 17th Intl. Ship and Offshore
Structures Congress, Seoul, Korea, 201257.
Nihei, Y. and Fujioka, H., 2010. Motion characteristics of
a TLP type offshore wind turbine in waves and wind.
Proc. 29th Intl. Conf. on Ocean, Offshore and Arctic Eng.,
OMAE-21126.
Roddier, D., Cermelli, C., and Weinstein, A., 2009. WINDFLOAT: a floating foundation for offshore wind turbines
Part I: Design basis and qualification process. Proc. 28th
Intl. Conf. on Ocean, Offshore and Arctic Eng., Honolulu,
Hawaii, USA, OMAE2009-79229.
Shimada, K., Ohyama, T., Miyakawa, M., Ishihara, T.,
Phuc, P.V. and Sukegawa, H. 2007. A study on a semisubmersible floating offshore wind energy conversion
system. Proc. 17th Intl. Offshore and Polar Eng. Conf.,
Lisbon, Portugal, 384354.
Skaare, B., Hanson, T., Nielsen, F.G. 2006. Integrated
Dynamic Analysis of Floating Offshore Wind Turbines.
Proc. European Wind Energy Conf. and Exhibition,
Athens, Greece.
Skaare, B., Hanson, T., Nielsen, F.G. 2007. Importance of
control strategies on fatigue life of floating wind turbines.
Proc. 26th Intl. Conf. on Offshore Mechanics and Arctic
Eng.. OMAE2007-29277.
Suzuki, H. and Sato, A., 2007. Load on turbine blade induced
by motion of floating platform and design requirement for
the platform. Proc. 26th Intl. Conf. on Offshore Mechanics
and Arctic Eng., OMAE2007-29500.
Suzuki, H., Kurimoto, M., Kitahara, Y. and Yukinari
Fukumoto, Y. 2009. Progressive drifting of floating wind
turbines in a wind farm. Proc. 28th Intl. Conf. on Ocean,
Offshore and Arctic Eng., Honolulu, Hawaii, OMAE200979634.
Tong, K.C., 1998. Technical and economic aspects of a floating offshore wind farm. J. of Wind Eng. and Industrial
Aerodynamics, 7476, 399410.
Utsunomiya, T., Sato, T., Matsukuma, H. and Yago, K.
2009. Experimental validation for motion of a spar-type
floating offshore wind turbine using 1/22.5 scale model.
Proc. 28th Intl. Conf. on Ocean, Offshore and Arctic Eng.,
Honolulu, Hawaii, USA, OMAE2009-79695.
Wang, C.M., Utsunomiya, T., Wee, S.C. and Choo, Y.S. 2010.
Research on floating wind turbines: a literature survey.
The IES Journal Part A: Civil and Structural Engineering,
Vol. 3(4), 267277.
Wayman, E.N., Sclavounos, P. D., Butterfield, S., Jason, J.,
Musial, W. 2006. Coupled dynamic modeling of floating wind turbine systems. Proc. Offshore Technology
Conference, Houston, Texas.
Weinzettel, J., Reenaas, M., Solli, C. and Hertwich, E.G.
2009. Life cycle assessment of a floating offshore wind
turbine. Renewable Energy, 34 (3), 742747.
Withee, J.E. and Sclavounos, P.D., 2004. Fully coupled
dynamic analysis of a floating wind turbine system. Proc.
8thWorld Renewable Energy Congress, Denver, Colorado,
USA.
Zambrano, T., MacCready, T., Kiceniuk, T. Jr.., Roddier,
D.G. and Cermelli, C., 2006. Dynamic modeling of
deepwater offshore wind turbine structures in Gulf of
Mexico storm conditions. Proc. 25th Intl. Con. on Offshore Mechanics and Arctic Eng., Hamburg, Germany,
OMAE2006-92029.
562

Modelling wave energy for the north coast of Spain

A. Rute Bento, Paulo Martinho & C. Guedes Soares
ABSTRACT: Two state-of-the-art spectral models, WAVEWATCH III (Tolman 2009) and SWAN (Booij et al.
1999), are used to create a wave prediction system with the aim to characterize wave energy conditions for the
Spanish North Coast. Direct comparisons are made and statistical results are obtained to assess the reliability of
the wave prediction system developed. The wave parameters considered for those comparisons are significant
wave heights, mean and peak periods. Theoretical values of wave power are estimated and the spatial distribution
of wave energy was evaluated considering a relevant scenario and some areas with greater energetic potential is
identified. Overall the wave prediction system gives reasonable results when compared with the buoy data and,
therefore, the results obtained for wave power can be accepted as being near to the real potential for the north
coast of Spain.
INTRODUCTION
World energy consumption has been rising considerably and since traditional methods of energy consumption are contributing seriously to environmental
problems there is an urgent need for non-polluting
power generation. Wave energy is considered as a clean
source of energy, with low environmental impacts,
in particular it is seen as a big source of energy not
involving large CO2 emissions. However, difficulties
facing wave power developments must be recognized,
such as irregularities in wave amplitude, phase and
direction, which makes difficult for a device to obtain
maximum efficiency, and structural loading in the
event of extreme weather conditions. Nevertheless,
the advantages of wave energy are clear; it combines
crucial economic, environmental, ethical and social
factors (Clment et al. 2002).
Wave energy results from wind energy transfer to
the water at the free surface of the ocean and, after
being created, waves can travel long distances with little energy loss. Energy fluxes occurring in deep water
sea waves can be very large and, though near the coastline the average energy intensity of a wave decreases,
reflection and refraction can still compensate that as
they lead to energy concentration spots (hot spots).
Situated at the end of the Atlantic fetch, the wave
climate at the western coast of Europe is known as
being highly energetic. The long-term annual wave
power level increases from about 25 kW/m off the
southernmost part of Europes Atlantic coastline up
to 75 kW/m off Ireland and Scotland. The total wave
energy resource for Europe results in 320 GW (Clment et al. 2002) (Fig. 1).
The potential of wave energy extraction can be
acquired from analysis of wave conditions. Buoy data
Figure 1. Wave resource distribution.
can give a general idea of the wave conditions and its

tendencies, but the time period of the measurements
is in general limited. Numerical models enable to
bridge that time gap and to predict wave conditions
in various scenarios, that is why it is essential to use
them to characterize wave climate.
Several approaches have been made in order to
use numerical models to evaluate wave conditions.
A nearshore wave energy atlas has been developed
for Portugal (Pontes et al. 2005) based on a directional wave spectrum hindcast for an 11-year period.
A 44 year hindcast was also made for the whole
European western coast in the context of HIPOCAS
project (Hindcast Dynamic Processes of the Ocean and
Coastal Areas of Europe) (Pilar et al. 2008). There, the
WAM model was used, modified for two-way nesting,
to provide results for processes of wave generation
and deep water propagation. The output parameters
563
were significant wave height, wave direction, mean

and peak period, wind speed and direction, Hs for wind
sea, direction of wind sea, Tm for wind sea, Hs for
swell, direction for swell and Tm for swell. In order to
extend the study referred to the coastal environment,
Rusu et al. (2008) used the SWAN (Booij et al. 1999)
model to make a hindcast for the Iberian coast, with
higher quality than the one of Pontes et al. (2005) that
used a ray model. A regional meteorological model
was also used in the coastal area and it was found that
the skill of the model was improved with the finer grid
wind fields (Rusu et al. 2008). After the HIPOCAS
project (Pilar et al. 2008) and its extension to finite
water depth (Rusu et al. 2008) the effectiveness of
such complex wave prediction system was demonstrated by Rusu & Guedes Soares (2009). In that work
the spatial distribution patterns of the wave energy in
the Portuguese nearshore were also assessed.
The present work aims to evaluate a similar wave
prediction system as the ones referred and analyze
the wave energy spatial distribution for the North
coast of Spain. WAVEWATCH III (Tolman 2009) and
SWAN (Booij et al. 1999) were the two state-of-the-art
models implemented. Two parallel studies were done
by Bento et al. (2011a, b), within the same project,
one for the Irish West Coast and another for UKs
southwest coast, known for having high average wave
power. A study of the wave conditions in coastal areas
was done for both places by coupling the wave models
SWAN and WAVEWATCH III. Validation tests were
carried and results for wave power were obtained for
specific tests sites: Galway and Belmullet in Irelands
case, Cornwall and Pembrokshire in UKs case.
The north coast of Spain was the area considered
for this study, from the NW corner of the Iberian
Peninsula to the SE Bay of Biscay. As a western
European coastline, the north coast of Spain is
exposed to powerful swells and it is also prominent
in wave energy exploitation, with annual mean values near to 40 kW/m in the section between Cape San
Adrin and Cape Ortegal, as can be seen in Figure 2.
Galicia has a very long coastline, approximately
1659 km, and it is known to have great potential in
deepwater energy. Two areas stand out in relation to
wave energy: the coast around Cape Estaca de Bares
and the Death Coast, a name given after the many
lives lost in shipwrecks caused by its energetic waves.
Regarding wave power, its values in the Death Coast
are in the order of 50 kW/m and annual wave energy
exceeds 400 MWh/m (Iglesias & Carballo 2009). For
Estaca de Bares, buoys give a mean value of 40 kW/m
and an annual mean around 360 MWh/m. Asturias,
the region adjacent to Galicia which extends between
Ra de Ribadeo and Ra de la Tina Mayor (250 km),
is battered by strong swells and its average wave
power exceeds 20 kW/m. In relation to annual wave
energy, it exceeds 200 MWh/m. The SE Bay of Biscay,
like the other regions of the northern seaboard of
Iberia, offers a noticeable potential for wave energy
exploitation. The annual wave energy range between
200250 MWh/m in deep waters, but due to irregular
Figure 2. Annual Wave Energy for the North Coast of Spain.
bottom contours in this area hot spots emerge and

they are considered important areas for wave farm
sites (Iglesias & Carballo 2010a, b, c).
2
DESCRIPTION OF THE WAVE HINDCAST

SYSTEM
Third-generation spectral wave models are the current

state-of-the-art for wave climate modeling and form
the essential structure of the majority of wave transformation models including WAM, WWIII, SWAN and
Mike 21-SW. All are based on solving the spectral
action balance equation, which determines the evolution of the action density in space and time. The
action density is defined as energy density divided by
wave frequency and is used because, unlike energy
density, it is conserved in the presence of currents.
The energy density is specified using the linear
two dimensional wave spectrum, with wave energy
distributed over frequency and propagation direction.
Here, E is the wave energy density, t is the time, x

and y are the Cartesian coordinates, is the wave
frequency, is the direction of wave propagation, C
is the speed of wave energy propagation and S refers
to the energy sources/sinks (Folley & Whittaker 2009)
being expressed in Equation 2.
S in a combination of nonlinear interaction wind

wave (Snl ), wind wave interaction (Sin ) and a dissipation term (Sds ). For model initialization it should also
consider a linear input (Sln ).
The wave hindcast system created uses the WWIII
model for wave generation, covering approximately
the entire North Atlantic basin (Fig. 3) while SWANs
domain covers Spains northern coastal environment
(Fig. 4) and connects the large scale to the coastal
simulations.
WWIII has been developed at the Marine Modeling
and Analysis Branch (MMAB) of the Environmental Modeling Center (EMC) of the National Centers
564
Figure 5. Illustration of the buoy locations.

Table 2.
Figure 3. Implemented area for WAVEWATCH III.
Coordinates and depth of the buoys considered.

Coordinates
Latitude
Buoy 1
Buoy 2
Buoy 3
Buoy 4
42.12 N
43.40 N
43.74 N
43.64 N
Longitude
Depth
600 m
386 m
450 m
600 m
9.43 W
9.21 W
6.17 W
3.05 W
Figure 4. Bathymetry of the north coast of Spain.

Table 1.
Computational grids for the wave hindcast system.

Limits
Bathymetries
Latitude
Longitude
Resolution
North Atlantic (15 N72 N) (66 W7 E) 1 1

Spain
(41 N45 N) (11 W1 W) 0.05 0.1
for Environmental Prediction (NCEP) and tends to be

more efficient at global scales, while SWAN, which
was developed at Delft University of Technology,
offers advantages at smaller scales and for shallow
water processes. Both are widely used spectral wave
models that have been validated in a various situations. Recent improvements in SWAN allow it to
be used effectively for sub oceanic scales. Therefore,
the wave prediction system implemented considers
a large SWAN domain that covers the entire coastal
environment.
Two NOAA datasets are used as inputs for WWIII.
The bathymetry came from GEODAS database and
the wind fields are taken from NCEPs Reanalysis
2, with time steps of 6 hours (4 daily data). The
results are generated with a time step of 3 hours.
Afterwards, WWIII data are used as boundary conditions for SWAN. For the SWAN runs, the wind
fields considered are the same as in WWIII, but the
bathymetries were taken from GEBCOs database.
Computational grids can be seen in Table 1.
Figure 6. Hs time series of Buoy 1 vs SWAN results.
3 VALIDATION TESTS IN THE TIME DOMAIN

Measurements from four wave buoys owned by
Puertos Del Estado were used. Their location is shown
in Figure 5 and specified in Table 2.
For the simulations, comparisons in the time
domain are carried out against the in situ measurements. For these validations, the time period considered is from October 1st to November 13th, 2010,
although some data from this time period may not be
included due to buoy failure. The time resolution of
the simulations was of 1 h.
Time series for Significant Wave Height (Hs) are
shown in Figures 69. Figures 1013 show the time
series for Mean Wave Period (Tm) while Figures 1417
illustrates the Peak Period. Each compare buoy results
with SWANs output.
565
Figure 10. Tm time series of Buoy 1 vs SWAN results.
It can be observed that, for the Hs time series,

simulation results by SWAN show the same behavior
as the buoy data, although SWAN performed better
for buoys 1, 2 and 3, than for buoy 4. In terms of
Mean Period (Tm) SWAN clearly underestimates real

data and for the Peak Period (Tp) simulations translate
better the buoy behavior, even though it still has the
tendency to underestimate.
566
Figure 16. Tp time series of Buoy 3 vs SWAN results.

For the analyses of the significant wave height,

scatter plots are also presented in the Figures 1821.
The scatter plots for buoys 1, 2 and 3 show a
more homogeneous distribution of results around the
Figure 18. Hs scatter plot for buoy 1.
567
observation line than the for buoy 4, where the results

are mostly bellow the observation line. Nevertheless,
this corroborates what was previously noted in the Hs
time series.
To better understand the performance of the model,
a statistical evaluation was made. The computed
statistics were the average values of measurements
(Bm) and simulations (Sm), the bias, root mean square
error (RMSE), scatter index (SI) and Pearsons Correlation Coefficient (r) and can be expressed by the
mathematical expressions:
Here Xi represent the measured values, Yi the simulated values and n the number of observations. The
results are shown in Table 3, 4 and 5.
In general, for Hs, the 3rd buoy has the best
results and the 4th buoy has the worst results. For
the first two buoys the statistical results for Hs are
good; it can be observed that the bias values are below
0.1 m, and all correlation coefficients are above 0.8.
For mean and peak periods the results were not as
good as for Hs, as expected. Nevertheless, buoy 4 has
the best results for both mean and peak periods. The
differences between measured periods and simulated
periods may be a result of the different processes used
in obtaining the values. For SWAN they result from
spectral analysis and for the buoys they result from a
direct analysis of the data.
568
Table 3.
Statistical results for Hs.
Significant Wave Height
Buoy 1 (n = 21)
Buoy 2 (n = 211)
Buoy 3 (n = 206)
Buoy 4 (n = 207)
Table 4.
Bm
Sm
Bias
RMSE
SI
2.513
2.738
1.944
1.909
2.564
2.660
1.778
1.257
0.050
0.079
0.166
0.652
0.626
0.647
0.430
0.948
0.249
0.236
0.221
0.496
0.899
0.875
0.902
0.813
Statistical results for Tm.
Mean Period
Buoy 1 (n = 21)
Buoy 2 (n = 211)
Buoy 3 (n = 206)
Buoy 4 (n = 207)
Table 5.
Bm
Sm
Bias
RMSE
SI
7.002
6.830
6.910
6.487
3.850
3.907
3.328
4.930
3.152
2.923
3.582
1.558
3.323
3.076
3.792
1.975
0.475
0.450
0.549
0.304
0.714
0.772
0.840
0.809
Bm
Sm
Bias
RMSE
SI
11.431
10.945
11.120
10.929
8.929
8.727
8.805
10.071
2.502
2.218
2.315
0.858
3.555
3.126
3.672
2.598
0.311
0.286
0.330
0.238
0.514
0.668
0.548
0.630
Statistical results for Tp.
Mean Period
Buoy 1 (n = 21)
Buoy 2 (n = 211)
Buoy 3 (n = 206)
Buoy 4 (n = 207)
4 WAVE ENERGY ASSESSEMENT

With the SWAN model, it is possible to calculate
energy transport per unit of wave front (W/m). It is
computed considering the following formula:
Here x, y are the problem coordinate system and cx ,

cy are the propagation velocities of wave energy in the
geographical space.
Wave power can afterwards be calculated using the
formula:
Afterwards the wave power is normalized dividing

each value by the maximum wave power computed
for that scenario.
An evaluation of significant wave heights spatial distribution (Fig. 22) and its wave power result
Figure 22. Hs spatial distribution. In the background significant wave height scalar fields and in foreground wave
vectors.
(Fig. 23) is made for wave energy assessment. One

case study is considered. It corresponds to 2010/11/09,
at 6 h, where one of the highest values of Hs is
encountered.
Evaluating theoretical results for wave energy, it
can be observed that, in general, the northwestern
coast registers higher values of wave energy. It is also
visible that half of the North Spanish coast is under
the shadow effect caused by Bay of Biscays northern
569
REFERENCES
Figure 23. Computed Normalized Wave power (W/m). In

background wave power scalar fields and in foreground
energy transport vectors.
area. The section between Cape San Adrin and Cape

Ortegal registers the highest values for wave power,
while Santander registers the lowest values for wave
power.
5
CONCLUSIONS
WWIII and SWAN, both state-of-the-art spectral

models, are used to create a wave prediction system to evaluate wave conditions for Spains Northern
coast. Having in mind that the important wave parameters for wave energy analysis are the significant wave
height and the wave period, the ability of the model to
simulate wave energy is tested. To have a perception of the accuracy of the numerical wave model,
a validation analysis of the computed results against
buoy measurements is made and theoretical values
for the energy transport components are obtained.
Taken as a whole, the results are good when
comparing significant wave height with buoy data.
For mean and peak period the results are not as good
as was expected. They have the same time evolution but have an almost constant bias. The results for
wave power depend mainly on significant wave height
and thus they should be fairly accurate.
ACKNOWLEDGEMENTS
Bento, A.R., Martinho, P. and Guedes Soares, C., 2011b,

Modelling Wave Energy Resources for UKs Southwest
Coast, Proceedings of the Oceans 2011 IEEE Santander
Conference, CD-ROM Edition, Spain.
Bento, A.R., Martinho, P., Campos, R. and Guedes Soares,
C., 2011a, Modelling Wave Energy Resources in the
Irish West Coast, Proceedings of the 30th International
Conference on Ocean, Offshore and Arctic Engineering,
Rotterdam, Netherlands, Paper OMAE2011-50346.
Booij, N., Ris, R. C. and Holthuijsen, L. H., 1999. A Third
Generation Wave Model For Coastal Regions. Part 1:
Model Description And Validation, J. Geophys. Res., 104,
C4, pp. 76497666.
Clment, A., McCullen, P., Falco, A., Fiorentino, A.,
Gardner, F., Hammarlund, K., et al., 2002, Wave Energy in
Europe: Current Status and Perspectives, Renewable and
Sustainable Energy Reviews 6, pp. 405431.
Folley, M. and Whittaker, T.J.T. 2009, Analysis of the
nearshore wave energy resource, Renewable Energy,
vol. 34, pp. 17091715.
Iglesias, G. and Carballo, R., 2009, Wave Energy Potential
Along the Death Coast, Energy 34, pp. 19631975.
Iglesias, G. and Carballo, R., 2010a, Wave Energy Resource
in the Estaca de Bares Area, Renewable Energy 35,
pp. 15741584.
Iglesias, G. and Carballo, R., 2010b, Offshore and Inshore
Wave Energy Assessment: Asturias (N Spain), Energy 35,
pp. 19641972.
Iglesias, G. and Carballo, R., 2010c, Wave Energy and
Nearshore Hot Spots: The Case of the SE Bay of Biscay,
Renewable Energy 35, pp. 24902500.
Pilar, P., Guedes Soares, C., Carretero, J.C., 2008, 44-year
Wave Hindcast for the North East Atlantic European coast,
2008, Coastal Engineering, 55, pp.861871.
Pontes, M.T., Aguiar, R., Oliveira Pires, H., 2005, A
Nearshore Wave Energy Atlas for Portugal, Journal of
Offshore Mechanics and Arctic Engineering, Vol. 127,
pp.249255.
Rusu, L.; Pilar, P., and Guedes Soares, C. 2008 Hindcast
of the Wave Conditions along the West Iberian Coast,
Coastal Engineering. 55(9), pp. 906919.
Rusu, E., Guedes Soares, C., 2009, Numerical Modelling
to Estimate the Spatial Distribution of the Wave Energy
in the Portuguese Nearshore, Renewable Energy, 34,
pp. 15011516.
Tolman, H. L, 2009. User manual and system documentation of WAVEWATCH IIITM version 3.14 , Environmental
Modeling Canter, Marine Modeling and Analysis Branch.

MAREN Marine Renewable Energy, Energy Extraction and Hydro-environmental Sustainability, which is
partially funded by the Atlantic Area Programme.
570

Reflection and transmission of gravity waves by a large floating

circular elastic plate
J. Bhattacharjee & C. Guedes Soares
ABSTRACT: The scattering of free surface gravity waves by a circular very large floating structure (VLFS)
in water of finite depth is studied. The circular platform is assumed to be a large elastic plate with constant
flexural rigidity and homogeneous stiffness. The edge of the plate is assumed free of shear forces, bending and
twisting moments. The geometry associated with the floating circular plate is assumed to be axi-symmetric.
Hence, the mathematical analysis describes the boundary value problem under the linearized water wave theory
in the two dimensional cylindrical polar coordinate systems. Using the continuity of velocity and pressure across
the vertical plate-water interface and then applying the higher order Fourier expansion formulae along with
the orthogonal properties of the eigenfunctions, the full solution is obtained. The reflection and transmission
characteristics of the incident wave and the behavior of flexural gravity wave propagation along the plate are
analyzed in detail.
INTRODUCTION
The hydroelastic analysis of a very large floating structure (VLFS) in water waves has gained a great deal
of attention due to the growing demand to use more
and more ocean space for various human activities.
These unconventional structures are often proposed as
a thin mat-like configuration extending in the horizontal direction up to several kilometers with a thickness
of less than 10 m. Such floating structures are very
flexible and the elastic deformations due to wave
action are more crucial than the rigid body motions. In
general, rectangular platforms are preferred as it is the
basic shape for construction. In reality, the geometry
of the platform is expected to be more complicated.
Various numerical methods have been developed over
the years to improve the hydroelastic predictions
of VLFS. On the other hand, analytical results are
always in need to provide the benchmarks. The extensive review of literature on the significant developments in the field of wave interaction with flexible
floating structures is provided by Kashiwagi (2000),
Watanabe et al. (2004) and Chen et al. (2006).
The study of VLFS attains more importance due
to its application in the analysis of wave interaction
with large floating ice sheets in the fields of Arctic
Engineering and the Cold Region Science and Technology. Squire et al. (1995) and Squire (2007) provided
extensive review of the work done in the field of wave
interaction with floating sea-ice, which is also modeled as a floating elastic plate. Squire (2008) discussed
the synergies between the two fields: wave interaction
with VLFS and the research on wave-ice interaction.
In general, the shape of the VLFS is expected to

be complex and arbitrary in nature. However, for the
preliminary analysis, simple regular rectangular or circular geometries are considered in most of the works
available in the literature. Meylan & Squire (1996)
analyzed the response of a floating circular ice floe to
ocean waves. They modeled the ice floe as a flexible
thin elastic plate and applied the eigenfunction expansion and the Greens function technique to predict
the displacement and the three dimensional scattering pattern around the floe. Zilman & Miloh (2000)
obtained a closed form solution for the hydroelastic
response of a floating circular elastic plate in shallow
water. This was further solved by Tsubogo (2001) with
the aid of advanced boundary element method.
Peter et al. (2003) studied the scattering of free
surface gravity waves by a circular plate in water of
finite depth and obtained a closed form solution by
using the angular eigenfunction expansion and the
matching of velocity and pressure across the interface. Sturova (2003) analyzed the transient problem of
floating circular elastic plate in shallow water under
the action of unsteady external load. Watanabe et al.
(2003) examined the pontoon type circular VLFS in
finite water depth by the Galerkin method. Korobkin
& Ohkusu (2004) studied the axi-symmetric unsteady
collision of two circular plates, one of which is floating
on a liquid of shallow depth.
Andrianov & Hermans (2005) considered the
diffraction of free surface gravity waves by a floating
circular elastic plate in water of both finite and infinite depths. They applied the Greens theorem to obtain
integro-differential equations and proposed numerical
571
schemes to obtain the solution. Watanabe et al.

(2006) obtained benchmark hydroelastic responses for
a floating circular VLFS under wave action. They
applied the eigenfunction expansion matching method
to solve the hydrodynamic radiation and diffraction
forces. The modal responses of the uniform plate are
obtained from the hydroelastic equation of motions via
Rayleigh-Ritz method.
Pham et al. (2008) performed the hydroelastic
analysis of pontoon type circular VLFS with an
attached submerged plate by the aid of eigenfunction
expansion and the Rayleigh-Ritz method. Wang et al.
(2004) presented the exact vibration mode shapes and
modal-stress resultants for freely vibrating circular
Mindlin plates with free edges. The circular geometry is suggested in the literature as an alternative for
the construction of floating airbase or a floating island.
In the present study, the scattering of free surface
gravity waves by a circular VLFS in water of finite
depth is studied. The circular platform is assumed to
be a large elastic plate with constant flexural rigidity and homogeneous stiffness. The edge of the plate
is assumed to be freely floating and hence with zero
shear forces, and zero bending and twisting moments.
The geometry of the structure is assumed to be axisymmetric with respect to the vertical axis. Thus, the
mathematical analysis describes the boundary value
problem under the linearized water wave theory in the
two dimensional cylindrical polar coordinate systems.
The whole domain of consideration is divided into
two regions, namely the free surface region and the
plate-covered region.
Applying the higher order Fourier expansion formulae, the velocity potentials are expanded in terms
of appropriate eigenfunctions. Using the continuity of
velocity and pressure across the vertical plate-water
interface, and applying the orthogonal properties of
the eigenfunctions, the boundary value problem is
then reduced to a linear system of algebraic equations.
Once the unknowns in the potential expansion are
determined through numerical methods, the reflection
and transmission characteristics of the incident wave
and the behavior of flexural gravity wave propagation
along the circular plate are analyzed in detail. Further,
the effects of the shape of the floating structure and
the change in water depth are also investigated. The
results are expected to provide useful benchmarks for
future studies.
The mathematical formulation of the physical problem is based on the cylindrical polar coordinate system
(r, , y) as shown in Figure 1. The circular elastic plate
is assumed to be floating in an infinite fluid domain.
For the sake of simplicity, it is assumed that the system
is symmetric in terms of rotation about the vertical axis y. Therefore, the axi-symmetric coordinate
system (r, y) is chosen that is independent of the rotational angle . The origin of the coordinate system is
Figure 1. Schematic diagram.
assumed to be at the centre of the circular plate of

radius a and small thickness d and the y- axis being
vertically downward positive as shown in Figure 1.
The bottom topography is assumed to be uniform at
y = h. The domain of consideration is divided into
two regions: region 1 is the plate covered region
r < a, 0 < y < h and region 2 is the undisturbed free
surface region r > a, 0 < y < h as shown in Figure 1.
Under the assumption of the linearized water
wave theory, the fluid is considered to be incompressible, inviscid and the motion is irrotational and
time harmonic with angular frequency . Therefore,
there exists a velocity potential (r, y, t) of the form
(r, y, t) = Re[(r, y)eit ], where the spatial velocity potential (r, y) satisfies the governing Laplaces
equation
where
The no flow condition at the rigid bottom boundary

yields
On the upper surface, the draught of the floating

flexible circular plate is assumed to be very small
compared to the radius of the plate. Thus, the circular
plate is considered as a part of the water surface with
different physical properties that are dominated by
the flexural rigidity of the structure. Using the EulerBernoulli beam theory, the boundary condition on the
plate covered surface is given by
where D = EI /(g i d2 ), I = d 3 /12(1 2 ), Q =

2 /(g i d2 ), E =Youngs modulus, = Poisson
ratio, = density of water, i = mass density of the ice
sheet and g = acceleration due to gravity.
572
In region 2, the free surface boundary condition

yields
kind of order zero, K0 is the modified Bessel function

(2)
(1)
of the second kind of order zero and H0 and H0 are
the Hankel functions of order zero. The eigenfunctions
fn (y) and gn (y) are given by
where k = 2 /g. The far field radiation condition is

given by
where B0 is the constant associated with the amplitude

of the outgoing progressive wave, k0 is the corresponding wave number. The function 2 represents the
(1)
(2)
velocity potential in region 2 and H0 and H0 are the
Hankel functions.
The continuity of velocity and pressure across the
plate water vertical interface at r = a, 0 < y < h
yields
At the finite edge of the elastic plate at r = a, the

free edge conditions of zero shear force and bending
moment are prescribed as
The above conditions will ensure the unique solution

of the aforementioned BVP.
3
METHOD OF SOLUTION
The method of solution is based on the eigenfunction expansion and the associated orthogonal relations
along with the matching technique. Using the generalized Fourier expansion formulae (see Manam et al.
2006), the velocity potentials 1 and 2 in the regions
1 (r < a) and 2 (r > a) respectively are expanded in
terms of appropriate eigenfunctions as follows:
with the eigenvalues kn n = 0, 1, 2, . . . and pn ,

n = 0, I , II , 1, 2, . . . satisfying the dispersion relations
In the present study, the constant associated with

the amplitude of the incident wave is taken as unity,
without any loss of generality. The expansions of
the potential functions contain the unknown constants An , n = 0, I , II , 1, 2, . . ., Bn , n = 0, 1, 2, . . . to be
determined by applying the matching of velocity and
pressure across the plate water interface along with
the orthogonal properties of the eigenfunctions. The
dispersion relation (10b) for the flexural gravity waves
contains two real roots of the form p0 , and four
complex roots pn , n = I , II , III , IV of the form piq
in the four quadrants of the complex plane. Further, the
dispersion equation has infinitely many purely imaginary roots of the form pn , n = 1, 2, . . .. On the other
hand, the free surface dispersion relation (10a) contains two real roots of the form k0 , infinitely many
purely imaginary roots of the form kn , n = 1, 2, . . ..
In the present study, only the positive real roots p0 , k0 ,
the positive roots +pn , +kn , n = 1, 2, . . . and the complex roots of the form p iq with positive real parts
are considered in order to obtain a bounded solution.
The eigenfunctions fn and gn satisfy the orthogonal relations (see Manam et al. 2006, Karmakar et al.
2009)
with
where J0 is the Bessel function of first kind of

order zero, I0 is the modified Bessel function of first
573
where the higher order mode-coupling relations

fm , fn 1 and gm , gn 2 are defined as
with denoting differentiation with respect to y.

The non-integral part present in the mode-coupling
relation for flexural gravity waves is the only difference with the mode-coupling relation for the free
surface gravity waves. It accounts for the higher order
term present in the upper surface boundary condition
in Eq. (4) due to the flexural rigidity of the structure.
Applying the mode coupling relation as in equation (12a) on 1r and fn and using the continuity
condition (6) along with the orthogonal relation as in
equation (11a), the expansion formula yields
One more equation is required to obtain a unique solution. The edge condition (7b) yields the final equation
as follows
Therefore, equations (13)(15) constitutes the linear system of algebraic equations to solve for the
unknown constants. The infinite series sums therefore present in the expansion formulae of the velocity
potentials are truncated up to a finite number of terms
N (say) in order to perform the numerical calculations.
Hence, equations (13)(15) yield (2N + 5) equations
for (2N + 5) unknowns. It may be noted that the number of terms N to be retained in each of the expansions
corresponding to the respective evanescent modes
depends on the convergence of the infinite series sums.
In the present study, it is observed that the number of evanescent modes required in the free surface
region is N = 15 whereas, N = 10 is sufficient for the
plate covered region. Therefore, for convenience in the
numerical calculation, in the present study the number
of evanescent modes in both the regions is taken as
N = 15. This further shows that the present method is
very efficient in terms of computer memory and time.
The surface displacements 1 and 2 in region 1 and
2 are given by
where
It may be noted that is an unknown to be determined

from the edge condition (7b). The other edge condition
(7a) is already used in deriving the above equations
(13a) and (13b). Once again, applying the orthogonal
relation (11b) upon using the mode coupling relation
as in equation (12b) on 2 and gn and the continuity
of pressure as in equation (6), gives
The reflection and transmission coefficients Kr and Kt ,

respectively, are defines as
where Kr is the ratio of amplitude of the reflected

and the incident progressive waves in the free surface
region. On the other hand, Kt is the ratio of transmitted flexural gravity wave amplitude in the plate
574
covered region to the incident wave amplitude in the

free surface region.
The surface strain and the shear force Xs along
the circular floating elastic plate are given by
Once the unknown constants are determined, substituting the potential expansion in the above relation
will provide the behavior of the strain and shear force
along the plate length.
Figure 2. Transmission coefficient Kt versus time period T

for different plate radius a with h = 30 m, d = 1 m.
In the present section, numerical results for the reflection and transmission coefficients Kr and Kt respectively, the free surface and elastic plate elevations 1
and 2 , respectively, the surface strain and the shear
force Xs are presented and discussed in brief.
The numerical values of the physical parameters that are kept fixed throughout the computations
are = 1025 kgm3 , i = 925 kgm3 , E = 5 GPa,
= 0.3 and g = 9.81 ms2 . It may be noted that due to
the assumption of the axi-symmetric cylindrical polar
coordinate system, the three dimensional problem has
been reduced to the two dimensional problem, which is
independent of the rotational angle . Assuming = 0,
the radial axis r coincides with the x axis in the two
dimensional Cartesian coordinate system (x, y), i.e.,
r = x, y = y. Therefore, all the parameters related to the
circular elastic plate are presented here for 0 < r < a,
i.e., for 0 < x < a.
As mentioned earlier, the numerical calculation
requires 10 to 15 numbers of evanescent modes in the
infinite series sums to converge and thus, the method
becomes efficient and robust in terms of computer
memory and time. The complex roots in the plate covered region play an important role in the behavior of
the flexural gravity wave propagation along the plate.
The numerical calculation shows that the reflection
coefficient is almost unity for all the incident wave
periods.
Figure 2 shows the variation of Kt with the period
T for different values of the plate radius a. It is
observed that the ratio of amplitudes of progressive
waves in the flexible plate and the amplitude of the
incident water wave increases with increase in the
period of the incident wave. Further, the ratio becomes
more than one after a certain incident wave period,
which implies that the amplitude of the transmitted
wave in the plate region is higher than the incident
wave amplitude. It may be further observed that the
change in the radius of the plate does not affect significantly in the behavior of the transmitted wave
Figure 3. The surface displacements 1 and 2 for different

plate radius a with h = 30 m, d = 1 m.
Figure 4. The surface strain for different plate radius a

with h = 30 m, d = 1 m.
amplitude as long as the dimension of the plate is comparable or much bigger than the incident wave length,
which is the case for VLFS structure.
Figures 3, 4 and 5 show the variation of free surface
and plate elevations, strain and shear force along the
plate for different values of the radius of the circular
plate a. It is observed that the amplitude of the flexural gravity wave that is propagating along the plate
is more than the free surface gravity waves. In the
free surface region, the maximum amplitude is attained
near the edge of the plate and it becomes steady away
from the plate. It may further be mentioned that the
flexural gravity wavelength is always greater than the
free surface gravity waves and for longer incident
wave periods, the flexural gravity wavelength becomes
575
Figure 5. The shear force Xs for different plate radius a

with h = 30 m, d = 1 m.

water depth h with a = 100 m, d = 1 m.
Figure 7. The surface strain for different water depth h

with a = 100 m, d = 1 m.
almost equal to the free surface gravity waves as also

observed by Tsubogo (1999).
Figure 4 shows that the strain along the circular plate
surface increases as the radius of the plate increases.
Further, the strain is maximum near the centre of the
plate. It becomes zero near the edge of the plate,
which also satisfies the free edge condition of zero
strain. A similar observation is made in the case of
shear force along the plate surface in Figure 5, which
becomes nearly zero at the edge. This is once again
in compliance with the assumption of free movement
of the edge. The average shear force increases with
increase in the radius of the plate.
Figures 6, 7 and 8 show the variation of the free surface and plate elevations, strain and shear force along
the plate for different values of the water depth h.
Figure 8. The shear force Xs for different water depth h with

a = 100 m, d = 1 m.

plate thickness d with h = 30 m, a = 100 m.
Figure 10. The surface strain for different plate thickness

d with h = 30 m, a = 100 m.
Figure 6 shows that the wave amplitude in both

regions of free surface and plate covered region
decreases as the water depth increases. On the other
hand, the wavelength of the progressive wave increases
with increase in the water depth. Figures 7 and 8 show
that the strain and the shear force along the plate
decrease as the water depth increases. The maximum
of the strain is attained at the centre of the plate and it
becomes zero at the edge. On the other hand the shear
force gradually increases away from the centre of the
plate and attains a maximum before dropping down to
zero at the edge of the plate.
The variation of free surface and plate elevations,
strain and shear force along the plate for different values of the plate thickness are depicted in
Figures 9, 10 and 11. It shows that the wave amplitude of the flexural as well as the free surface gravity
576
the plate. The effects of the change in the dimension

of the plate on these parameters are investigated by
varying the radius and the thickness of the circular
elastic plate. It is observed that the strain increases
away from the edge of the plate. The wavelength in
the plate covered region is greater than the free surface gravity waves. Further, the effect of the change
in water depth is studied. The present work provides
some benchmarks for future studies that will include
the three dimensional structure with various motion
characteristics.
Figure 11. The shear force Xs for different plate thickness
d with h = 30 m, a = 100 m.
waves decreases significantly as the thickness of the

plate increases. This is due to the fact that the increase
in the plate thickness eventually increases the flexural
rigidity of the plate. On the other hand, the wavelength increases with increase in the thickness of the
plate.A similar observation is made byTsubogo (1999)
for hydroelastic waves in flexible floating structures.
The strain near the centre of the plate increases with
increase in the thickness of the plate, whereas the
reverse phenomenon is observed as it approaches near
the edge of the plate. The numerical results agree well
with the assumption of zero shear force and bending
moments at the edge of the plate. The shear force at
the edge for the case of d = 2 m shows a little deviation from the assumed edge condition. This may be
due to the fact that the flexural rigidity rose beyond a
particular value that does not comply with the present
flexible plate theory and the analysis needs to be
modified to consider the plate as a rigid structure.
5
CONCLUSIONS
The interaction of water waves with a circular VLFS

is analyzed in the linearized water wave theory.
Considering the axi-symmetric nature of the geometry,
the problem is studied in the two dimensional cylindrical polar coordinate systems. The circular floating
structure is modeled as an Euler-Bernoulli thin elastic plate with free edges. Applying the higher order
Fourier expansion formula and the associated orthogonal mode coupling relations, the boundary value
problem is reduced to a linear system of algebraic
equations. The expansion formula contains infinite
series sums associated with the evanescent modes in
both the free surface and plate covered regions.
The numerical calculation show that both the series
sums converge for 10 to 15 terms. Therefore, the
present method provides an efficient tool to analyze
floating flexible structures having regular geometries
even in the cylindrical polar coordinate system. It may
be mentioned that the present method can be easily extended to analyze the non axi-symmetric case
in the three dimensional cylindrical polar coordinate
systems. The nature of the flexural gravity waves that
propagate along the circular plate is investigated via
the wave elevation, strain and the shear force along
ACKNOWLEDGEMENTS
The first author has been funded by the Portuguese Foundation for Science and Technology (FCT,
Fundao para a Cincia e a Tecnologia, Ministrio
da Cincia, Tecnologia e Ensino Superior) through
a Post Doctoral Fellowship under the contract no.
SFRH/BPD/47210/2008.
The work has been performed within the project
FLOATERS Wave Interaction with Floating
Structures in Shallow Water financed by the
PTDC/ECM/111242/2009.
REFERENCES
Andrianov, A.I. & Hermans, A.J. 2005. Hydroelasticity of a
circular plate on water of finite or infinite depth. J. Fluids
and Structures. 20: 719733.
Karmakar, D., Bhattacharjee, J. & Sahoo, T. 2009. Wave interaction with multiple articulated floating elastic plates.
J. Fluids Struct. 25: 10651078.
Korobkin, A.A. & Ohkusu, M. 2004. Impact of two circular
platesone of which is floating on a thin layer of liquid.
J. Engng. Math. 50: 343358.
Manam, S. R., Bhattacharjee, J. & Sahoo, T. 2006. Expansion
formulae in wave structure interaction problems. Proc.
Roy. Soc. Lond. Ser. A 462(2065): 263287.
Meylan, M. & Squire, V. 1996. Response of a circular ice floe
to ocean waves. J. Geophys. Res. 101(C4): 88698884.
Peter, M.A., Meylan, M. & Chung, H. 2003. Wave scattering
by a circular plate in water of finite depth: a closed form
solution. Proc. 13th Intl. Offshore & Polar Engng. Conf.
Honolulu, Hawaii, USA. 180185.
Pham, D.C., Wang, C.M. & Utsunomiya, T. 2008. Hydroelastic analysis of pontoon-type circular VLFS with an
attached submerged plate. Appl. Ocean Res. 30: 287296.
Sturova, I.V. 2003. The action of an unsteady external load
on a circular elastic plate floating on shallow water. J.
Appl. Maths. Mechs. 67(3): 407416.
Squire, V. A. 2007. Of ocean waves and sea-ice revisited,
Cold Reg. Sc. Tech., 49(2): 110133.
Squire, V. A., Dugan, J. P., Wadhams, P., Rottier, P. J. &
Liu, A. K. 1995. Of ocean waves and sea ice, Annual Rev.
Fluid Mech., 27: 115168.
Squire, V. A. 2008. Synergies between VLFS hydroelasticity
and sea-ice research. Intl. J. of Offshore and Polar Engng.
18(3): 113.
Tsubogo, T. 1999. On the dispersion relation of hydroelastic
waves in a plate. J. Marine Sc. & Tech. 4(2): 7683.
577
Tsubogo, T. 2001. The Motion of an elastic disk on shallow

water in waves. Proceedings of 11th Intl. Offshore & Polar
Engng. Conf., vol. 1, Stavanger, Norway, 229233.
Wang, C.M., Xiang,Y., Watanabe, E. & Utsunomiya, T. 2004.
Mode shapes and stress-resultants of circular Mindlin
plates with free edges. J Sound &Vibration. 276: 511525.
Watanabe, E., Utsunomiya, T., Wang, C.M. & Xiang,Y. 2003.
Hydroelastic analysis of pontoon-type circular VLFS.
Proc. of 13th Intl. Offshore & Polar Engng. Conf., vol. 1,

Honolulu, USA, 9399.
Watanabe, E., Utsunomiya, T., Wang, C.M. & Hang, L.T.T.
2006. Benchmark hydroelastic responses of a circular
VLFS under wave action. Engng. Structures. 28: 423430.
Zilman, G. & Miloh, T. 2000. Hydroelastic buoyant circular
plate in shallow water: a closed form solution.Appl. Ocean
Res. 22: 191198.
578

Computational simulation of a buoy behaviour for automation and

power generation in aquaculture facilities
Sara Ferreo & Alicia Munn
Cetnaga Centro Tecnolgico del Naval Gallego, Spain
V. Daz-Casas, Pablo Farias & F. Lpez Pea

Integrated Group for Engineering Research, University of Corua, Spain
ABSTRACT: In this paper the computational simulation of the behaviour of a buoy intended for automation
and power generation in aquaculture facilities has been studied. The main purpose for this buoy is to serve
as power supply for an aquaculture farm to support a system that could operate independently. The dynamic
behaviour of the buoy results to be of crucial importance in order to guarantee the correct operation of both the
power generation equipment and the farm control system. To investigate this dynamic behaviour a computational
simulation, through a 3DOF solver and a mooring simulator, have been combined. Since the fish farm will be
installed offshore, a key point to be considered during its design is that the installation must be capable to undergo
adverse weather conditions without damage. In this way all its elements susceptible to suffer damage or loss
are designed and installing such to be stowed automatically (i.e. windmill), and the cages (including both the
structure and the networks) have a suitable strength to withstand in the event of extreme weather conditions.
INTRODUCTION
Nowadays coastal aquaculture presents many challenges that affect the production and require a
change in business model and in engineering solutions
(Halwart et al., 2007). Limitations associated with
the available space in coastal areas, the poor quality
of water in some of these areas and the fulfilment
of environmental standards and food safety regulations often prone to consider strategies for growth
that eventually moves production offshore. However,
in order to transfer aquaculture production offshore
one has to face several challenges, not only technological but also related to market positioning and scale of
business.
Although offshore aquaculture is one of the main
routes to meet growing global demand for fish products (PTEPA, 2009), facing a significant increasing in offshore production still requires substantial research efforts and multidisciplinary studies in
order to achieve this qualitative and quantitative leap
in production without taking unnecessary technical,
economical or personnel risks.
Therefore research efforts must be aimed not only
towards the design of structures able to withstand
the hard environmental conditions that may occur
in remote sea areas apart from the coast, but also
to resolve some issues related to the management
and maintenance of these structures while minimizing risk and associated costs derived from these
tasks: cleaning, monitoring of biomass, feeding,

prevention of leakage, adult fishes, etc.
In this context, one possible solution is to provide a
buoy equipped with all control systems for the automation of the aquaculture farm, and that also be able to
generate the energy needed for the operation of all the
systems that are part of the aquaculture facilities. The
development of the automation and power generation
buoy requires a correct design aiming to guarantee
the correct operation of all the systems inside it.
As the buoy design criteria are depending on the
sea conditions at the buoy emplacement, a particular
location in the Galician coast is going to be used to
serve as a testing work bench. This doesnt diminish
the generality of this work as the application to any
other place can be achieved straight forward by just
adjusting the criteria to the oceanographic conditions
of the new area.
The dynamic behaviour of offshore mooring
buoys (employed in oil & gas offshore sector) has been
investigated in model tests and numerical simulations
for many years (Cozjin and Bunnik, 2004). Recently,
studies about oceanographic buoys have also been
developed, (Wallinga et al., 2003).
Since the aquaculture farm will be a facility located
in the open sea and considering the severe environmental conditions which take place in the Galician coast,
makes the study of the dynamics of the buoy an issue of
great importance in order to ensure that its behaviour
is within acceptable limits.
579
Figure 4. Current direction distribution.
Figure 1. Wave height distribution.
The facility design concerning dimensions, structural

strength, and optimum systems for power generation
(wind and solar characteristics) has been addressed
in regard of this study results. Figures 1 to 4 show
some graphs summarizing the oceanographic conditions in the area (waves and currents) Meteogalicia1
and Puertos del Estado2 .
3
Figure 2. Wave peak period distribution.
Figure 3. Current speed distribution.
DESIGN CRITERIA
The buoy will be placed in an aquaculture facility

designed to be installed in the Galician coast (specifically in the region called area of Rias Baixas in
the Pontevedra province). The facility design criteria,
engineering analysis and specification of components
are set specifically to match those of this coastal area.
The farm need be located in a place having adequate water characteristics concerning PH and temperature, and having a flow renovation suitable for
fish survival. A full environmental, meteorological
and oceanographic study of the site has been made
regarding sea conditions (waves and currents), solar
radiation, and wind resource among other parameters.
OFFSHORE AQUACULTURE FARM
In this section, a reference fish farm and its operation have been modelled. The general distribution
and main characteristics of the farm are taken to be
representative of the ones used in the region.
The proposed aquaculture facilities under the scope
of this work are designed according to the monoculture of flat fish, as might be turbot. This kind of
fish is considered the priority objective of the aquaculture industry in Galicia, and can be presented as
the most representative specie to be cultivated in this
region, (Beaz-Paleo, 2008).
The fish farm considered for the present study consists of 12 floating cylindrical cages, flexible type, and
made of plastic. Although some cages are anchored
individually, the most common practice is to group
them, not just because it simplifies and minimizes the
costs of funding, but also for reasons of operation and
maintenance.
Circular offshore aquaculture cages are usually
attached to a cable network. This structure can have
multiple cells; any cage is placed in one of these cells
and secured with ropes to the network. Hawsers must
be sufficiently tight to keep the cage centred within the
cell, but allowing it to have some movement in such a
way that it can damp the blows of the sea.
In this study, the farm has 12 cages of 38 meters in
diameter and 5 meters height, with which, according
with our preliminary calculations, it can be achieved
an annual fish production of about 1800 tons.
It is clear that aquaculture farming in offshore areas
increases the difficult to perform the maintenance of
the farm and the control of its facilities. For that reason
1
2
580
www.meteogalicia.es/
w3.puertos.es/es/oceanografia_y_meteorologia/index.html
Figure 5. Sketch of the arrangement of cages with connecting lines and mooring.
Figure 6. Basic external appearance of the buoy.
a system for monitoring all farm processes combined

with an autonomous control system appears to be a
good option to be developed in order to allow a remote
operation of the farm.
Generally, the tasks associated with fish farming activities are: maintenance of facilities, control
of physicochemical parameters (alkalinity, temperature, marine currents. . .), supplying the appropriate
percentage of food, and control of biomass. A control system has been developed so that the status of
the buoy can be easily monitored and the feeding
schedules can be remotely modified through Internet.
The control system will have the ability to monitor
buoy position as well as environmental and system
diagnostic data (Irish and Boduch, 2006).
The remote monitoring system will be based on
software vision cameras, which are going to be
installed in each of the cages. The system will be
able to identify and separate the particles found in
the environment through the evaluation of the geometry and collapse behaviour parameters, among others.
This monitoring system will made possible differentiate between fish, food, faecal deposits, and foreign
particles.
With this technology it will be possible to control
the systems used in the feeding of farmed species
by using specific software developed ad-hoc. Thus,
feeding may be programmed from a computer and
controlled by the cameras.
In addition of providing real time information
on the status of the facilities, the surveillance system will also provide environmental, biological and
oceanographic information.
Its geometry is basically composed of:
A main body having a lower section (submerged

part) of 0.5 m height, and a conical truncated shape
with an upper diameter of 6.276 m and a lower diameter of 2.2 m (it should be such that the electric
batteries can be disposed on the bottom, to get a
better centre of gravity position). This lowest part
has also a reserved height of 0.300 m at the very
bottom, which could be filled with fixed ballast.
On the top of this lower body is a cylindrical part
having a diameter equal to the maximum diameter
of the truncated cone, and a height of 0.9 m.
The upper body is a hexagonal truncated pyramid.
It holds 20 solar panels at a rate of 3 panels in each
face of the truncated pyramid (plus two more in the
buoy deck). Likewise, a small wind turbine (with
two blades) is to be placed in the buoy deck. This
wind turbine has a diameter of 3 meters.
The design of the buoy geometry has been conditioned

by solar panels (and its optimal position) and (although
a lesser extent) a wind generator.
Inside the buoy, as mentioned before, the batteries
and also electrical equipment, as well as all control
systems, will be located.
BUOY MOORING CONFIGURATION
Taking into account the forces exerted to the buoy

by currents, wind and the chain mooring, the first
approach to the parameterization of the buoy mooring
has been made by assuming it as a catenary line. So
starting with the known catenary differential equation:
BASIC BUOY CONFIGURATION
Based on the specific requirements of our project, a

design of the buoy geometry has been developed as
shown. Later it has been optimized in order to obtain
the basic design that we see in Figure 6.
The buoy consists of a steel float container, which
is 3.6 m depth, and having a maximum diameter of
6.276 m. It encloses all the required electrical, power
generation, and control equipment.
where Fd is the sum of Drag Forces.
and w is the chain weight per unit length.
581
The definition of the wind and current drag forces

has been made by using the following expressions
(IALA, 1998):
Cw = Wind force coefficient

Cs = Shape coefficient
Ch = Height coefficient
A = Vertical projected area of each surface exposed to
the wind
V = Design wind speed
where
Css = Current force coefficient
Cd = Drag coefficient
and the chain drag force has been defined by its normal (Fnchain) and tangential (Ftchain) components.
That is:
Figure 7. Buoys mooring behaviour.
where the normal (Cn) and tangential (Ct) coefficients

are calculated as:
Therefore, the total chain drag force can be defined

as (Ikeda et al., 1978):
Figure 8. Buoys dimensions (m).
The results obtained to different current and wind

velocities indicate that a single line mooring would
submit the buoy to large excursions, the displacement
due the external forced is shown in Figure 7. So,
four mooring lines are recommended to keep the buoy
stable and in position. The point where the mooring
lines are attached is the middle of the lower part of the
structure.
6
Table 1. Weights and gravity centers of the buoy.
Steel Weight
Solar Panels
Wind Generator
Batteries
Other systems
Total
Weight
KG
4.811 t
0.340 t
0.055 t
1.612 t
0.189 t
7.007 t
1.423 m
2.269 m
6.930 m
0.532 m
2.265 m
1.325 m
BUOYS NAVAL ARQUITECTURE

CALCULATIONS
Figure 8 represents the buoy and its principal

dimensions.
Once the forms of the buoy are defined, the second
step is to fulfil all necessary hydrostatic and hydrodynamic calculations. To achieve this, the weight and
centre of gravity are estimated as in Table 1.
Thus the principal hydrostatic characteristics
of the buoy should be summarized in Table 2. All
hydrostatic calculations have been made with Maxsurf

Software (developed by FormSys Formation Design
Systems Pty Ltd).
In order to evaluate the large angle response of
the boy a large angle stability of the buoy has been
calculated, as shown in Figure 9.
To calculate the response of the buoy respect to
defined sea conditions, wave spectrum parametersheights and periods in the area under study are
582
Table 2.
Hydrostatic characteristics of the buoy.
(Displacement)
V (Volume)
T (Draft to Baseline)
Lwl (Waterline length)
Bwl (Waterline beam)
WSA (Wet surface area)
AT (Max cross sect area)
Aw (Waterplane area)
Cp (Prismatic Coefficient)
Cb (Block Coefficient)
Cm (Midship Coefficient)
Cwp (Waterplane coefficient)
KB (Centre of buoyancy height)
KG (Centre of gravity height)
BMt (Transverse metacentric radius)
BMl (Longitudinal metacentric radius)
GMt (Transverse metacentric height)
GMl (Longitudinal metacentric height)
KMt (Transverse Metacentre)
KMl (Longitudinal Metacentre)
TPc (Tons per cm Immersion)
MTc (Moment to change trim 1 cm)
RM (Restoring Moment) at
1 deg = GMt.Disp.sin(1)
7.007 t
6.836 m3
0.475 m
6.072 m
6.064 m
29.653 m2
1.961 m2
28.825 m2
0.574
0.391
0.681
0.783
0.307 m
1.325 m
9.717 m
9.640 m
8.698 m
8.621 m
10.02 m
9.946 m
0.295 t/cm
0.099 t cm
1.064 t cm
Figure 10. Heave RAO (Heading Sea, 180 ).
Figure 11. Pitch RAO (Heading Sea, 180 ).
SIMULATION
Due to the radial symmetry of the buoy, and with

the design weather conditions (which consider a wave
direction of 0 ) its DOF can be reduced from six to
three (pitch, surge and heave).
From the results in the previous section the equation
of the buoy dynamics can be defined as:
Figure 9. Righting lever, GZ (m).
introduced as input. Data have been obtained from an

oceanographic buoy located near the farm emplacement (Silleiro Buoy).
Figures 10 and 11 show the graphs corresponding to
the RAOs (Heave, Pitch and Roll RAOs) for a bow sea
(angle of incidence of the waves equal to 180 degrees).
This graphics have been drawn with data obtained
in Hydromax/Maxsurf and MorisonCalc Form (developed by Lognoter, and based in Morison Equation).
It can be also seen how the buoy heave motion has
a peak greater than unity at frequencies close to the
buoy natural one.This peak is due to resonance.A RAO
value greater than unity indicates that the amplitude of
the buoy response is greater than the wave amplitude
itself.
where M = the mass and added mass matrix, D = the

damping matrix, C = restoring matrix and fext = the
external forces. These matrixes are defined as:
583
offshore aquaculture sector to be consolidated and

become increasingly competitive.
In order to maximize the automation in these facilities, this work focusses in the development of a
buoy carrying all the automation and control systems
needed for the farm, as well as the necessary electric
power generation equipment.
The design criteria, buoy configuration, and mooring configuration are presented through an example
of application to an offshore fish farm placed at the
Galician region coast. We have chosen this example as a representative one because the hard weather
conditions present in Galician coasts made of vital
importance the question of the study of the behaviour
of this buoy in the sea, and its mooring arrangement.
For the conditions and characteristics of this farm,
a cylindrical buoy with 4 mooring points have
been designed. Both, hydrostatic and hydrodynamics
analysis of the buoy have been made, and its righting
lever and RAOs graphics have been obtained.
Figure 12. Times series for X displacement.
ACKNOWLEDGEMENTS
This work was partially funded by the Xunta de Galicia
through project 10REM007CT and by Feder founds
by the MICIIN through project ENE2010-20680C03-03.
REFERENCES
where A = added mass, APz = added inertia product

and AI = added inertia. The damping coefficients for
x and z displacements have been calculated as the
integral of drag of elementary shapes. The pitch damping have been estimated though the method of Ikeda
et al. (1978).
The external forces vector is calculated as the
composition of the following forces:
Wave forces
Wind forces (including wind turbine)
Marine current forces
Thus, a time depend simulator has been implemented

in order to simulate the buoy behaviour in a real condition. Then, the results of the simulator should be used
in order to improve the design and to develop life cycle
studies.
8
CONCLUSIONS
The ultimate goal of the work presented here is

to optimize fish production in offshore aquaculture
plants through the optimization of its technology.
This optimization aims to the reduction of costs and
the increasing of profitability, and thus will help the
Beaz-Paleo J.D. 2008. Ingeniera de la Acuicultura Marina:

Instalaciones de Peces en el mar. OESA (Observatorio
Espaol de Acuicultura) Technical Paper.
Cozjin, J.L. and Bunnik, T.H.J., 2004. Coupled Mooring
Analysis for a Deep Water CALM Buoy, Proceedings of
OMAE 2004, 23rd International Conference on Offshore
Mechanics and Artic Engineering.
Halwart M., Soto D. and Arthur J.R., (Eds.) 2007. Cage
aquaculture. Regional reviews and global overview.
FAO Fisheries and Aquaculture Technical Paper n. 498.
IALA, 1998. Recommendation E-107 on the designs of
normal moorings.
Ikeda, Y., Himeno, Y. and Tanaka, N., 1978. A Prediction
Method for Ship Roll Damping, Report No. 00405 of
Department of Naval Architecture, University of Osaka
Prefecture.
Irish J.D. and Boduch S.J. 2006. Aquaculture Feed Buoy
Control Part 1: System Controller. OCEANS 2006
Conference Proceedings.
Irish J.D. and Boduch S.J. 2006. Aquaculture Feed Buoy
Control Part 2: Telemetry, Data Handling and ShoreBased Control. OCEANS 2006 Conference Proceedings.
Plataforma Tecnologica Espaola de la Pesca y la Acuicultura
(PTEPA), 2009, Documento Visin 2020: Tendencias y
prioridades cientficas y tecnolgicas en el sector de la
pesca y la acuicultura. PTEPA Publication.
Wallinga, J.P., Neal R. Pettirew, N.R. and Irish, J.D. 2003.
The GoMOOs Moored Buoy Design, Proceedings of
OCEANS 2003.
584

Review and classification of wave energy converters

C. Guedes Soares, J. Bhattacharjee, M. Tello & L. Pietra
ABSTRACT: An overview of existing Wave Energy Converters (WECs) is presented, concentrating on the
main features of the devices dynamics, the power take off mechanisms and the mooring systems. The aim is to
identify the most promising technologies that are worth to be considered when planning the installation of wave
energy offshore. A common approach to classify WECs is based on their mode of operation, but it is recognized
that some devices have additional features that influence their performance and cost. The definition of a power
conversion chain is recognized to be a useful tool. Advantages and disadvantages of each class are outlined,
and concepts presenting distinguishing features are pointed out. Parameters are identified, on which to base an
evaluation and comparison between devices.
INTRODUCTION
Several concepts of Wave Energy Converters (WECs)

exist: some are well-proven by abundant literature, and
already implemented in working devices, others are
just at a preliminary design phase. Up to date, many
devices have been proposed, several built and tested
as models, and a few have been put in experimental
operation at sea.
In view of the diversity of technology available an
institution that may want to invest in the installation
of a wave energy infrastructure will need to compare
different designs, in terms of their technical performance and feasibility, and economic viability. This
paper takes this type of approach in looking at the
available technologies to identify the major principles
of their mode of operation so that it becomes easier to
compare their performance.
To identify the devices of interest, eventually closest
to commercial applicability, several are the aspects that
need to be considered and to serve as basis to classify
the existing devices, as several are the key aspects contributing to the technical and economic feasibility and
success of a project. The main aspects are the dynamics of the device, the principle according to which
it interacts and extracts energy from waves (PTO
Power Take Off) and the mooring and station keeping characteristics. Furthermore, aspects like building,
installation and maintenance are also important for
decision making.
Description of the devices dynamics gives information on the shape of moving and fixed parts of
the WEC, their interaction with waves, and indirectly
on their size, material and structural requirements, as
well as costs. Most importantly, mathematical models (including any control strategy implemented) will
point out upper limits in energy extraction efficiency.
Information on the Power Take Off system include

its operating principle, working fluid (if any), positioning (with respect to the machine, to shore and
surface), components, and therefore reliability, maintenance requirements, components and costs (e.g. if
off-shelf or custom-made). The knowledge of any
power-limiting strategy and the efficiency curve are
fundamental to determine the production curve of the
whole device.
The mooring system is subject to different requirements according to the mode(s) of motion of the WEC,
but also the machines position (energy flux, extreme
events), bathymetry and seafloor characteristics.
Different station keeping solutions will comply best
with different issues.
This paper adopts commonly used schemes,
which divide devices according to working principle,
mode(s) of motion, position relative to shore and surface, PTO system and stage of development. These
criteria were used in various review papers through
the years as for example in Thorpe (1999), Clment
et al. (2002), Falnes (2007), Drew et al. (2009) and
Falco (2010).
2
DEVICES DYNAMICS
There is a large variety of wave energy converters

that can be grouped based on their working principle, a concept that includes mode(s) of motion (i.e.
degrees of freedom) and orientation with respect to
wave direction as well as the type of PTO.
Energy conversion, from kinetic and potential in
wave motion to electric delivered to the grid, can be
divided into a series of subsequent steps, each operated in a sub-system. Generally speaking, these can
be combined in different ways: four main variations
585
Figure 1. Power conversion chains.
of this power conversion chain are found in existing or proposed devices (not considering the option
of having other final products than electricity, i.e.
typically, desalinated water): these are schematically
represented in Figure 1.
Each of the mentioned conversion steps can be
performed in more than one way, nonetheless a general scheme can be outlined, with a primary interface
representing the conversion operated by the part of
the device in direct contact with the waves (applying a damping force to them), that has in turn to be
referenced to either the seabed, the shore or another
component of the WEC, fixed or moving out of phase
from it; there can be intermediate transformations or
storage/accumulation of energy, and eventually the
conversion to electrical energy.
In the following, a review of the principles proposed so far is given, both the most common and the
unique ones, divided according to the primary conversion step they operate, with reference to Figure 1;
advantages and disadvantages common to all the
machines included in each of the classes are pointed
out; machines having peculiar features are individually
described.
2.1
Potential energy
2.1.1 Overtopping
These devices consist of a sloped structure (either
floating or fixed to shore), over which the waves topple; one or more reservoirs where the water is collected
and temporarily stored, create a static head of water;
and low-head hydro turbine(s) installed at the bottom
of the reservoir, through which the water flows out,
back to sea. These machines make use mainly of wellestablished technologies, thus reducing development
costs and risks; the only moving parts are the turbine(s)
and there are no components moving with the waves
(the device is not wave activated), thus minimizing
loadings and breaking risks, as well as reducing the
requirements on the structure and moorings (in floating devices). They are easily scalable without the need
for reconfiguration/tuning. The presence of a reservoir, acting as a buffer, helps smoothing the power
output.
One important device of this type is the Wave

Dragon, which consists of two wave reflectors focusing the incoming waves towards a ramp, a reservoir
for collecting the overtopping water and a number of
hydro turbines for converting the pressure head into
power. A prototype has been deployed in Denmark
an extensive measuring program has been carried out
(Kofoed et al. 2006).
The Seawave Slot-cone Generator (SSG), is an
extension of this concept to a multi caisson device,
which is installed on the coast with caissons at different
height. The water is collected in them through horizontal openings on the breakwater sloping wall and then
is run through a multistage turbine to produce electric
energy (Margheritini et al. 2009).
Another concept of a floating device, although of a
much smaller size is based on using the pitch motion
of a ship like floater. The water that overtops the device
is then collected in a tank (Buchner, 2011).
2.2
Pressure or kinetic energy
2.2.1 Oscillating water column (OWC)

Devices of this type are composed of a partly submerged, hollow structure (floating, bottom-standing
or shore-mounted), open below the water surface so
that it contains a column of water. This chamber can
have either vertical or sloped walls, to induce resonance of the water column. Air is trapped in a chamber
above the surface of the water column. As waves enter
and exit the collector, the water column moves up and
down and acts like a piston on the mass of air, pushing
it back and forth through a self-rectifying air turbine
(or a one-way turbine, if the flow is rectified), open to
the atmospheric pressure at the other end.
These devices have been among the first ones
to be proposed for the conversion of wave energy,
thus the literature available is abundant and many
studies have been dedicated both to the dimensioning of the air chamber and to the turbines. Still,
fixed-geometry, self-rectifying air turbines have a low
time-averaged efficiency (although peak efficiency is
relatively good); the proposed alternatives (impulse
wheel, variable pitch), while increasing the performance, are more expensive and delicate to operate
(Folley et al. 2006). Being the turbine the only moving
part in the machine, and being it not in direct contact
with water, issues related to maintenance and access,
and protection against extreme seas, are made easier;
still, requirements on blades and discs materials have
indeed to account for water and salt in the air column.
Several fixed OWC prototypes have been in operation in shore areas as in Toftestallen, Bergan, Norway
(1985), Sakata, Japan (1990), Vizhinjan, Kerala, India
(1990), Pico, Azores, Portugal (1999), Islay Island,
Scotland (2000). It is interesting to note the advantages
of incorporating the power plant into an existing breakwater, as done in Sakata, Japan. Another concept for
such option is due to Boccotti (2007), who also adopted
multichambers, as done by others (Thiruvenkatasamy
and Neelamani 1997).
586
Solutions of floating OWC devices have also been

proposed and various studies are available on the
Backward Bent Duct Buoy (BBDB), which has a duct
bent backward from the wave directions (e.g. Hong
et al., 2004). This allows the length of the water column to be as large as required for resonance to be
achieved at a relatively shallow draft.
The Mighty Whale (Hotta et al. 1996) is another
floating OWC which was developed in Japan and a
prototype of around 50 m length has been tested for
several years, with three chambers connected to Wells
turbines that drive electric generators.
Several studies have also been made for SPAR type
OWC (McCormick 1976, Korde 2000). These are a relatively long vertical tubes opened at both ends, which
have a heave motion. The pressure of the inner OWC
with respect to the floater pressurizes a air chamber
that is connected to a turbine. Solutions with multiple
tubes have also been considered (Lye et al. 2009).
2.2.2 Compressed air between subsequent waves
A unique device, the OWEL (Leybourne, et al. 2011),
makes use of this principle: wave crests enclose a
volume of air in a converging duct, thus compressing it and feeding it to a pressure reservoir and to a
unidirectional air turbine.
This is basically a pre-compressed Oscillating
Water Column. As in OWCs, here too the turbine is
the only moving part and its not in direct contact with
water, giving advantages in terms of maintenance and
damage prevention. The main structure is made of
concrete, of cheap construction, and its movements
are minimized, making for a stable platform to potentially accommodate wind turbines, with advantages in
terms of shared mooring, cable and deployment costs.
Requirements on moorings will be stricter. On the
other hand, the presence of a big floating structure
could interfere with other maritime activities, leading
to a more complex authorization process and fewer
available deployment areas.
2.2.3 Flexible membranes
Energy is extracted from the sea via a structure that
changes its shape/volume, acted upon by the incident
or travelling water waves (Chaplin et al. 2007). Such
structures are made of lightweight materials, reducing thus the transportation cost and mooring loads,
while increasing survivability. Manufacturing costs
could turn out to be high if special materials have to
be developed for the purpose.
2.3
Mechanical energy
2.3.1 Point absorbers

Are defined as wave energy devices whose characteristic length (typically, the diameter of a floater) is small
compared to the incident wave length. They are floating structures, whose wave-activated motion reacts
against a reference of some kind: either the sea floor or
another body with different motion characteristics (a
coupled body positioned so that it does not experience
wave force or a non-coupled body experiencing different exciting forces). Point absorbers can collect energy
from all directions and may be designed or controlled
so as to resonate that is, to move with larger amplitudes
than the waves themselves, in order to maximize the
amount of power captured. Their motion mode can be
heave, pitch or roll, or a combination of these. Thanks
to resonance, a point absorber is capable of collecting
more energy than that present in a wave front of its
same width. This same phenomenon, though, exposes
the machines and moorings to high and potentially
destructive loads, making over-engineering necessary.
Vantorre et al. (2004) have proposed a heaving point
absorber that moves with respect to a platform.
Point absorbers that are connected to the sea bottom are not the most frequent ones but reference can
be made to one taut-moored buoy concept that uses a
linear electric generator on the bottom (Eriksson et al.
2005, Leijon et al. 2006).
Elwood et al. (2010) have also developed a wave
energy conversion system that consists of a deep draft
spar and a taurus-shaped buoy having a saucer-shaped
profile. The outer buoy is free to heave relative to the
spar but is constrained in all other degrees of freedom
by a linear bearing system. Wave forces are transferred from the buoy to the spar using a contact-less
force transmission system. The wave forces imposed
on the spar along with the relative velocity between
the two floating bodies is converted into electricity by
a permanent magnet linear generator.
The systems that depend on the relative motion of
two bodies (Falnes 1999) are more frequent. Examples
are the Aquabuoy (Weinstein et al. 2004), the Wavebob (Weber et al. 2009), which are floating buoys that
have parts moving vertically. The Archimedes Wave
Swing (AWS) has a bottom fixed cylinder and oscillating one on the upper part, which responds to the
pressures of the waves as they pass. The system was
tested in Portugal in 2004 and its performance can be
improved with control strategies (Valrio et al. 2007).
Another device depends on its pitch motion is the
PS Frog MK.5, which is a floater that has a vertical
floater with a ballast tank hanging below (McCabe
et al. 2005). When it pitches there is a mass that moves
sideways above the sea level, providing the motion for
the PTO.
Installation of arrays of devices is often foreseen,
with consequent higher costs of deployment (and manufacturing, although economies of scale will occur) as
compared to a single, bigger unit; a positive issue is
that the failure or temporary shutoff of one machine
will not affect the overall production significantly. The
hydrodynamic interaction between devices within an
array is complex and it can have negative effects on
absorbed power of some of the devices in the array,
and positive on others.
Point absorbers have been among the first kind
of devices to be proposed and investigated, and
there are several studies on their interaction. Falnes
(1980) neglected the scattered wave potential under the
assumption that the device dimension is much smaller
587
than the wave length. McIver (1994) considered that

all scattered and radiated waves are approximated by a
plane wave, under the assumption that the device spacing is many wave lengths. Mavrakos and Kalofonos
(1997) adopted a multiple scattering method that deal
appropriately with the hydrodynamic interactions. An
overview of the various methods can be found in
Mavrakos and McIver (1997).
Solutions for arrays of linear generator converters can be found in Thorburn and Leijon (2006), for
OWCs in Falco (2002) and for overtopping devices
in Beels et al. (2010).
2.3.2 Pressure differential
These devices comprise floater(s) submerged a little
below the surface, or anchored to the bottom, being
activated in their heave/sway movements by the pressure variations associated with surface waves. Their
location below surface guarantees better survivability and no visual impact; on the other hand, access
for maintenance and installation/decommissioning is
more complex and costly.
2.3.3 Terminators
Structures that move at or near the water surface,
absorbing energy only in the direction of the incoming waves. The devices main dimension extends in the
direction normal to the predominant wave direction so
that, as waves arrive, the device restrains them (this
originates quite high forces on the device). Resonance
may be employed. A perfect terminator would create
waves exactly in anti-phase with those arriving at it,
thus extracting 100% of their energy; a real terminator reflects some energy, transmits some and extracts
the rest.
An example of this type is the famous Salter Duck,
which has undergone different design improvements
with time (Salter, 1993).
Other type of structures has one main platform
where one PTO is installed and then several small
floating point absorbers are deployed and their motion
will react with the platform. One example is the FO3,
which is based on a square floating structure with a
hydraulic PTO around which there is an array of 21
axisymmetric buoys (Lendenmann et al. 2007).
The Wave Star is made of two rectilinear arrays of
floaters located on both sides of a bottom mounted
steel structure aligned with the dominant wave direction. The swings of the buoys pump oil into the
hydraulic system. A prototype with 24m long has been
tested for some years connected to the grid (Marquis
et al. 2010). A similar system for wave energy absorption was adopted by Estefen et al. (2008), although in
this case it is fixed to a vertical breakwater. Other differences of this concept are that it uses a hyperbaric
chamber and pumps water through a Pelton turbine.
2.3.4 Attenuators
Are long floating structures oriented (or self-orienting,
thanks to slack moorings) parallel to the dominant
wave direction: this gives a smaller area normal to
wave direction (as compared to terminator devices),

and thus lower forces experienced. Relative pitch
movements between the devices hinged sections are
restrained to extract energy. The power absorption is
based on the pitch angle rather than on the height of
oscillation; absorption efficiency drops with increasing angles, thus being self-limited and safer in high
seas.
The McCabe Wave Pump is a two-body system
whose behaviour is based on pitch motion. This is
made of a central pontoon aimed at having very small
heave due to a damper plate. There is then a forward
and an aft pontoon, which rotate around the hinge in
the central pontoon. A 40 m long prototype has been
tested in Ireland in 1996 (McCormick et al. 1998).
Multi body systems like floating raft (Newman
1979, Haren and Mei 1982) have been considered but
have not been much explored recently. A derived concept is the Pelamis, which is based on making use of
the relative pitch motion between sections of a long
segmented cylinder (Retzler 2006, Henderson 2006).
Sea trials of these devices have been made in Scotland
in 2004 and in Portugal in 2008 for relatively short
periods.
2.3.5 Oscillating wave surge converters
These converters consist of pitching flap(s) anchored
and hinged to the sea floor, moved back and forth
by the surging movement of water particles close to
the bottom in shallow waters. This surge motion is
amplified in shallow water due to the shoaling effect
induced as the wave travels over the gradient of the
seabed. Their location closer to shore guarantees maximum loads to be reduced (and structural requirements
and costs with them) with reduced incident wave
power; the exploitable power is nevertheless often
only marginally lower than that found offshore (Henry
et al. 2010).
The kinetic energy of the moving flap is converted
into hydraulic energy using double-acting hydraulic
cylinders to pressurise water. High pressure water is
transmitted to the onshore hydroelectric plant through
conventional directionally drilled pipelines.
The Oyster full scale proof-of-concept device has
been at EMEC in Orkney in the summer of 2009 and an
improved design is on-going (Cameron, et al. 2010).
Another system with apparently similar features,
the Wave Roller is planned to be installed in Peniche,
Portugal in the near future but no details were found
in the technical literature.
2.3.6 Rotors
The water particles circular movement, or a component of it, is used to drive directly a water wheel
of different possible configurations (Darrieus, Wells,
propeller etc.). The components are off-the-shelf and
the technology is in principle well known, although
the application is new. There is no available body of
studies consistent enough to give a comprehensive
evaluation of such devices.
588
POWER TAKE OFF
3.2
A variety of Power Take Off System (PTO) arrangements have been proposed: these are not univocally
associated to a specific device or family, but can in
general be matched to several. Some PTOs developed by independent parts are proposed as off-theshelf components to device developers (Trident linear
generator, Artemis hydraulic systems, Sara Magnetohydrodynamic generator, Electroactive Polymer Artificial Muscle EPAM). There is no agreement on which
method is best, as each method has advantages and
disadvantages, and depending on device and site specifications different solutions will result to be better
than others. A brief description of different PTOs,
with their main advantages and drawbacks, follows.
A more extensive treatment can be found in Mynett
et al. (1979) and Salter et al. (2002).
Compressed fluids
3.2.1 High-pressure oil

Power transmission through hydraulic systems is generally mechanically quite inefficient, besides adding
one more conversion step; the system includes many
moving parts, all introducing losses and prone to failure; furthermore, possible leakages from the piping
are potentially harmful for health or the environment.
These systems are used because it is a well-known
technology, and they allow grouping several units to
work with a single PTO; also, big part of the machinery can be installed ashore for ease of installation and
maintenance.
3.2.2 Low-pressure water (sea or fresh, open or
closed loop)
With similar advantages to the previous ones, these
systems eliminate the risk of dangerous leakages;
sea water imposes requirements on materials of pumps
and (Pelton) turbines, due to corrosion/erosion issues.
3.1 Turbines
3.1.1 Air turbines (Wells, Denniss-Auld, airwave,
HydroAir)
These are used in Oscillating Water Column devices
and are in most cases of the Wells type, either single or twin rotor, with or without guide vanes. Other
kinds of turbines are also used and studied: namely
variable-pitch and impulse wheels. The Wells turbine
is characterized by an efficiency curve dropping quite
fast away from optimal rotational speed; variablepitch machines perform better but add in construction
complexity and are more prone to failure (requiring
therefore more maintenance). A review of this type of
turbines is given by Raghunathan (1995), while Curran
and Gato (1997) have compared the performance of
several types of turbines.
Another type of turbine is the self-rectifying
impulse turbine (Setoguchi et al. 2001), but it is not
clear if this type of turbine has advantages over the
Wells turbine.
An advantage with air turbines is that they are
away from, and not in direct contact with, the potentially corrosive salty water and potentially destructive high waves; also, they are easily accessible for
maintenance.
3.1.2 Low-head Kaplan (reaction)
A well-known technology used in overtopping devices;
it requires for an efficient control strategy to work at
optimal flows, if the head is variable. These axial flow
reaction turbines are used in mini-hydroelectric plants
and they operate with small heads of water of few
meters (Paish 2002).
3.1.3 High-head (impulse)
These impulse turbines operate with heads of tens
of meters, in some cases of Pelton type, are used in
some devices (Estefen et al. 2008, Henry et al. 2010)
where pressurized water is the power transmission
medium.
3.3
Direct drive
3.3.1 Linear generators

The rotor is substituted by a translator, moving in an
alternating fashion through the stator coils or permanent magnets, and directly coupled to the floater.
Thanks to the absence of gearbox, this is a more efficient solution than others, but efficiencies and costs of
power conditioning equipment (needed after the generator and before grid connection) have to be taken
into account; further cost factors relate to the fact that
physically big machines are needed to withstand the
high forces involved, and one generator is needed for
each point absorber (Rhinefrank et al. 2006).
3.3.2 Aegir dynamo
Transforms the alternating movement of a floater in
continuous rotational movement of a conventional
generator. The device is claimed to be commercial,
but no applications are documented yet.
3.3.3 Magnetohydrodynamic generator
The water particles movement induced by waves is
used to force a flow of conducting fluid (the sea water)
through a duct crossing magnetic field, thus inducing
electric current in coils wrapped around the duct. A
clear advantage is the absence of solid moving parts
(near zero maintenance), as well as the elimination of
intermediate conversion steps.
3.3.4 Electroactive polymer artificial muscle
This material has the characteristic of inducing a voltage between two electrodes connected to it, when
stretched or compressed. It can be incorporated in
a buoy to generate electricity directly from heave
motion. Its cheap and lightweight construction and
absence of intermediate conversion steps make it an
interesting solution especially for small scale applications, if satisfying power densities can be achieved.
589
For the moment, its application to wave power is still

at an initial phase (Chiba et al. 2009).
3.3.5 Contactless force transmission system
Transmission of the heaving movement from the
floater to the nut of a ball screw happens without
contact through permanent magnets; the movement of
the nut causes the ball screw to rotate, connected to a
conventional generator (Agamloh et al. 2008).
MOORING AND STATION KEEPING
The design of the mooring systems for floating structures has been made for some time within the offshore
industry, in particular as it has been moving towards
deep waters and relying more on floating than on fixed
structures.
As compared to the offshore industry, the WECs
tend to be smaller and to operate in shallower waters.
Furthermore, some types, as the point absorbers, have
part or the whole of the structure that is aimed at
operating in resonance and thus to undergo relatively
large displacements, which goes against one of the
common purposes of mooring systems, which are to
keep the structure in its original position (Johanning
et al. 2006).
Thus, mooring systems for this kind of WECs need
to be designed in such a way that they do not hamper the oscillatory motion of the device. Additionally,
the mooring system will have a dynamic response to
wave or wave group loading, and this may be critical when the WEC and its moorings are considered
together as a coupled system. For some WECs this
dynamic response, or the lack of it, is a key element in the mooring system design. Thus, just as
there is diversity among WEC configurations, there is
diversity among the associated mooring systems and
the requirements upon them. Therefore, design issues
on mooring systems have to include new constraints
with respect to what has already been addressed for
offshore structures.
However in some cases of larger non oscillating structures, such as floating overtopping or OWC
devices, mooring systems designed for oil platforms
can be applied without much difference. In these cases
use can be made of the available guidelines and regulations for mooring systems of offshore oil platforms,
published by various authorities around the world such
as the American Petroleum Institute or any of the Classification Societies. Some have specific documents for
mooring and in other cases they are incorporated in
the rules for the design, analysis and maintenance for
floating structures in the offshore oil and gas industry.
There are also in some cases specific rules for floating
fish farms and for mobile offshore units, which can
support the design of mooring systems for WECs.
Harris et al. (2004) listed standard requirements
for mooring lines applied to WECs, (mostly derived
from the offshore industry), including other issues
strictly related to the challenge of harnessing ocean
wave energy: the mooring system must be sufficiently

compliant to the environmental loading, to reduce the
forces acting on anchors and the device itself to a minimum; the mooring should not adversely affect the
devices efficiency, and if it is part of an active control
system it must also be designed dynamically as part of
the overall WEC system.
It is obvious that general conclusions regarding
mooring lines for all WECs cannot be drawn. However, based on the work of Harris et al. (2004) and the
studies of Fitzgerald and Bergdahl (2008) and Tello
et al. (2012), recommendations addressing the specific problem for floating WECs can be made, such as
the need for light materials (synthetic ropes) and the
use of spring buoys to avoid hampering the oscillatory
motion of devices.
Concerning the economic aspects, for a floating
device the cost of the system will be directly related
to meeting these requirements, including that of resistance to fatigue and abrasion loading, occurring over a
design life of thirty or more years (Harris et al. 2004).
DESIGN CONSIDERATIONS
Important information are conveyed by the position

of a device relative to shore (onshore, near shore and
offshore): although not directly related to the working
principle and modes/equations of motion, due to big
changes in energy flux and wave loads, the distance
from shore greatly influences the requirements on the
devices and their moorings, in terms of survivability
as well as average available power. Devices located
in similar positions will face similar conditions and
engineering issues.
Loosely speaking, to the shore-wise position a
surface-wise one can be associated, being that most
offshore devices are floating or near-surface, while
the near-shore ones are mostly bottom-mounted
(either piercing the surface or completely submerged;
although floating, near shore devices exist). There
are exceptions to this informal rule, and also shoremounted devices have to be included. Potentially,
almost all combinations of working principle and
position could take place; as a matter of fact, though,
these combinations are limited.
Each option about the position of a wave energy
conversion device carries advantages and disadvantages, some intrinsic and some relative to the technology being used. Waves offshore are known to carry
more power per meter of front, and this is theoretically
available for conversion to devices located there; on the
other hand, this also implies higher structural loads on
the device and on its mooring system (with consequent
over-engineering and increased material costs), and a
higher risk of damage in case of storms (indeed, some
devices include an actively or passively activated survival mode). Furthermore, it is important to remember
that the economically exploitable power is often lower
than the total one contained in the sea, as a significant
portion of it might not be available for extraction, if
590
carried by waves too big for the device to extract power

from them (Henry et al. 2010). Short-term variability
in sea conditions is higher in open sea than in shallow
waters, calling for faster responding control systems.
Finally, devices located further from shore are more
expensive to install and maintain, and the limitedness
of suitable weather windows to perform such activities
leads to decreased availability and missed incomes.
Also, longer transmission cables are more expensive
to buy and deploy, and imply higher losses.
An advantage of deploying machines far from shore,
especially if in farms and therefore taking up significant areas of sea surface, is the limited or null
visual impact, which can instead constitute a significant hurdle to near shore or onshore projects, unless
the machines are completely submerged or integrated
in breakwater structures (see following paragraph).
Issues related to interference with other marine activities (typically fishery and recreation industries, or
navigation routes) can arise both near and offshore,
and have to be evaluated individually for a specific
project and area.
Concerning the position of a device relative to the
surface, having it completely submerged has advantages in terms of avoided visual impact, limited interference with other activities, and protection against
extreme sea states; but involves more difficult and
costly deployment and maintenance operations (generally speaking, but there are exceptions where special
procedures have been designed to limit costs and
complexity).
There are other features common to a number of
devices, and it can give some insights to make comparisons between machines within such sub-groups.
For instance, it is noticed that some devices are
suitable for integration with breakwater structures
(SSG, WECA, S.D.E.); others can be used as floating platforms, or be otherwise integrated with offshore
wind turbines (Grampus, W2 Power, Wave Treader,
Poseidon, Wave Rotor). In both cases, there are advantages in terms of reduced (shared) manufacturing (of
platforms or other carrying structures), installation
(of cables, moorings and the devices themselves) and
transmission costs; in the case of integration with wind
turbines, a further advantage is the increased overall
availability and smoother (more constant) electricity
production.
Some machines can be deployed in farms
(arrays); this brings advantages in terms of cost of
(shared) transmission cables (and in some cases mooring systems), although potentially higher material and
manufacturing costs as compared to a single, bigger
machine of analogous rated power. It is recognized that
the hydrodynamic interaction between devices within
an array (specially point absorbers) could lead to
decreased performance in some of them, but increase
the absorption of others: more investigations on this
subject are needed and are being carried on, and specific evaluations need to be made on a case-by-case
basis, both based on the type of machines considered
and on the specific site and wave climate.
ECONOMIC CONSIDERATIONS
In addition to the technical differences between wave

converters, the choice of one for a specific location will
ultimately be governed by economic considerations
and thus it is necessary to be able to provide this type
of comparison between concepts.
One of the early treatments of this problem was by
Thorpe (1999), who adopted the concept of the cost of
energy associated with a device.
The process starts with the identification of the site
to be studied and a wave resource assessment for the
location is required, with the appropriate information.
The effective captured power depends on the capture efficiency of each converter, which is normally
described by a Power Matrix, which indicates for each
sea state the amount of energy that the device is able
to collect.
The power chain from the device to the grid needs
to be identified and this allows the estimation of the
maximum annual output. It is then necessary to determine the availability of the device, which will then
allow the estimation of the actual output.
For that annual output, the cost of electricity
depends on the capital costs and on the operational
and maintenance ones.
The capital costs depend on four major cost items
for any wave power scheme:
Generation Device (Fabrication of the structure)

Mechanical and electrical plant
Foundations or moorings
Transportation and installation
Electrical transmission to the grid
There are two main factors which make up the

annual running cost of any converter: repayment of
capital costs and payment of recurrent costs such as
insurance and maintenance. The amortisation of the
capital costs is normally considered over the complete
lifetime of the system using various discount rates.
Callaghan and Boud (2006) have also adopted the
concept of price of electricity and have used it to determine the cost associated to wave farms. They assessed
the relative importance of the various components of
capital cost as:
Structure 27%
Mechanical and electrical 49%
Mooring 5%
Installation 13%
Grid connection 4%
Project management 2%
For the operation and maintenance costs for a wave
farm the breakdown would be:
591
Licences 1%
Insurance14%
Monitoring 4%
Planned maintenance 29%
Unplanned maintenance 28%

Refit 24%
These cost estimates are obviously related to a given
size of farm and type of device, but at any rate they
can be indicative of the relative importance of the cost
items.
Callaghan and Boud (2006) have also introduced
the important concept that the cost of energy will
decrease with time due to the maturing of the technology. They suggest that this will be achieved through
four routes to cost reduction: concept design developments; detailed design optimisations; economies
of scale; and learning in production, construction,
installation, operation and maintenance.
Skoglund et al. (2010a,b) did not restrict their study
to wave energy and have broadened the use of economic considerations to the choice of the renewable
energy type and introduced the concept of Degree of
Utilization, which is crucial for the power that can
be obtained from the device. It depends on how the
device responds to weather conditions and also to the
availability of its PTO.
In addition to the predicted cost of electricity,
another important economic measure is the internal
rate of return for a project (IRR), which is the interest
rate received for an investment consisting of payments
(negative values) and income (positive values) that
occur at regular periods. This is calculated against the
income for which the electricity generated by a wave
power plant is sold (Menanteau et al. 2003). This will
dictate whether it is worthwhile to go ahead with an
investment in a wave converting system composed of
a number of wave converters.
Previsic et al. (2004) considered the concept of cost
of electricity and has studied how a project would
depend on estimates of future market value of the electricity and how it varies among regions, as well as other
aspects like financing and income tax regulations. In
fact, Leijon et al. (2003) have argued that it is the market value of the electricity that dictates the economic
viability of wave energy projects.
Finally approaches for assessing the project uncertainties need to be used as the lack of experience and
operational data on these systems imposes significant
uncertainties on the cost estimates and this needs to be
appropriately handled.
For a specific wave farm development, the choice
of the type of wave converter to be adopted can be
made with the approach provided by Stallard et al.
(2008) who used the concept of Data Envelopment
Analysis (Charnes et al., 1978), which had been
applied to a wide variety of fields, (e.g. Khouja 1995,
Ramanathan 2001). This allows the assessment of one
measure of efficiency of how one system converts
multiple inputs to multiple outputs and thus can be
used to compare the performance of different systems. When applied to a set of wave energy converters
it will give measures of their technical efficiency
despite the fact that converters are of different type
and require different input to produce electricity to
the grid.
CONCLUSIONS
A variety of wave energy systems are at several stages

of development, and it is still not clear whether there
will be a final winner; more likely, different solutions
will result to be better adapted to different conditions.
Meaningful ways to describe devices have been
developed, thanks to which the relative performance
of different subsystems can be evaluated.
It was noticed how some devices, although less
advanced, are worth special interest, for exhibiting
unique features, which make them more than just
a replication of already proposed working principles (this being especially true for flexible membrane
devices).
Eventually, the most meaningful comparison is
based on the cost of energy that a device is capable
of delivering in a specific installation site, or the Net
Present Value of a project. Nevertheless, this requires
the knowledge of site-related factors and economic
parameters that are prone to change significantly in
the future.
ACKNOWLEDGEMENTS
This work has been performed within the research
project WAVE CONVERTERS Hydrodynamic
analysis of wave energy converters with different
motion characteristics, which has been financed by
the Foundation for Science and Technology (Fundao
para a Cincia e a Tecnologia), from the Portuguese
Ministry of Science and Technology, under contract
The second author has been funded by the
(FCT, Fundao para a Cincia e a Tecnologia, Ministrio da Cincia, Tecnologia e Ensino Superior)
through a Post-Doctoral Fellowship under the contract
no. SFRH/BPD/47210/2008.
REFERENCES
Agamloh, E., A. Wallace, A. von Jouanne. A Novel DirectDrive Ocean Wave Energy Extraction Concept with Contactless Force Transmission System. Renewable Energy.
33 (2008), pp. 520529.
Beels, C., Troch, P., De Visch, K., Kofoed, J.P., De Backer,
G., 2010. Application of the time-dependent mild-slope
equations for the simulation of wake effects in the lee of a
farm of Wave Dragon wave energy converters, Renewable
Energy, 35, 16441661.
Boccotti P. 2007. Caisson breakwaters embodying an OWC
with a small opening. Part I: Theory. Ocean Eng; 34:
80619.
Buchner, B. 2011, Model tests and simulations on a wave
energy converter based on inverse offshore engineering,
Marine Technology and Engineering, C. Guedes Soares
et al. (Eds.), Taylor & Francis Group, London, pp. 349
367.
Callaghan, J. and Boud, R., 2006. The future marine energy
Technical report, The Carbon Trust.
592
Cameron, L., Doherty, R., Henry, A., Doherty, K., Vant Hoff,
J., Kaye D., Naylor, D., Bourdier, S., Whittaker, T., 2010,
Design of the Next Generation of the Oyster Wave Energy
Converter, Proc. 3rd International Conference on Ocean
Energy (ICOE2010), 6 October, Bilbao.
Chaplin, J.R., Farley, F.J.M., Prentice, M.E., Rainey, R.C.T.,
Rimmer, S.J. and Roach, A.T. 2007. Development of
the Anaconda all-rubber WEC. Proceeding of the 7th
European Wave and Tidal Energy Conference, Lisbon.
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the
efficiency of decision making units. European Journal of
Operations Research 2 (6), 429444.
Chiba, S., Waki, M., Kornbluh, R., Pelrine, R., 2009. Innovative Wave Power Generation System Using Electroactive
Polymer Artificial Muscles. Proc. IEEE Oceans 09,
Bremen, 1114 May 2009.
Clment A, McCullen P, Falco A, et al. 2002. Wave energy
in Europe: current status and perspectives.Renewable and
Sustainable Energy Reviews 6:40531.
Curran R, Gato LMC. 1997. The energy conversion performance of several types of Wells turbine designs. J Power
Energy, 211:13345.
Drew, B., Plummer, A. R. and Sahinkaya M.N. 2009. A
review of wave energy converter technology, J. Power and
Energy, 223, 887902.
Elwood, D., J. Prudell, C. Stillinger, A. Brown, S. Yim, A.
von Jouanne, T. Brekken, R. Paasch. 2010. Design, Construction, and Ocean Testing of a Taut- Moored Dual-Body
Wave Energy Converter with a Linear Generator Power
Take-off., Renewable Energy, 35, 348354.
Eriksson M, Isberg J, Leijon M. 2005. Hydrodynamic modelling of a direct drive wave energy converter. International Journal of Engineering Science; 43:137787.
Estefen SF, Esperana PTT, Ricarte E, Costa PR, Pinheiro MM, Clemente CH, et al. 2008. Experimental and
numerical studies of the wave energy hyperbaric device
for electricity production. Proc. of 27th Int. Conf. Offshore Mechanics Arctic Engineering, Estoril, Portugal;
paper no. OMAE2008-57891.
Falco, A.F.de O., 2002. Wave power absorption by a periodic linear array of oscillating water columns. Ocean
Engineering 29, 11631186.
Falco, A.F.de O., 2010. Wave energy utilization: A review
of the technologies. Renewable and Sustainable Energy
Reviews 14 (2010):899918.
Falnes, J., 1980. Radiation impedance matrix and optimum
power absorption for interacting oscillators in surface
waves. Applied Ocean Research 2, 7580.
Falnes J. 1999. Wave-energy conversion through relative
motion between two single-mode oscillating bodies.
J Offshore Mech Arctic Eng 121: 328.
Falnes J. 2007. A review of wave-energy extraction. Marine
Structures 20:18520.
Fitzgerald, J. & Bergdahl, L. 2008. Including moorings in the
assessment of a generic offshore wave energy converter:
A frequency domain approach.Marine Structures 21(1):
2346.
Folley, M., Curran, R., Whittaker, T. 2006. Comparison
of LIMPET contra-rotating wells turbine with theoretical and model test predictions. Ocean Engineering 33
10561069.
Haren, P. and Mei, C. C. 1982. An array of Hagen-Cockerell
wave power absorbers in head seas, Applied Ocean
Research, 4, 5156.
Harris, R.E., Johanning I., Wolfram J., 2004, Mooring systems for wave energy converters: a review of design
issues and choices.Proc. 3rd International Conference on
Marine Renewable Energy (MAREC). Blyth, UK.
Henderson, R., 2006. Design, simulation, and testing of a

novel hydraulic power take-off system for the Pelamis
wave energy converter, Renewable Energy 31, 271283.
Henry, A. Doherty, K. Cameron, L. Whittaker, T.J.T. and
Doherty, R. 2010. Advances in the design of the Oyster wave energy converter. Proc. Marine and Offshore
Renewable Energy Conference. RINA, London April
2010.
Hong, D. C., Hong, S. Y., and Hong, S. W., 2004 Numerical
Study on the Reverse Drift Force of Floating BBDB Wave
Energy Absorbers, Ocean Engineering, Vol. 31, No. 10,
pp. 12571294.
Hotta, H., Washio, Y., Yokozawa, H. Miyazaki, T 1996,
R&D on wave power device Mighty Whale, Renewable
Energy, 9, 12231226.
Johanning I, Smith GH, Wolfram J., 2006. Mooring design
approach for wave energy converters. Journal of Engineering for the Maritime Environment 220:15974.
Khouja, M., 1995. The use of data envelopment analysis for technology selection. Computers and Industrial
Engineering 28 (1), 123132.
Kofoed JP, Frigaard P, Friis-Madsen E, Srensen HC. 2006.
Prototype testing of the wave energy converter wave
dragon. Renewable Energy, 31, 181189.
Korde UA. 2000. A note on the hydrodynamics of a tail tube
buoy. Ocean Eng 27:14731484.
Leijon M, Bernhoff H, Berg M, Aagren O. 2003. Economical considerations of renewable electric energy production especially development of wave energy. Renewable
Energy, 28:12011209.
Leijon M, Danielsson O, Eriksson M, Thorburn K, Bernhoff
H, Isberg J, et al. 2006. An electrical approach to wave
energy conversion. Renewable Energy, 31(9):130919.
Lendenmann H, Strmsem K-C, Dai Pre M, Arshad W,
Leirbukt A, Tjensvoll G, Gulli T. 2007. Direct generation wave energy converters for optimized electrical
power production. Proceedings of 7th European Wave
Tidal Energy Conference.
Leybourne, M., Batten, W., Bahaj, A.S., Minns N., and
ONians J., 2011, Preliminary design of the OWEL
wave energy converter commercial demonstrator, Proc.
World Renewable Energy Congress, Linkoping, Sweden
2214222.
Lye JL, Brown DT, Johnson F. 2009. An investigation into
the non-linear effects resulting from air cushions in the
Orecon oscillating water column (OWC) device. Proc. of
28th Int. Conf. on Ocean Offshore Arctic Engineering,
Honolulu, Hawaii; paper no. OMAE2009-79115.
Margheritini, L., Vicinanza, D., Frigaard, P. 2009. SSG wave
energy converter: Design, reliability and hydraulic performance of an innovative overtopping device, Renewable
Energy, 34, 13711380.
Marquis, L., Kramer, M., Frigaard, P., 2010. First Power Production Results from the Wave Star Roshage Wave Energy
Converter, Proc. 3rd Int. Conf. on Ocean Energy (ICOE
2010), 6 October, Bilbao.
Mavrakos, S.A., Kalofonos, A., 1997. Power absorption by
arrays of interacting vertical axisymmetric wave-energy
devices. Journal of Offshore Mechanics and Artic Engineering 119, 244249.
Mavrakos, S.A., McIver, P., 1997. Comparison of methods
for computing hydro-dynamic characteristics of arrays of
wave power devices. Appl. Ocean Res. 19, 283291.
McCabe AP, Bradshaw A, Meadowcroft JAC, Aggidis G.
2005. Developments in the design of the PS Frog Mk 5
wave energy converter. Renewable Energy, 31:14151.
McCormick ME. 1976. A modified linear analysis of a waveenergy conversion buoy, Ocean Engineering 3:133144.
593
McCormick ME, Murthagh J, McCabe P. 1998. Largescale experimental study of a hinged-barge wave energy
conversion system. Proc. 3rd European Wave Energy
Conference; p. 21522.
McIver, P., 1994. Some hydrodynamic aspects of arrays
of wave-energy devices, Applied Ocean Research, 16,
6169.
Menanteau, P., Finon, D., Lamy, ML, 2003. Prices versus
quantities: choosing policies for promoting the development of renewable energy, Energy Policy, 31 799812.
Mynett, A. E., Serman, D. D. and Mei, C. C. 1979. Characteristics of Salters cam for extracting energy from ocean
waves, Appl. Ocean Res., 1, 13.
Newman, J.N. 1979. Absorption of wave energy by elongated
bodies. Applied Ocean Research 1(4): 189196.
Paish O. 2002. Small hydro power: technology and current
status. Renewable and Sustainable Energy Reviews 6:
53756.
Previsic, M., Bedard, R., Hagerman G., 2004. Offshore
Wave Energy Conversion Devices. E2I EPRI WP-004-USRev 1.
Raghunathan S. 1995. The Wells air turbine for wave energy
conversion. Prog Aerospace Sci 31:33586.
Ramanathan, R., 2001. Comparative risk assessment of
energy supply technologies: A data envelopment analysis
approach. Energy 26, 197203.
Retzler, C 2006. Measurements of the slow drift dynamics of a
model Pelamis wave energy converter, Renewable Energy
31 257269.
Rhinefrank K, Agamloh EB, von Jouanne A, Wallace AK,
Prudell J, Kimble K, et al. 2006. Novel ocean energy
permanent magnet linear generator buoy, Renewable
Energy 31:127998.
Salter, S.H. 1993. Changes to the 1981 Design of Spinebased Ducks, Proceedings of the First European Wave
Power Conference, Edinburgh, UK.
Salter, S. H., Taylor, JRM, and Caldwell, NJ, 2002. Power
conversion mechanisms for wave energy. Journal of
Engineering for the Maritime Environment 216:127.
Setoguchi T, Santhakumar S, Maeda H, Takao M, Kaneko
K. 2001. A review of impulse turbines for wave energy
conversion. Renewable Energy 23:26192.
Skoglund A, Leijon M, Rehn Al, Lindahl M, Waters R. 2010a.
On the physics of power, energy and economics of renewable electric energy sources - Part I, Renewable Energy 35
17291734.
Skoglund A, Leijon M, Rehn Al, Lindahl M, Waters R.
2010b. On the physics of power, energy and economics of
renewable electric energy sources Part II, Renewable
Energy 35 17351740.
Stallard, T. Rothschild, R. Aggidis, G.A. 2008. A comparative approach to the economic modeling of a large-scale
wave power scheme, European Journal of Operational
Research 185, 884898.
Tello, M., Bhattacharjee J., Guedes Soares C. 2012. Dynamics and hydrodynamics of ship like wave energy converters
in roll. Maritime Technology and Engineering, C. Guedes
Soares et al. (Eds.), Taylor & Francis Group, London
(this volume).
Thiruvenkatasamy, K. Neelamani, S. 1997 On the efficiency
of wave energy caissons in array, Applied Ocean Research,
19, 6172.
Thorburn K, Leijon M. 2006, Farm size comparison with
analytical model of linear generator wave energy converters. Ocean Engineering 34:90816.
Thorpe, T.W., 1999. A brief review of wave energy. UK
Department of Trade and Industry. Report ETSU-R120.
Valrio D, Beiro P, S da Costa J. 2007, Optimisation of
wave energy extraction with the Archimedes Wave Swing.
Ocean Engineering; 34(17-18):233044.
Vantorre M, Banasiak R, Verhoeven R. 2004. Modelling of
hydraulic performance and wave energy extraction by a
point absorber in heave. Applied Ocean Research, 26:
672.
Weber J, Mouwen F, Parrish A, Robertson D. 2009.
Wavebobresearch & development network and tools in
the context of systems engineering. In: Proceedings of 8th
European Wave Tidal Energy Conference; pp. 416420.
Weinstein, A., Fredrikson, G., Parks, M. J., Nielsen, K., 2004,
Aquabuoy, the offshore wave energy converter numerical
modeling and optimization Proc. MTTS/IEEE TechnoOcean 04 Conf., Kobe, Japan, vol. 4, pp. 18541859.
REFERENCES FROM THE WEB

Aegir dynamo: [www.oceannavitas.com].
Artemis hydraulics: [www.artemisip.com].
MHD generator:[www.sara.com /RAE/ocean_wave.htm].
Poseidon: [www.poseidonorgan.com/].
Seawave Slotcone Generator: [www.waveenergy.no].
S.D.E.: [www.sde.co.il].
Trident linear generator: [www.tridentenergy.co.uk].
W2 Power : [www.pelagicpower.no/].
Waveplane: [www.waveplane.com].
Wave Treader: [www.greenoceanenergy.com/].
Wave Rotor: [www.ecofys.com].
594

Analysis of water depth effects on the wave power resource

and the energy captured by a wave energy converter
ABSTRACT: Due to economic reasons related to the mooring system characteristics and to the electrical grid
connection, most floating wave energy converters under development are planned for operation in relatively
shallow water, typically 30 m to 70 m. The limited water depth reduces the wave energy resource as compared to
offshore deep water. This paper reviews the existing simplified methods for wave energy spectrum modification
as the water depth reduces. The TMA method (Bouws, 1985) is implemented and applied with the aim of
assessing both the reduction in the available wave power resource and the energy converted by a floating wave
energy converter. This method is based on the hypothesis that the similarity principle for the saturation range of
shallow-water spectrum (Kitaigororodskii, 1975) is not restricted to the equilibrium range, but is valid across
the entire spectrum. This hypothesis has been proven, thus demonstrating that a finite depth spectrum can be
obtained by simply multiplying the corresponding deep water spectrum by a function of the angular frequency.
The Pelamis machine and the wave climatology of Figueira da Foz (Portugal) are used as a case study.
1
INTRODUCTION
The interest in extracting energy from renewable

resources has been growing in the last years, thus
bringing to the development of a wide range of wave
energy converters, most of which are intended to be
installed in relatively shallow water. This characteristic of the operational site brings the attention on the
necessity to have some knowledge of the spectrum in
shallow water, in order to optimize the power take off
and to better design both the structure and the mooring
system of the device. In fact the existing sea spectra
have been derived for deep water, however the waves
modify their characteristics as they approach shallow
water.
The most accurate method to assess the near shore
seastate characteristics consists on applying wave
models which propagate the waves in time and space
accounting for the wave generation by wind action,
refraction and frequency shifting due to depth effects
and due to currents, nonlinear wave-wave interactions, dissipation effects due to whitecapping and wave
breaking and diffraction effects. The free software
SWAN is probably the most well-known wave model
computer code (Booij et al., 1999). When reliable wave
measurements are not available, wave propagation
models are in fact recommended to assess site specific
characteristics prior to the design and installation of a
wave energy converter.
Wave propagation models have been used to assess
the wave energy resource near the Portuguese coast by
Pontes (2005) and Rusu and Guedes Soares (2009),
however the objectives of these studies were different
from the objectives of the present work therefore the

information produced is not adequate for the present
study. Pontes presents a near shore wave energy Atlas
for the Portuguese coast with comprehensive information regarding the wave characteristics for 78 points
at 20 m water depth. Although the information is site
specific and valuable, we are interested in the (average) evolution of the of the wave characteristics as the
water depth decreases including the spectral shape so
that the performance of the wave energy converter can
be assessed. Rusu and Guedes Soares (2009) present
spatial distributions of the wave energy for a small set
of wave conditions. The results are presented for large
areas without information regarding the evolution of
wave characteristics and related spectra for the water
depth range of interest for the present study (between
100 m and 20 m of water depth).
The objective of the present work is to obtain a
general overview of the energy resource reduction
as the water depth decreases and of the reduction of
the energy production by a wave energy converter.
The later aspect is interesting since, on one hand it
is expected that the largest reduction of the resource
occurs for the higher seastates, but on the other hand
the power production of wave energy converters is usually limited to some level (nominal power) meaning
that above a specific value of the wave heights the
converted power does not increase. For this reason it
might happen that the reduction of the energy resource
is not important for the annual energy production.
With the mentioned objective in mind, it seems reasonable to use a robust, simple and general method
to represent the wave spectrum modification as the
595
water depth decreases. The method has been proposed

by Bouws et al. (1985) based on the analysis of the
results from a huge campaign of field measurements.
This method has the additional advantage of representing average bottom topography conditions, which
is interesting given the objective presented before.
Besides the conclusions presented in the paper
regarding the reduction of energy resource and production, the proposed method can be easily applied
for the initial cost/benefit studies (meaning the preliminary design of the device and the predictions of
annual energy production).
Regarding the simplified method, Bouws et al.
(1985) proposed transformation of the off-shore wave
spectra to a finite depth site by a model for spectral wave-energy transfer. This spectrum shape, named
TMA, is a transformation of the JONSWAP off-shore
one and is considered in this paper as the most realistic
spectrum to point out the loss of energy resource and
energy production when moving from deep water to
shallower waters.
The aim of the first two sections of the paper is
to review the standard off shore spectra and the theory
that brought to the definition of theTMA spectra, while
the third section investigates the finite depth effects on
the energy resource of Figueira da Foz (West coast of
Portugal) and on the energy produced by a converter.
the wave generation physics should consist not only of

wind source terms but also of some other mechanism.
In this context Phillips postulated that such mechanism in waves would be wave breaking, which with its
strong local effect on waves would fix a universal limiting value to the energy densities within a spectrum. As
a consequence of this theory a saturated region of spectral energy densities, called equilibrium range of the
spectrum, is assumed to exist in some region from near
the peak to frequencies sufficiently high that viscous
effect would begin to be significant.
On dimensional grounds Phillips (1958) suggested
the following expression for the equilibrium range of
the spectrum for a fully developed sea in deep water:
where E(f ) is the spectral energy density in unit length

squared per hertz and was understood to be a universal (dimensionless) constant approximately equal
to 0.0081.
On the basis of Phillips theory Kitaigorodskii
(1962) extended the similarity arguments of Phillips to
distinct regions throughout the entire spectrum where
different mechanisms might be of dominant importance. Pierson and Moskowitz (1964) proposed a new
empirically developed spectral form based on the similarity theory of Kitaigorodskii and on measured data
in North Atlantic:
2 A REVIEW ON THE WIND-WAVE

SPECTRUM MODELS
The theory of wave generation has had a long and rich
history. Beginning with the works of Kelvin (1887) and
Helmholtz (1888) in the 1800s many scientists have
focused their studies in defining various forms of water
wave motion and interaction with the wind. However,
it was not until the World War II that organized wave
prediction models were formulated. The first models
were based on the idea that interrelationships among
dimensionless wave parameters will be governed by
universal laws. In this context the most important law
was the fetch-growth law, widely studied by Sverdrup,
Munk and Bretscheneider, which suggested that a fully
developed wave height would evolve under the action
of the wind.
Later in the 1950s, scientist, mathematics and
engineering began to recognize that the wave generation process was best described as a spectral phenomenon (e.g. Pierson, Neumann and James, 1955).
Re-examining the previous ideas of wave generation
and growth Phillips (1958) and Miles (1957) advanced
two theories that constituted the basis of the understanding of wave generation physics for many years.
Phillips theory predicted linear wave growth and was
believed to control the early stages of wave growth
while Miles theory predicted the exponential growth
and was believe to control the major portion of wave
growth observed in nature. From basic concepts of
energy conservation and the fact that wave do attain
limiting fully developed wave heights, it came out that
where Ep (f ) is the Phillips equilibrium range formula

(eq. 1) while PM is the shape function derived from
measurements at the sea and fp is the peak frequency:
Since the Pierson-Moskowitz spectrum (or PM spectrum) is assumed to represent fully-developed conditions (both the fetch and the duration are infinite) in
deep water, the peak frequency can depend only on
the wind speed (fPM = 0.13g/U10 ) (Holthuijsen, 2007,
pp. 159).
The relation between the peak frequency and the
significant wave height has the form:
while the relation between the peak frequency and the

mean zero-up-corssing frequency is:
Based on the concepts of spectral wave growth due

to wind inputs via MilesPhilips mechanisms and on
the limiting Pierson-Moskowitz spectral form, firstgeneration (1G) wave models were developed in the
596
United States. This models recognized that wave in

nature are not only made up of infinite (continuous)
sum of infinitesimal wave components at different frequencies but that each frequency component is made
up of an infinite (continuous) sum of components
traveling in different directions. It was clear how this
behavior cannot be modeled properly in terms of significant wave height, bringing the development of
these models to include all wave components in a
spectrum individually (discrete-spectral models).
In the late 1960s evidence of spectral behavior
began to emerge which suggested that the equilibrium
range in wave spectra did not have universal value for
, instead it was observed that the energy scale () varied as a function of nondimensional fetch (Mitsuyasu,
1968). The theoretical foundation to explain this discrepancy already existed: Hasselmann (1961) showed
using minimal assumptions that wave in nature should
interact with each other in such a way as to spread
energy throughout a spectrum. This wave-wave interaction was finally confirmed by Hasselmann et al.
(1973), thanks to the extensive collection of data set
in the context of the Join North Sea Wave Project
(JONSWAP, 1968-1969). They found out that the
Phillips had a clear relationship with the nondimensional fetch and that the wave spectra had a tendency to
be more peaked than the Pierson-Moskowitz spectrum
during active generation.
The impossibility of explaining these phenomena in
terms of first-generation supported the development of
the second-generation wave models, which included
the nonlinear interactions between waves. These
models are called parametrical models (Hasselmann
el al., 1976).
A new modified spectral shape was then proposed,
named JONSWAP spectrum (Hasselmann et al., 1973)
from the name of the experiment. The spectrum
represents developing sea states with fetch limitation:
The peak-width parameter sharpens the spectral

peak, but has no effect on the other parts of the spectrum. The average values of the and parameters
in the JONSWAP data set were found to be = 3.3,
a = 0.07, b = 0.09. More over Hasselmann (1973,
1976) noticed a slight dependence on the fetch on the
peak frequency and on the equilibrium range parame2
ter () and used the dimensionless fetch (gX /U10
) to
create the following empirical relationships:
JONSWAP spectrum can also be defined in terms

of significant wave height, peak period and peakness
factor (Gachet and Kherian, 2008):
where:
Hs is the significant wave height,
p = 2fp is the angular spectral peak frequency,
A = 1 0.287 ln() is a normalizing factor.
Approximate expressions for the significant wave
height and peak period, as functions of the peak-shape
parameter and the mean zero-up-crossing period Tz
were numerically found (see Tucker 1991, Chakrabarti
1984).
for = 1 from eq. 12 and 13.b
where:
: dimensionless spectral width parameter, with
value a for f < fp and b for f > fp to account the
slightly different width on the two sides of the spectral
peak;
: peak-enhancement factor (random variable,
approximately Gaussian, with mean 3.3 and variance
0.62);
fp : dimensionless peak frequency
which gives an error of less than 1% for Hs , and less

than 0,1% for Tz , compared to P-M.
For = 3.3:
If no particular values are given for the peak shape

parameter , the following value may be applied:
: scale parameter (Lewis and Allos, 1990) defined

as follows
where TP is in seconds and HS is in meters.
597
It is easy to note how all these spectra show a

characteristic development. With growing energy the
rather sharp peak migrates toward lower frequencies
(Bouws et al., 1985). Whereas JONSWAP and PM
spectra equilibrium range follows the f5 shape proposed by Phillips, more recent measurements show the
decrease may be better described by a negative fourth
power frequency (e.g. Toba, 1973). In the later years
Resio and Perrie (1989) determined that, although certain spectral growth characteristics were somewhat
different between the f4 and f5 formulations, the
basic energy-growth equation were quite similar for
the two formulations. Further studies of the high frequency tail of measured spectra were presented by
Prevosto et al. (1996) and Rodriguez et al. (1999),
while the uncertainty of the high frequency tail slope of
wave spectra was examined by Rodriguez and Guedes
Soares (1999).
Finally in the early 1980s the third-generation wave
model class was born (Hasselmann et al., 1985). These
models explicitly represent all the physics relevant for
the development of the sea state in two dimensions.
If compared with the second-generation wave models the third-generation models use a more detailed
parameterization of the nonlinear wave-wave interaction source terms and relax most of the constraints
on spectral shape in simulating wave growth. Some
tuning parameters are however introduced, in order to
make the spectral distribution and migration of energy
resemble theoretical and experimental results (Komen
et al., 1984). This procedure implies the computation of five-dimensional Boltzmann integrals. Exact
computation of these expressions demands much computing time; therefore more efficient numerical methods have been proposed (Hasselmann & Hasselmann
1985). The third generation WAM model (Komen
et al., 1994) uses an exact non-linear transfer source
function parameterization in order to save computer
time (Caetano and Innocentini, 2003). Improvements
in this model include two-way coupling between wind
and waves, assimilation of satellite wave data, and
medium-range operational forecasting (Gmez and
Carretero, 1997).
3 A REVIEW ON THE WIND WAVE
SPECTRUM TRANSFORMATION IN
FINITE DEPTH WATER
If, on one hand it is possible to assert that in the early
70s the process in understanding the wave-growth in
deep water was already well developed, on the other
hand no theories were still developed to describe the
complicating effect of variable water depth and dissipation of wave energy due to the existence of a bottom
boundary layer.
3.1
finite constant depth. He observed that in idealized

shallow-water conditions, the wave spectrum evolves
essentially as in deep water: from the high frequencies to lower frequencies while the spectral density
relative to the peak period and so the area under
the spectrum increases. But in contrast to the situation in deep water, the spectrum does not retain its
shape along the fetch. Kitaigorodskii (1975) demonstrated, supported by observations, that if one adopts
the wave number (k = 2/L) form as universal, it can
then be shown that for shallow water the equilibrium range in the spectrum of wind-generated surface
waves is E(); 3 instead of E(); 5 adopted in
deep water (Kitaigorodskii et al., 1975). The saturation
range spectrum in finite water depth was obtained by
multiplying that in deep water (proposed by Phillips,
1958) by a factor:
where d is the water depth, H = d 1/2 /g 1/2 , and

k (H ) is an universal nondimensional function
expressed by:
This function can also be written (assuming that the

equilibrium range is determined by spectral components for which the long-wave approximation is
valid the phase speed is c = /k and ground velocity
cg = d/dk) as:
where n is the ratio of group velocity over phase

velocity (Holthuijsen, 2007):
Since 0 kd and therefore 0 2kd/sinh(2kd) 1,

this expression for n shows that n varies between
n = 1/2 (deep water) and n = 1 (very shallow water).
This implies that the speed of an individual wave
(the phase speed) is always larger than or equal to the
speed of the group: c cg .
This relation can be converted into angular frequency trough the dispersion relation:
The use of an iteration process yields to the approximate result:
Kitaigorodskii
The first to approach this topic was Kitaigorodskii

(1975) who reexamined Phillips equilibrium range
concept for deep-water and extended it to water of
598
Figure 1. The universal dimensionless function K (K )

(Kitaigorodskii, 1975).
The depth function K can also be written as:
3.2 The TMA spectrum, Bouws et al.

Since the Kitaigorodskii expression suggest a similarity principle for the saturation range of shallow-water
spectrum analogous to the deep-water case, Bouws
et al. (1985) hypothesized that a shallow-water self
similar shape may be obtained by replacing the Ep (f )
in the JONSWAP expression of the spectrum (eq. 6)
with Ek (f , d), yielding:
The principle hypothesis used to derive ETMA is that

the Kitaigorodskii scaling parameter is not restricted
just to the saturation range but is valid across the
entire spectrum (Bouws et al., 1985). Several hundreds
of examples where analyzed by Bouws to prove this
hypothesis reaching the conclusion that TMA spectrum fits wind wave spectra in finite water depth very
well.
As waves move from deep water to shallow water,
the f 5 -shape slowly replaces the f 3 -shape, starting
at the lower frequencies (where the depth effects take
hold first). Apparently a universal f n - tail of the spectrum would be valid only either in deep water (with a
frequency-independent value of n, e.g., n = 5) or in
very shallow water (e.g., n = 3).
This model was considered as suitable for predicting
developing wind waves in any water depth by Hughes
(1984) (form deep to shallow and even in the surf
zone) and by Guedes Soares and Caires (1995). However Goda (1990) argued: the TMA spectral model is
intended for use in wave hindcasting and forecasting
in water of finite depth. Various mechanisms of wave
attenuation by bottom friction, percolation, breaking,
and others are supposed to be included in the function
K (K ) implicitly. Use of TMA model in general wave
Figure 2. Transformation of the deep-water JONSWAP

spectrum in the shallow-water TMA spectrum (Holthuijsen,
2007).
transformation problems in shallow water should be

made with caution, because the model is essentially
for wind waves at the growing stage. (R. Silvester,
1997).
4
CASE STUDY
4.1 The effect of water depth in Figueira da Foz

incident wave power resource
Figueira da Foz scatter wave climate diagram (tab. 1,
see Guedes Soares et al. 1988) was taken by the authors
as case study to analyze the effect of the finite depth
on the energy resource. The choice of this site is of particular interest because of its proximity to the 320 km2
Portuguese pilot zone (located in So Pedro de Moel),
which will be an important testing and demonstrating
area for renewable energy converters in the next years.
The scatter diagram is the result of the statistical
analysis carried out by the Portuguese Hydrographic
Institute. The wave records obtained from a waverider
station located at 89 m depth. Considering the average
zero-up-crossing period 7.829 s and applying the linear
theory to a first approximation the wave length can
be assumed to be L0 = 1.56T02 = 95.62 m. This result
points out that it is possible to refer to the data from
the buoy as deep water (d/L = 89/95.62 > 1/2, deep
water boundary condition).
599
Table 1.
Deep water scatter wave climte diagram of Figueira da Foz.
Table 2.
Deep water power resource matrix [kW/m].
Table 3.
50 m depth power resource matrix [kW/m].
On the basis of the above observation, for each sea

state in the scatter diagram (defined by a significant
wave height and by a zero up-crossing period) a
JONSWAP spectrum was constructed to define the
off-shore power spectral density distribution. Later,
by applying the TMA method the spectra were
transformed to represent the energy density in 50 m

and 30 m depth, chosen as representative depths of
WECs operational sites.
In table 2, 3 and 4 the wave power flux (kW/m of
wave crest width) for each sea state and water-depth
(deep water, 50 m and 30 m) is given. The values are
600
Table 4.
30 m depth power resource matrix [kW/m].
Table 5. Average incident power reduction per year.
Deep water
50 m
30 m
Power [kW/m]
% lost incident power/year
29
21
15
0.0%
26.89%
47.15%
cg is the group velocity [m/s]

E(, d) is the power spectral density [m2 /Hz].
The values obtained by this integration were compared with the approximate formula for wave power
density, based on statistically derived sea state parameters [IEA, 2003]:
Figure 3a & b. Significant wave height transformation due

to the reduction of water depth.
calculated by using trapezoidal rule to integrate the

following expression:
where
is the water density [kg/m3 ]
g is the gravity acceleration [m/s2 ]
and they result to be similar.

These tables (tab. 2, 3 and 4) give a first insight in the
quantity of energy flux lost by passing from deep water
to shallower waters. What comes out is that the TMA
transformation affects more significantly the sea states
with higher period than those whit small periods. In
figure 3.a and 3.b two representative significant wave
highs are plotted with their transformation from deep
water to 50 and 30 meters depth. It can be noticed how
considering sea states having the same deep water Hs
the reduction passing to shallow waters is higher for
those with higher periods.
The total annual incident power decrement due to
finite depth is schematized in table 5. These values are calculated by summing all the sea states
power weighted with their yearly probability of
occurrence.
601
Table 6.
50 m depth wave climate scatter diagram.
Table 7.
30 m depth wave climate scatter diagram.
Table 8. Total capture energy reduction per year.
Deep Water
50 m
30 m
Caputure Energy
[GWh/year]
% Energy
captured
% Energy
lost
0.99
0.87
0.67
100.00%
87.47%
67.76%
0.00%
12.53%
32.24%
the amount of energy lost is confirmed by the data

published by Folley (2006) relative to the Western
coast of Scotland.
4.2 The effect of finite water depth in the Pelamis
power capture
The deep water incident power (29 kW/m) matches

pretty well the results obtained by Pontes (2005) thus
confirming that the JONSWAP spectrum ( = 3.3,
a = 0.07, b = 0.09) correctly describe the deep water
power spectral density of Figueira da Foz. Moreover
In this section Pelamis (www.pelamiswave.com) is

taken as an example of floating device for wave energy
extraction. The authors aim to investigate the effect
on its Power Matrix, and on the power capture, of
finite water depth installation. This matrix shows the
power generated by the device in a range of sea spectra
defined by significant wave height and energy period
(Te = 1.2Tz ).
602
Table 9.
Table 10.
Energy captured by the Pelamis in deep water for each sea state in one year [kWh].
Energy captured by the Pelamis in 50 m depth water for each sea state in one year [kWh].
A bilinear interpolation is applied to the Pleamis

power matrix in order to adapt it to the zero up-crossing
periods and significant heights of Figueira da Foz
scatter diagram.
The TMA transformation of each of the deep water
spectra in the scatter diagram yields to a reduction
of its significant wave height (while the peak period
remains the same as in deep water), and by consequence to an alteration of the probability of occurrence
of each Hs,Tz combination. Based on this observation,
two different wave climate scatter diagrams are constructed, relative to 50 m and 30 m depth, respectively
(tab. 6 and tab. 7). Due to the water depth decrement, which implies the action of bottom friction,
percolation and breaking waves, on one hand the
sea states having the highest significant are no more
present in the diagram and on the other the probability
of occurrence of the sea states having lower heights is
increased.
By multiplying these new scatters by the deep water
Pelamis power matrix and by the hours in one year it is
possible to know the annual amount of energy captured
by each sea state and to compare it with the one in deep
water.
Table 9, 10 and 11 represent the capture energy scatter for the three depths, respectively. The results clearly
indicate a decrement of the energy captured especially,

as expected, at high periods and heights.
Passing from deep water to 50 m 100% of the of the
energy lost is for periods higher than 7.75s, while in
30 m depth 19 % is lost for periods lower than 7.75s
and 81% for periods higher than 7.75s.
In conclusion table 8 plots the total amount of
energy lost in one year by considering the Pelamis
operating in 50 m and 30 m depth.
This result partially explain the choice of installing
devices in 50 m depth: it is easy to note how the incident wave energy decrease significantly from deep
water to finite depth but the decrement in terms of
capture energy is remarkable smaller, especially passing from deep to 50 m depth waters. In this instance the
loss of capture energy can be compensated by lower
installation and maintenance costs, by a smaller power
loss in the cables and by a less damaging wave climate,
which guarantee a longer life to the structure.
5
CONCLUSIONS
Most of the recently proposed wave energy converters

are planned to be installed in relatively shallow water
depths where the wave energy resource is reduced
603
Table 11.
Energy captured by the Pelamis in 30 m depth water for each sea state in one year [kWh].
compared to deep water. The present work uses a

simplified method to represent the wave spectrum
modification as the water depth reduces and present
an analysis of the energy resource reduction, as well
as the impact on the energy produced by a floating
wave energy converter. The case study uses the waves
statistics from the West coast of Portugal and the power
matrix of the Pelamis machine.
It is concluded that there is a significant reduction
on the wave energy resource: compared to deep water,
the energy resource reduces about 26% at 50m and
47% at 30m. The impact on the annual energy production of Pelamis is not as large, because the capability
of the machine is limited by the nominal power therefore the most energetic deep water seastates would not
be used at the full potential, but still the reduction of
produced energy is 13% at 50m and 32% at 30m.
The procedure presented can be useful for a preliminary cost benefit analysis of the best water depth
to install a floating wave energy converter. For the
final design and accurate assessment of the system
site specific wave data should be used.
ACKNOWLEDGMENTS
The work was performed in the scope of the project
Design of mooring systems for floating wave energy
converters, which is financed by Fundao para
a Cincia e a Tecnologia (FCT) under contract
REFERENCES
Bergdahl, L., 2009. Comparison of measured shallow-water
wave spectra with theoretical spectra, Proceedings of
the 8th European Wave and Tidal Energy Conference,
Uppsala, Sweden.
Booij, N.; Ris, R.C.; Holthuijsen, L.H., 1999. A thirdgeneration wave model for coastal regions 1. Model
description and validation. Journal of Geophysical
Research 104(C4), 76497666.
Bouws, E., Gunther, H., Rosenthal, W. et al., 1985. Similarity of the wind wave spectrum in finite depth water. 1.
Spectral form. J.Geophy& Re., 90: 975986.
Caetano, E. and Innocentini, V., 2003. An improved second
generation wave model. Braz. j. oceanogr. 51: 121.
Chakrabarti, S. K., 1984. Technical note on the formulation
of the JONSWAP spectrum, Applied Wave Research 6(3):
175176.
Folley, M., T.J.T. Whittaker, A. Henry, 2006. The effect of
water depth on the performance of a small surging wave
energy converter, Ocean Engineering, 34: 12651274.
Gachet, M. and J. Kherian, 2008. Impact of linearization
of bilge keel damping on the early assessment of vessel
operability, Proceedings of the ASME 27th International
Conference OMAE
Goda, Y., 1990. Random waves and spectra. Handbook of
coastal and ocean engineering, J.B. Herbich. Houston,
Texas: Gulf Pubblishing Co., 1: 175212.
Gmez, M., and Carretero, J.C., 1997. A Two-ay Nesting Procedure for the WAM model: Application to the Spanish
Coast, J. Offshore Mechanics and Arctic Engineering,
119:2024.
Guedes Soares, C. and Caires, S. Changes in Spectral Shape
Due to the Effect of Finite Water Depth. Proceedings of
the 14th International Conference on Offshore Mechanics
andArctic Engineering (OMAE); NewYork. Copenhagen,
Denmark: ASME; 1995: 547556.
Guedes Soares, C.; Lopes, L. C., and Costa, M. Wave Climate
Modelling for Engineering Purposes. Computer Modelling in Ocean Engineering. Rotterdam: A.A. Balkema
Pub.; 1988; pp. 169175.
Hasselmann K. 1961. On the non-linear energy transfer in a
gravity-wave spectrum Part 1. General theory. Journal of
Fluid Mechanics 12 (4): 481500.
Hasselmann K., T.P. Barnett, E. Bouws, H. Carlson, D.E.
Cartwright, K. Enke, J.A. Ewing, H. Gienapp, D.E.
Hasselmann, P. Kruseman, A. Meerburg, P. Mller, D.J.
Olbers, K. Richter, W. Sell, and H. Walden, 1973. Measurements of wind-wave growth and swell decay during
the Joint North Sea Wave Project (JONSWAP)Ergnzungsheft zur Deutschen Hydrographischen Zeitschrift Reihe,
A(8), 12:95
Hasselmann, S., K. Hasselmann, J. H. Allender and T. P.
Barnett, 1985. Computations and parameterizations of
the nonlinear energy transfer in a gravity wave spectrum. Part II: parameterizations of the nonlinear transfer
for application in wave models, J. Phys. Oceanogr.,
15(11):13781391
604
Holthuijsen L. H., 2007. Waves in Ocean and Coastal

Waters, Delft University of Technology and UNESCOIHE, Cambridge University Press.
Hughes, S.A., 1984. The TMA shallow-water spectrum
description and applications. U.S. Army Corps of Engrs.
Coastal Eng. Res. Center, Waterways Expt. Station,
Vicksburg, Miss., Tech. Rep. CERC-84-7.
IEA | OCEAN ENERGY SYSTEMS. ANNEX II REPORT
2003, Implementing Agreement on Ocean Energy
Systems Engineering 34 (2007) 12651274.
Kelvin, Lord, 1887. On the Waves Produced by a Single
Impulse in Water of any Depth, nr in a Dispersive Medium.
Proc., Roy. Soc., 42, 80.
Kitaigorodskii, S.A., 1962. Application of the theory of similarity to the analysis of wind-generated water waves as
a stochastic process, Bull. Acad. Sci. USSR Geophys. Ser.
no. 1, 73p.
Kitaigorodskii, S. A., Krastiskii, V. P., Zaslavskii, M. M.,
1975. On Phillips theory of equilibrium range in the
spectra of windted gravity waves. J. Phy. Oceanog. 5:
410420.
Komen, G. J., Hasselmann, S., Hasselmann, K. 1984. On the
existence of a fully developed wind-sea spectrum. J. phys.
Oceanogr., 14(8):12711285.
Komen, G. .J: Cavaleri, L.; Donelan. M.; llasselmann, K.:
Hassclmann, S. & Janssen, P. A. E. M. 1994. Dynamics
and modelling of ocean waves. Cambridge. Cambridge
University Press. 532 p.
Lewis, A.W., and R. N. Allos, 1990, JONSWAPs parameters: sorting out the inconsistencies, Ocean Engineering,
17(4):409415
Miles, J. W., 1957, On the generation of surface waves by
shear flows, J. Fluid Mech., 3:185204
Mitsuyasu, H. (1968): On the growth of the spectrum of
wind- generated waves (1). Rep. Res. Inst. Appl. Mech.,
Kyushu Univ., 16: 459482.
Phillips, O. M., 1958. The equilibrium range in the spectrum
of wind-generated waves. J. Fluid Mechanics, 4: 426434.
Pierson, W. J., and L. Moskowitz, 1964. A proposed spectral
form for fully developed wind seas based on the similarity
theory of S. A. Kitaigorodskii, J. Geophys. Res., 69:5181
5190.
Pierson, W. J., G. Neumann and R. W. James, 1955, Practical

Methods for Observing and Forecasting Ocean Waves by
Means of Wave Spectra and Statistics, Pub. by the Hydrographic Office under authority of the Secretary of the
Navy, U.S.
Pontes, M.T., R. Aguiar, H. Oliveira Pires, 2005. A Nearshore
Wave Energy Atlas for Portugal, Journal of Offshore
Mechanics and Arctic Engineering, 127:249255.
Prevosto, M., Krogstad, H. E., Barstow, S., and Guedes
Soares, C. 1996. Observations of the High Frequency
Range of the Wave Spectrum. Journal of Offshore
Mechanics and Arctic Engineering. 118:8995.
Resio, D. T., and Perrie, W. 1989. Implications d and f
Equilibrium Range for Wind-Generated Waves, J. Phys.
Oceanogr., 19:193204.
Rodriguez, G. R., Guedes Soares, C., and Ocampos, F. J.
Experimental Evidence of the Transition Between Power
Law Models in the High Frequency Range of the Gravity
Wave Spectrum. Coastal Engineering. 1999; 38:249259.
Rodriguez, G. R. and Guedes Soares, C. Uncertainty in the
Estimation of the Slope of the High Frequency Tail of
Wave Spectra.Applied Ocean Research. 1999; 21(4): 207
213.
Rusu, E. and Guedes Soares, C. Numerical Modelling to
Estimate The Spatial Distribution Of The Wave Energy
in The Portuguese Nearshore. Renewable Energy. 2009;
34(6):15011516.
Rusu, L., Pilar, P., and Guedes Soares, C. Hindcast of the
Wave Conditions along the West Iberian Coast. Coastal
Engineering. 2008; 55(11):906919.
Silvester, R., Hsu, J.R.C., 1997. Coastal Stabilization, World
Scientific. 578 pp.
Toba, Y.,1973, Local balance in the airsea boundary processes, III. On the spectrum of wind waves, J. Oceanogr.
Soc. Japan, 29(2):209220
Tucker, M. J., 1991. Waves in Ocean Engineering:
Measurement, Analysis, Interpretation., Ellis Horwood,
Chichester, U. K, 431 pp.
605

Conceptual design of an ocean current turbine for deep waters

L.R. Nuez Rivas, A. Lopez Pieiro, J.A. Somolinos Sanchez & F. Robledo de Miguel
E.T.S. Ingenieros Navales, Technical University of Madrid (UPM), Madrid, Spain
ABSTRACT: This paper shows the result of a research project that had as main objective the development of
the conceptual design of a new type of submarine electrical generator to get the exploitation of marine currents
that are sited on many places in the world into deeper waters, with good performance. The proposed device can
work into the water and gets the right depth and position by itself only using hydrodynamics forces that allow it
to go the surface or dive without any other help. The generator has not fixed foundations and it is placed on an
intermediate level between the surface and the bottom through a simple anchor system, the entire device even
the anchor system included will be able to be removed without any damage to the marine environment.
INTRODUCTION
In a world scene that it shows a large growing of the

demand of energy and also the decrease of the CO2
emissions to the atmosphere, its absolutely necessary
to look at the renewable energetic sources and into
them, the marines.
Between the different marine renewable energies
is the kinetic energy of marine currents flow that
have origin in seas and oceans by several superposing
effects that are , the tides, different density between
waters, the wind and the Coriolis forces.
Mainly, the marine currents that have economical
interest are originated by tides and only those where
the mean spring peak velocity is high it is possible to
get enough kinetic energy for harnessing.
Nowadays the Tidal Stream Technological Developments (King & Tryfonas 2009) are yet in the beginning
of their life and there are only devices used mainly
as a test to improve the technology and not to generate or sell electricity. At present time there is over
seventy devices into this technology and only one, the
Marine Current Turbines Sea Gen with 1,2 MW generation power, is connected to the electrical network. The
others technological concepts are in different stages
of development but not in industrial exploitation and
any of them, Sea Gen included are capable to exploit
currents that are than more of 40 meters depth sites
(Fraenkel 2010).
But most of the more interesting places for harnessing of this energetic source are with depths over
40 meters and will be necessary a new generation of
capable converters to operate in these deeper waters
and this is the main objective of the new device that
its conceptual design this paper shows.
It begins with a brief explanation about the characteristics of the source and also with the sites where
the velocity of the currents allows the installation of
an exploitation farm. After it an analysis of the state
of the art that this technology has at present time it

is done.
Afterwards it is explained a large description of the
shape and the main elements of the generator and is
finished with the description of the design of the mooring system, the connections to transport the energy to
the shore, the installation, anchoring and removing,
of the device and also the maintenance procedure and
the way to transport it from the shore to marine site for
operation.
2
2.1
CHARACTERISTICS OF THE SOURCE AND

PLACES OF INTEREST FOR EXPLOITATION
Characteristics of the source
The marine currents, ones from inertial causes and

others from tides, are renewable energetic sources with
a very high power. The quantification of the energy
that will be extracted from them is over 800 Twh/year
(about 4% of world electrical consume) but at present
time it is not possible to exploit a significant part of
this immense energetic potential.
This source has a very reliable long term prediction
of power because the marine currents have their flows
well known along the year and maintain a large energetic capability so that the density of marine waters
is over 1015 kg/m2 and a simple comparison exhibits
that one current with mean velocity of 2 m/s has the
same energetic flow by m2 of transversal area than one
wind current with mean velocity of 18 m/s.
Mainly marine currents that are situated into
European sole rights are originated by tides and on
those of them that the mean velocity of the stream
would be high is possible to get enough kinetic energy
and they are an important renewable sources of energy.
The tides are originated by the gravitational attraction forces over the marine waters by the moon and the
607
sun, if the moon and the sun are aligned the attraction
is larger (spring tides) and is short when the directions
take the form of 90 (nape tide).
The main characteristics of the tidal energy
source are:
Reliable long term prediction of power and speed
Very short environmental impact
High reliability
2.2
Places of interest for exploitation
Nowadays, all of the devices that are developed or

in development need a mean spring peak velocity
over 2.0 m/s to be effective and there are into the
European sole rights exploitation many sites where the
main velocity of the stream is high enough and have
a large energetic potential associated. It happens on
the Scottish coast line and also on others places of the
United Kingdoms marine coast.
There are others good places into the Bretagne in
France and in others places of the European Atlantic
marine coast.
In Italy, the Messina Strait is a very good place and
at the south of Spain there is the Gibraltars Strait
(Nuez-Rivas L.R. & Herreros-Sierra M.A. 2006)
that is the natural way to join Mediterranean Sea
and Atlantic Ocean, its limited by a section between
Punta Europe and Punta Cires (Africa) at east. It has
a variable width, with one maximum of 44 Km at
the west entrance and one minimum of 14 Km at the
Strait Narrows between Tarifa and Punta Cires.
The Straits axis goes in east-west direction along
60 km over 36 N parallel and the bathymetric profile
is irregular with average depth of 550 m in the main
channel and zones where 90 m are only and another
ones with maximum depth of 960 m.
The energetic resource that the Strait performs
is making up by a double current that it is originated by several superposing effects that are, different
density between both masses of water, the different
level between the ocean and the sea, the tides that are
generating in the Atlantic side of the Strait and go into
the Mediterranean Sea and the currents originated by
local winds.
The principal part of the energy associated to this
tides (Novoa-Rojas & Nuez-Rivas 2010) is on the
upper Atlantic waters current and it is going along an
hypothetic channel surrounding by the sea surface and
one level that is situated over 100 meters depth.
Another big part of the energy is associated to the
low waters current that is going along the channel
worked in the sea bed by the waters.
On the Strait there are several places with a maximum currents speed up 2 m/s but normally they are in
deeper waters, more of 40 meters depth, usually over
80 or 100 meters.
Figure 1. Bathymetric profile of the Strait.
there are only developments used mainly as a test to

learn and improve the knowledgement than to generate
electricity to an industrial harnessing.
Nowadays there are over seventy devices into the
Tidal Stream Technology and the majority of concepts have the utilization of a horizontal axis rotational
method of generation and there are not signs that the
industry is going to converge towards a single configuration. There is only one device connected to the
electrical network, the Marine Current Turbines Sea
Gen with 1.2 MW generation power and it is located
into the sea at the Strangford Narrows. The others
devices and technological concepts, are in different
stages of development but not on industrial exploitation and all of them are not capable to exploit submarine currents at more of 40 meters depth (Nuez-Rivas,
Lopez-Pieiro, Leo-Mena, Moran 2010).
The classification of the existing devices, that is
possible to call Tidal Energy Converters (TEC) of
first generation and normally fixed to the bottom of
the sea, is presented below:
Open horizontal axis rotor devices, as the same
shape than the machines used in wind power but
with two or three blades, rotors, the axis is parallel
to the current lines of the flow.
Horizontal axis rotor into closed conduct, devices
and with this shaping is gets an acceleration of the
stream and the protection of the rotor.
Vertical axis rotor devices, here the axis of the rotor
is perpendicular to the flow lines and the generator
is over on a floating platform.
Devices with blades of oscillatory or linear, movements. With this configuration the oscillation of
the blades by the stream is the pump that moves an
electro hydraulic generator.
4 THE CONCEPTUAL DESIGN
STATE OF THE ART OF THE TECHNOLOGY
At present time as it is known, tidal stream exploitation technology is yet in the beginning of its life and
It is known that the gross power than can be obtained

from one marine current source with a device that has
horizontal axis rotor is related to the velocity of water
608
and to the diameters of the rotor. The power and the

energy generated during a given period of time can be
obtained through the following equations (1) and (2).
where A stands for the cross sectional area of the rotor

and the maximum power to be obtained in this case
is given by equation (3) being Cp a coefficient power
whose valor can reach 0.44 and as can be seen, the
maximum power is proportional to the area and to the
cubic value of the velocity
Because many of the better places to get energy

from a marine current source are on depth waters over
40 m as it happens on Gibraltars Strait, was the challenge to exploit the energy of these current the task that
this new type of device (First of a second generation
of TECs) would do it with a good performance.
And was the conceptual design of this type of device
the project main objective developing it with a low
life cycle cost device and designed for Gibraltars
Strait and other world sites with waters depth over 40 m
and with the following goals:
Figure 2. General View of GESMEY U1M.
Simplified deployment
Minimum environmental impact
Without surface elements on operation
Robust and simple construction
Easily scalable (depth, stream, speed, nominal
power)
Integration of mature technologies
With these requests a new conceptual design,
the Submarine Electrical Generator with Y Shape
Framework (GESMEY), was done.
The Project was germinated from an initial idea of
Professor Lopez Pieiro, patented by the Technical
University of Madrid (UPM) (Lopez-Pieiro 2007)
and it has been developing in collaboration between
research teams, one from the E.T.S. Ingenieros Navales
of the UPM and other one from the Foundation Technological Centre SOERMAR. The Project has been
supported by funds from The National Research
Program 2008/2011 of the Spanish Administration.
This new concept of generator will have one rotor
with three blades that it propels one electrical generator, placed into a pod and joined it there are three
columns, situated on a perpendicular plane with the
axis of the rotor and at the end of each column there
is one float with torpedo shape than has its axis in a
parallel way of the rotors one.
There were developed two options; one is a generator GESMEY A6.7 with 600 kW of power that its
optimum for places with bidirectional currents flux
and the other one GESMEY U1M with 1 MW of power
for places where the one way flux of the stream is
highest. Both of them are shown on the Figures 23.
Figure 3. General View of GEMEY A6.7.
Both concepts have a rotor with fixed pitch blades

because they has better reliability and with very
improve efficiency it the flux is unidirectional as
happens in the case of U1M generator, on the other
hand the generator A6.7 is optimized by one of the
two directions of the bidirectional flow and for the
other has an acceptable performance. This last case is
optimum for the currents on Gibraltars Strait (NuezRivas, Lopez-Pieiro, Novoa-Rojas, Carneros-Lozano
2010).
The other main parts of two concepts of GESMEY
are, one central pod with the power take off components and the auxiliary systems, the columns are main
structural parts and also are used as auxiliary ballast tanks and the end torpedoes are the main ballast
tanks than lets the stability of the device on operation
with asymmetric loads and its stability on flotation
like a semisumergibles platform.
609
Table 1. General Specifications of the concepts of

GESMEY.
Description
Power (kW)
Stream velocity (m/s)
Sea bed depth (m)
Rotor diameter (m)
Minimum end of the
Blade depth (m)
Rotor (RPM)/Gear Output
(RPM)
Structure steel steel Number of
devices in a generations park
GESMEY
A6.7
GESMEY
U1
600
2.0
60/100
20
15
1000
1.8
80
32
34
12/1500
12/750
20/50
20/50
Figure 4. Anchoring of one generator with the flow by bow.
5 THE INSTALLATION AND ELECTRICAL

CONNECTION
5.1
Figure 5. Transport on float of the device.
Installation
It would be necessary, as first step of the technical process, the definition of a specific mooring system and
its installation in the sea (Lopez-Pieiro, SomolinosSanchez, Nuez-Rivas, Santamaria-Romojaro 2010).
This system will fix the generator on the right position against the flow of the current during electrical
generation and it is formed by the following elements:
Submerged buoys with cylindrical shape, three by
generator
Three mooring lines that fixed the buoys to the
generator
One mooring line that anchor the buoy to sea bed
The anchorage into the sea bed
These cylindrical buoys are conformed from steel
plates on a very simple way of construction, the
anchorage is formed by a concrete monopile and the
mooring lines are steel or synthetic fiber, wires.
With this shape, all the mooring system is sufficiently simple, easy of install and with a very less
impact over the marine environment and gets the
following objectives:
Maximum independence between the mooring and
the generator
Easy access to the hook point
High reliability and safety of the device because
the system support one line failed
The joint of the mooring system to the generator doesnt produce large mechanical forces that
would can do the lack of balance of the generator
Modular conform that allows by a simple concentration the anchoring of all the generator of a park
or only one of them
With this system there are three submerged buoys
by generator, one by the bow part and two at the rear,
the bow wire of mooring is 94 m long and the two
rear wires are 101 m each one and with this configuration into a park the generators will be separated
over 70 m on the transversal way of the flow and

about 200 m along the same direction of the current
has. The electrical connection can be easily seen from
Figure 4.
The generator is completely formed in a yard or
on a shore platform, near the point where it will be
installed on the sea and is transported on float by a
tug vessel and when it reaches the location over the
vertical of the place of operation, it is engaged to the
mooring system and connected to the electrical line to
transmit the energy ashore, using one auxiliary vessel.
Then the immersion procedure begins and the device
is placed on the right position for operation, only with
hydrodynamics forces and by controlling the ballast
level, at last it is fixed on the mooring buoy using a
ROV or by a diver that makes the connection.
5.2 Electrical connection
The transport of the electrical energy from the generator to the shore is possible to do it on different
configuration options.
The first one that will be used during the experimental trials of the prototype, is a simple system consisting
on a cable for transfer electrical AC current from
the generator to the shore, the cable have three wires
for three phase current and it is connected at the rear
of the pod and is supported by the mooring line to
the buoy and then lay along the anchoring line to the
bottom, at side of the anchor pile there is the connector with the power cable that on the sea bed transfer
the energy to step up transformer and there reach the
right high voltage than is necessary to connect into the
electrical network. This arrangement is also suitable in
the case of a low power generation park sited near off
shore.
In the second option, it will be used when a
generation farm would be constructed, each one of
the generators is connected with a dedicated power
610
Figure 6. Work position of the electrical current transfer

cable.
Figure 7. Immersion procedure.
Figure 9. Pre-emersion position.
converter. This configuration would provide a high

level of flexibility of the system, that allows generators rotating at different speeds and also that several
of the generator be shut.
The power cable of each and by a submerged connector is connected to an only power line that lays
along the sea bed to the shore and at its end there is
the connection with one step out transformer and then
with the right voltage is transferred to the electrical
network.
Other alternative option is the use of a HVDC power
line, in this case there is a power converter AC/DC
at the take off from each generator and all of them
are connected in a parallel way to a big section power
cable that transport the DC to the shore and there is
a DC/AC converter and behind is the step up transformer to reach the right voltage and the shore grid
connection.
6
6.1
Figure 8. Anchoring of the generator in operation.
OPERATION, MAINTENANCE, REMOVAL

Operation
The generator is positioned on the right depth by

itself, connected to the mooring lines and is operating
standing its position with a very less deviation.
As one example choosing the A6.7 device to build
one exploitation farm in a place of Gibraltars Strait
on 80 m depth with 2.0 m/s mean velocity and thirty
machines, each one with rotors diameter of 20 m, it

will be possible have the power of 16.8 MW.
The array of this farm with the thirty generators
needs a distance between devices over 140 m in the
transversal direction of the current and about 840
meters along the flow way and all the surface occupied
on the sea bed is about 12 hectares.
6.2 Maintenance and removal
The usual maintenance is doing on float and when it
is necessary do it, the emersion procedure begins and
put the generator on flotation.
The rope that hooks the device to the bow buoy
is detached and the ballast water is expelled out by
compressed air changing the level of the water into
the ballast tanks and the device goes up to the surface.
The generator emerges to the surface with vertical
position and the emersion time is over ten minutes
as has been probed with a simulation procedure and
arrive to the surface floating in vertical way.
At the surface it is necessary to change the main
axis position from horizontal to vertical direction by
expelling more ballast water it gets this new position
and the device lays floating over the waters with the
rotor out over four meters above the surface of the
sea and on this way is possible to do the necessary
operation of usual maintenance task with the help of an
auxiliary vessel (Somolinos-Sanchez, Lopez-Pieiro,
Nuez-Rivas, Novoa-Rojas 2009).
611
It has a low cycle cost because the installation,

maintenance operations and the final remove will
be cheaper than other devices that at present
time are.
It has a simple system to emerge or submerge into
the sea only with hydrodynamics forces.
It has less weight and volume than other devices
with similar power generation.
Figure 10. Position in the arrive to the surface.
REFERENCES
Figure 11. Final emersion position.
With this new position the device can be transported

to one port or one yard if it is necessary for make
especial maintenance duties.
And when the device finishes its life cycle over
25 years, the removal of the machine is towing it on
float by a tug vessel, and also with the help of one Rov
or one diver is easy to take down the buoys and the
anchoring and mooring lines and put them away in a
vessel, laying only the three concrete piles into the sea
bed, without any environmental damage.
7
CONCLUSIONS
This electrical generator is a new concept that will

be able to exploit marine currents energy on sites
with rotor depth over 40 meters.
It has a robust and simple construction.
It has the possibility to exploit currents with stream
speed below 2 m/s.
It not need especial and expensive vessel or offshore artefacts to transport it or positioned in the
operation sites and also to emerge and removal.
It needs a simply mooring and anchoring system
that has minimum environment impact.
It has a minimum environment impact because it
works submerged but not lies over the sea bed.
Fraenkel P.L. 2010. Developments and Testing of marine

current turbines Sea Gen 1,2 MW tidal stream turbine.
Proceedings of the International Conference on Ocean
Energy ICOE 2010, October, Bilbao, Spain.
King J. & Tryfonas T. 2009. Tidal Stream Power TechnologyState of Art. OCEANS 2009 Europe conference, May,
Bremen, Germany.
Lopez-Pieiro A. 2007. Sistema Sumergible para el
aprovechamiento energtico de la corriente marina.
Patente P200700987 UPM, BOPI 16707/08, Spain.
Lopez-Pieiro A., Somolinos-Sanchez J. A., NuezRivas L.R., Santamaria-Romojaro M. 2010. Modelado y Simulacin de Dispositivos Fondeados para
el Aprovechamiento de la Energa de las Corrientes
Marinas. IV Jornadas de Automtica Marina AUTOMAR
2010, 2122 October, Murcia, Spain.
Lopez-Pieiro A. et al. 2010. Generador elctrico submarino
para el aprovechamiento de las corrientes de flujo bidireccional. Spanish patent ES 234131311B2 Nov. 2010.
Spain.
Nuez-Rivas L.R. & Novoa-Rojas E.2010. The exploitation of marine currents energy on the Strait of Gibraltar.
Journal of Marine Technology and Environment, Vol. II
pages 125130 November, ISSN 1844-6116, Constanta,
Romania.
Nuez-Rivas L.R. & Herreros-Sierra M.A. 2006. Gibraltar
Strait a marine renewable energy source. Proceedings
World Maritime Technology Conference WMTC 2006,
May, IMAREST, London, U.K.
Nuez-Rivas L.R., Lopez-Pieiro A., Novoa-Rojas E.,
Carneros-Lozano A. 2010. The GESMEY Ocean Current
Turbine. A proposal for marine current energy extraction on deeper waters. Proceedings of the International
Conference on Ocean Energy ICOE 2010, October,
ISBN 978-84-693-5467-4, Bilbao, Spain.
Nuez-Rivas L.R., Lopez-Pieiro A., Leo-Mena T., MoranGonzlez J.L. 2010. Journal of marine Technology
and Environment, Vol. II pages 131/138 November,
ISSN 1844-6116, Constanta, Romania.
Somolinos-Sanchez J.A., Lopez-Pieiro A., NuezRivas L.R., Novoa-Rojas E. 2009. Simulation of the
emersion procedure for a new underwater electrical
generator. Proceedings 21th European Modeling &
Simulation Symposium MAS 2009, ISBN 978-84-6925417-2, October, Puerto de La Cruz, Tenerife, Spain.
612

Assessment of the wave energy in two enclosed seas

Eugen Rusu & C. Guedes Soares
ABSTRACT: The objective of the present work is to evaluate the wave energy resources in two enclosed seas,
the Black Sea and the Caspian Sea. A first perspective of the wave and wind climate in these areas is given
by a medium term analysis of some recent satellite data. This allowed the identification of the most energetic
sectors. As a further step, wave prediction systems based on the SWAN spectral phase averaged model were
implemented and validated in the two seas. Comparisons carried out against remotely sensed data show that these
wave prediction systems generally provide accurate results in terms of significant wave heights. Finally, using
the numerical wave models, the spatial distribution of the wave energy was analyzed for some case studies that
correspond to the most relevant wave patterns. The present work gives some insight about the wave conditions,
and the wave energy potential, in the basins of the Black and the Caspian Seas. Moreover, the perspective of the
development of the wave energy devices for small amplitude waves is expected to be very dynamic in the near
future and to bring technological breakthroughs. Thus following such tendencies, the problem of extracting this
type of renewable energy in the Black and the Caspian Seas, most probable coupled in hybrid farms wind-waves,
might become of actuality.
INTRODUCTION
Extraction of wave energy became in the last decade

one of the most challenging engineering problems.
Wave energy is abundant and using numerical models
it can be predicted with a good accuracy in a time
window of a few days. On the other hand, wave energy
is not only more predictable than wind or solar energy
but it has also a higher energetic density allowing
extraction of more energy in smaller areas.
Various devices to extract this energy have been
designed, but at this moment there is yet no technology that can be considered as being the most effective.
A very important inconvenient that is usually encountered in the areas traditionally considered as having
high potential in wave energy is that very often these
areas are subjected to strong wave conditions that may
destroy the devices operating for extracting the wave
energy.
From this reason an alternative solution would be
to explore the possibility of implementing the energy
farms, eventually together with wind farms, in areas
with smaller wave amplitudes, but where the extreme
wave conditions are usually not so severe, allowing
the functionality of the wave extraction systems for
considerably longer time intervals.
In this respect the objective of the present work
is to evaluate in parallel the wave conditions and the
wave energy resources in two enclosed seas using both
satellite data and results coming from a wave prediction system based on numerical models. These are the
Black Sea and the Caspian Sea.
The Black Sea is an enclosed sea located deeply

inside the continent that represents the most isolated
part of the World Ocean while the Caspian Sea can
be considered the largest inland body of water in the
world and accounts for 40% to 44% of the total lake
waters of the world.
The maximal length of the Black Sea (along the
latitude 42 29 N) is 1148 km, while its minimal width
along the meridian from Crimea to the coast of
Turkey is only 258 km. The principal characteristics
of the Black Sea are: 423000 km2 for the sea area,
555000 km3 for its volume, and 1315 and 2258 meters
for the mean and maximal depths, respectively. Three
principal structures: the shelf, the continental slope,
and the deep-water basin, can be clearly distinguished
in the bottom topography of the sea. The shelf occupies up to 25% of the total area of the seafloor and,
on average, is restricted to sea depths of 100200 m. It
reaches its greatest width (more than 200 km), in the
northwestern part of the sea, which is entirely located
within the shelf zone. Almost over the entire extension of the eastern and southern coasts of the sea, the
shelf is very narrow (only a few kilometers wide); in
the western part of the sea, it is wider (a few tens of
kilometers).
The Caspian Sea extends approximately 1200 km
from north to south, with an average width of
325 km east to west, covering a total area of about
400000 km2 . Most of the northern Caspian is shallow, with water depths averaging only 4 m. The central
Caspian approaches depths of 800 m, while the southern Caspian has a maximum depth of slightly over
613
1000 meters. The bathymetric features together also

with some geographical characteristics of the surrounding relief are very different from north to south
and as a consequence the climate of the Caspian region
varies also substantially.
The present work points to find out some preliminary answers to the question of the availability of the
wave energy resources in these two seas.
2 ANALYSIS OF REMOTELY SENSED DATA
In the last years more accurate satellite data became
available on various internet sites (as for example http://las.aviso.oceanobs.com that was the main
source considered in the present work). An altimeter
node gives for each day at zero hours near real time
multi-mission merged non interpolated values of the
significant wave height (Hs) and wind speed (Vw).
These are time (for the last 48 hours) and space (for
1 squares centered in the node) averaged data sets.
As regards measuring the sea waves, the estimates
are obtained using empirical models derived from
analyses of the altimeter data. The algorithm used
to deduce the significant wave height is based on
the initial results of Gaspar et al. (1998) with the
new parametric fits by Labroue et al. (2002). For
a typical significant wave height of 2 meters, the
error in the sea state bias correction is approximately
12 cm, i.e., 0.5% to 1.0% of the effect. For measuring the wind speed, a mathematical relationship
that considers the Ku-band backscatter coefficient
together with the Vandemark and Chapron algorithm
is used (Vandemark et al., 2000). The wind speed
model function is evaluated for 10 meters above the
sea surface.
In order to provide a more recent picture of the
characteristics and dynamics of the most relevant
parameters (Hs and Vw ) concerning the wave and wind
conditions in the two seas, some synthetic results
coming from the analysis of the remotely sensed
data, corresponding to the time interval December
2005June 2010 will be presented bellow.
Some definition relations for the main statistical
parameters evaluated are given bellow. In probability theory and statistics, kurtosis is a measure of the
peakedness of the distribution of a real-valued random variable. This is equal to the fourth moment
around the mean divided by the square of the variance (or the fourth power of the standard deviation) of
the distribution minus 3.
4 / 4 is also called the fourth standardized moment.

The minus 3 at the end of this formula is often
explained as a correction to make the kurtosis of the
normal distribution equal to zero. The kth moment
about the mean is defined as:
Table 1. The Black Sea; overall statistics for the Hs data in

the eight reference points for the period 2005 to June 2010.
Pt
Poz
Mean
(m)
Max
(m)
Std
Kurt
Skew
A1
A2
A3
A4
A5
A6
A7
A8
40 E, 42 N
40 E, 43 N
32 E, 42 N
36 E, 42 N
34 E, 43 N
38 E, 43 N
36 E, 44 N
38 E, 44 N
0.85
0.85
0.97
0.91
0.94
0.87
0.89
0.86
5.4
5.2
5.1
5.1
5.1
5.1
5.1
5.1
0.54
0.53
0.55
0.50
0.50
0.50
0.48
0.48
11.75
10.41
5.58
7.25
6.12
8.17
6.90
7.82
2.47
2.33
1.84
1.92
1.79
2.02
1.85
1.95
where E is the expectation operator. If X is a discrete

random variable with probability mass function
p(x),

then the expected value becomes E(X ) = xi p(xi ).
The standard deviation () of X is the quantity:
with = E[X ] the mean value.

Skewness is a measure of the asymmetry of the
distribution of a real-valued random variable. The
skewness value can be negative or positive. Qualitatively, a negative skew indicates that the tail on the
left side of probability density function is longer than
the right side and the bulk of the values (including the
median) lie to the right of the mean. A positive skew
indicates a reversed tendency. The skewness of a random variable X is defined as the third standardized
moment:
Considering the statistical parameters above defined,

Table 1 presents the characteristics of the Hs data in
eight selected points from the Black Sea where this
parameter has greater values. These reference points
were denoted with A1, A2,, A8. The monthly average Hs values are illustrated in Figure 1. The second
parameter analyzed is the wind speed (Vwin ) and its
monthly average values are illustrated in Figure 2 while
Table 2 presents the wind characteristics in statistical
terms.
For the Caspian Sea, 14 reference points covering
the entire basin of the sea were defined. They were
denoted as P1 to P14 and they are counted from south
to north. Since the objective of the present work is to
estimate the wave energy, more points were defined
in the southern and central parts of the sea where
due to the deeper water depth larger waves are usually expected. The reason for selecting in the Caspian
Sea a greater coverage (more points covering the most
relevant parts of the basin) is that while for the Black
Sea previous analyses were performed by the authors
(see for example Rusu, 2009), in the Caspian sea
results of such previous analyses were not available.
614
Table 3. The Caspian Sea; overall statistics for the Hs data

in fourteen reference points for the period 2005 to June 2010.
Figure 1. The Black Sea; monthly averaged values of Hs in

eight reference points for the period 2005 to June 2010.
Pt
Poz
Mean
(m)
Max
(m)
Std
Kurt
Skew
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
P11
P12
P13
P14
50 E, 38 N
51 E, 38 N
52 E, 38 N
51 E, 39 N
51 E, 40 N
51 E, 41 N
50 E, 41 N
51 E, 42 N
50 E, 42 N
49 E, 42 N
51 E, 43 N
50 E, 46 N
51 E, 46 N
52 E, 46 N
0.79
0.79
0.78
0.86
0.91
0.95
0.95
0.95
0.95
0.95
0.88
0.60
0.59
0.58
3.85
3.81
3.76
4.09
4.24
4.43
4.35
4.18
4.01
3.90
4.10
3.33
3.27
3.25
0.51
0.51
0.51
0.58
0.61
0.64
0.64
0.65
0.64
0.65
0.62
0.41
0.41
0.41
3.29
3.21
3.09
4.01
3.63
3.10
3.02
2.75
2.65
2.56
3.23
9.57
9.63
9.59
1.56
1.55
1.54
1.68
1.62
1.54
1.53
1.51
1.50
1.49
1.62
2.50
2.52
2.52
Table 4. The Caspian Sea; overall statistics for the Vw data

in fourteen reference points for the period 2005 to June 2010.
Figure 2. The Black Sea; monthly averaged values of Vwin

(in m/s) in eight reference points for the period 2005 to
June 2010.
Table 2. The Black Sea; overall statistics for the Vw data in
the eight reference points for the period 2005 to June 2010.
Pt
Poz
Mean
(m)
Max
(m)
Std
Kurt
Skew
A1
A2
A3
A4
A5
A6
A7
A8
40 E, 42 N
40 E, 43 N
32 E, 42 N
36 E, 42 N
34 E, 43 N
38 E, 43 N
36 E, 44 N
38 E, 44 N
3.86
3.96
3.97
4.76
4.18
4.07
4.28
4.42
16.4
16.4
16.4
16.3
16.3
16.3
16.2
16.2
2.27
2.22
2.25
2.37
2.12
2.20
2.09
2.11
2.55
2.20
2.10
1.22
1.50
1.66
1.42
1.28
1.27
1.20
1.18
1.03
1.05
1.08
1.03
1.00
Pt
Poz
Mean
(m)
Max
(m)
Std
Kurt
Skew
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
P11
P12
P13
P14
50 E, 38 N
51 E, 38 N
52 E, 38 N
51 E, 39 N
51 E, 40 N
51 E, 41 N
50 E, 41 N
51 E, 42 N
50 E, 42 N
49 E, 42 N
51 E, 43 N
50 E, 46 N
51 E, 46 N
52 E, 46 N
4.21
4.21
4.20
4.58
4.86
5.14
5.15
5.32
5.31
5.33
5.29
4.72
4.70
4.68
14.00
13.93
13.78
15.96
17.1
18.27
18.15
17.77
17.43
17.19
17.57
16.30
16.27
16.23
2.35
2.34
2.34
2.52
2.69
2.84
2.84
2.92
2.92
2.93
2.85
2.56
2.56
2.57
0.74
0.70
0.68
0.70
0.60
0.38
0.40
0.24
0.23
0.20
0.49
1.11
1.13
1.14
0.91
0.90
0.90
0.88
0.83
0.77
0.77
0.73
0.73
0.72
0.81
0.97
0.98
0.98
points P8, P9 and P10. The histograms for these two

parameters (Hs and Vw ) related to the same data set
(related to the time interval December 2005June
2010) are presented in Figure 3 and 4, respectively.
The data from the histograms was structured in total
and winter time respectively, where winter time is considered here the interval between the beginning of
October until the end of March.
3
Tables 3 and 4 present the overall Hs and Vw
statistics for the remotely sensed data registered in
the Caspian Sea in the fourteen reference points
considered.
The results presented in Tables 3 and 4 show that
the central part of the Caspian Sea is more energetic,
especially the locations corresponding to the reference
IMPLEMENTATION OF THE WAVE MODEL
A wave prediction system based on the SWAN model

was implemented and evaluated separately in each
sea. As regards the Black Sea basin, validation tests
were previously performed against buoy data as presented by Guedes Soares & Rusu (2005) and Rusu
(2009). Further validations of the above modeling system were performed also by Rusu and Ivan (2010) and
615
Table 5. The Caspian Sea; wave and wind statistics,

remotely sensed data against SWAN and wind model outputs.
Results in five reference points (P4, P8, P9, P10, P12); time
period considered 2009/01/012009/12/31. a) Hs statistics;
b) Vw statistics.
Pt
Xm
a) Hs (m)
P4
0.91
P8
1.03
P9
1.04
P10
1.01
P12
0.60
b) Vw (m/s)
P4
4.49
P8
5.33
P9
5.29
P10
5.20
P12
5.08
Figure 3. Satellite data, Hs histograms for the reference

points P8, P9 and P10. Daily records for the time interval
December 2005June 2010; a), c) and e) total time, b), d)
and f) winter time.
Figure 4. Satellite data, Vw histograms for the reference

points P8, P9 and P10. Daily records for the time interval
December 2005June 2010; a), c) and e) total time, b), d)
and f) winter time.
the above system was used to provide the support in the

case of the environmental alerts (Rusu, 2010a) and to
assess the wave-current interactions at the mouths of
the Danube (Rusu, 2010b).
Ym
bias
rmse
si
1.04
1.18
1.16
1.02
0.44
0.13
0.14
0.12
0.01
0.16
0.37
0.39
0.38
0.34
0.33
0.41
0.37
0.37
0.34
0.55
0.79
0.83
0.81
0.82
0.77
4.37
5.60
5.26
4.83
5.44
0.12
0.27
0.03
0.37
0.36
1.26
1.40
1.28
1.29
1.28
0.28
0.26
0.24
0.25
0.25
0.83
0.86
0.88
0.89
0.90
Some results concerning the implementation in the

Caspian Sea of a wave modeling system SWAN based
are presented bellow. The system origin corresponds
to the lower left corner point and has the coordinates (46.7 E, 36.2 N), whereas the lengths are 8 in
x-direction (longitude) and 11.2 in y-direction (latitude). In the geographical space the computational
grid was chosen identically with the bathymetric grid
and has 100 points in x direction and 140 points
in y direction, the points are equally spaced with
x = y = 0.08 . In the spectral space, 36 directions and 30 frequencies were assumed. The frequency
range considered is between 0.06 Hz and 1.2 Hz. The
computations were performed in the non stationary
mode with a 5 minutes time step. Wave model simulations were carried out for the entire year 2009
and in statistical terms the Hs comparisons against
remotely sensed data are presented in Table 5 for
the five reference points considered (P4, P8, P9, P10
and P12).
For the three points that were found as the most energetic in the Caspian Sea (P8, P9 and P10) Figure 5
illustrates direct comparisons for Hs , SWAN results
against remotely sensed data, while Figure 6 (a, c
and e) presents the Hs scatter plots for the same
three locations. Since the accuracy of the input wind
field represent a fundamental issue in obtaining better
results in wave modeling, Table 5 presents also the statistical analysis for Vw (ECWMF model wind against
the corresponding remotely sensed data) at the same
five locations (for the reference points P4, P8, P9, P10
and P12) and Figure 6 (b, d and f) presents the Vw
scatter plots for the points P8, P9 and P10.
EVALUATION OF TWO CASE STUDIES
Using the wave modeling systems implemented in the

two seas, two case studies will be considered for each
616
length of wave front), are computed with the relationships:
where: x, y are the problem coordinate system (for

the spherical coordinates x axis corresponds to longitude and y axis to latitude), E(, ) the wave energy
spectrum, the relative wave frequency, the wave
direction and cx , cy are the propagation velocities of
the wave energy in the geographical space defined as:
Figure 5. The Caspian Sea; Hs direct comparisons, SWAN

results against satellite data for the entire year 2009. a) Reference point 8; b) Reference point 9; c) Reference point 10.
Hence the absolute value of the energy transport

(denoted also as wave power) will be:
The non dimensional normalized wave power is

expressed as:
In the present work ETR max was defined separately for

each individual case study. This is a round off value
approximated the maximum value corresponding to
the computational domain.
4.1 Black Sea, Case study 1 1997/01/12/h12
Figure 6. The Caspian Sea; Scatter plots for the parameters Hs and Vw model (wave and wind) against satellite data
for the entire year 2009 a) Hs for reference point 8; b) Vw
for reference point 8; c) Hs for reference point 9; d) Vw for
reference point 9; e) Hs for reference point 10; f) Vw for
reference point 10.
sea to evaluate and analyze the most relevant patterns concerning the spatial distribution of the wave
energy.
In SWAN, the energy transport components
(expressed in W/m, i.e., energy transport per unit
This case study, denoted as BS1, provides an average

energy distribution for the entire Black Sea basin and
is illustrated in Figure 7a. This figure shows the normalized wave power in background and the energy
transport vectors (in kW/m of wave front) in the foreground. The locations, for this computational domain,
of the maximum values of the wave power are marked
with circles. For this case study the value of ETR max was
set at 20 kW/m. The quantities ETR max 1 and ETR max 2
represent the effective values of the most energetic
points and their locations are also illustrated in the figure. Although there is an obvious relationship between
significant wave height and wave power, the energetic
peak in a computational domain is not necessarily
located at the same point as in the present case.
4.2 Black Sea, Case study 2 1997/01/21/h18
This second case study presents some storm conditions
in the Black Sea (BS2) and is illustrated in Figure 7b.
It should be noted that this is not an extreme event, but
a regular storm (a typical storm when the western part
of the Black Sea is more energetic due to the dominant
wind patterns). For this case the value of ETR max was
set at 150 kW/m.
617
Figure 7a. The Black Sea, Case study 1 1997/01/12/h12, average energetic situation, representation for the entire Black
Sea. In background normalized wave power, in foreground energy transport vectors (in kW/m of wave front).
Figure 7b. The Black Sea, Case study 2 1997/01/12/h12, average energetic situation, representation for the entire Black
Sea basin. In background normalized wave power, in foreground energy transport vectors (in kW/m of wave front).
618
Figure 8. The Caspian Sea; a) Case study 1 2009/10/02/h18, average energetic situation, b) Case study 2 2009/11/27/h03,
high energy conditions. In background normalized wave power, in foreground energy transport vectors (in kW/m of wave
front).
4.3
Caspian Sea, Case study 1 2009/10/02/h18
The first case study in the Caspian Sea, denoted as

CS1, reflects wave conditions with average energy
(for the winter time period) in the Caspian basin.
The background of Figure 8a shows the normalized
wave power (ETR /ETR max ) for CS1 and the foreground
shows energy transport vectors (represented with red
arrows in kilowatts per meter of wave front). The
location in the computational domain of the maximum
value for the wave power is marked with a red circle.
For the present case study the value of ETR max was set
to 20 kW/m while the maximum wave energy in the
sea was 22.4 kW/m.
highest energetic conditions that can be expected in

the basin of the Caspian Sea.
Figure 8b shows in background the normalized
wave power (ETR /ETR max ) and in foreground the energy
transport vectors (represented with red arrows in kilowatts per meter of wave front). The location in the
computational domain of the maximum value for the
wave power is marked with a red circle. For the present
situation the value of ETR max was set to 100 kW/m
while the maximum wave energy in the sea was
98.9 kW/m.
5
4.4
Caspian Sea, Case study 2 2009/11/27/h03
The second case study considered in the Caspian sea,

denoted as CS2, reflects one of the highest energetic
conditions that were encountered in the central part of
the Caspian Sea for the entire five-year period analyzed (the time interval December 2005June 2010).
Thus, although it can not be considered as an extreme
event this situation can give a good perspective on the
At this point, a discussion will be employed first in

relationship with the results presented in Tables 14
that reflect the altimeter measurements. Thus in the
Black Sea the kurtosis are higher than in the Caspian
Sea for both parameters analyzed (Hs and Vw). This
indicates the fact the energetic peaks are higher in the
Balck Sea. Only the three points from the north of
the Caspian Sea (P12, P13 and P14) that are located in
619
very shallow water have values more elevated for this

parameter. Another observation is that for the points
where the kurtosis is higher also the skewness has more
elevated values.
Some remarks will be made also in relationship
with the accuracy of the results provided by the
wave prediction system that was implemented herewith. A comparison of these wave predictions with
some results obtained in similar environments (semienclosed or enclosed seas) coming also from SWAN
model simulations will be made first.
Dykes et al. (2009) evaluated a SWAN based wave
forecasting system in the Adriatic Sea forced with
ALADIN wind model (acronym from AROME Limited Area Decentralized International Network). For
Hs nowcast against buoy data S.I . was in general
between 0.220.3 and r about 0.9, the accuracy of
the results decreasing in the case of the forecast products. Nevertheless for the case of the wave hindcast
against altimeter data the same system provided Hs
results with lower accuracy (S.I . about 0.320.34 and
r about 0.7278). The accuracy of the wind velocity
was also evaluated against satellite measurements and
there resulted S.I . about 0.3436 and r about 0.7879.
In the Black Sea using ECMWF wind to force
SWAN, Rusu (2009) obtained in a hindcast study for
Hs against buoy data RMSE in the interval 0.320.36,
S.I . 0.360.42 and r 0.780.88. Against satellite data
the same prediction system provided RMSE 0.370.4,
S.I . 0.310.34 and r 0.770.8 (Rusu, 2010b).
Looking at the results presented in Table 5 and
Figure 6 (a, c and e) it can be noticed that from a statistical point of view the estimations provided by the
wave prediction system implemented in the Caspian
Sea are compatible with those coming from similar
systems in the Adriatic Sea or the Black Sea. Thus
RMSE has values between 0.330.39, S.I . between
0.340.41 and r about 0.790.83.
An exception is related with the reference point
P12 that is located in the north. The lower accuracy
concerning S.I . (0.55) and r (0.77) in the conditions
when the wind accuracy is even better in that region
than in other parts of the Caspian Sea (S.I . = 0.25 and
r = 0.90 for Vw ) are probably related with the fact that
the entire northern region of the Caspian Sea is characterized by very shallow water (about 4 m depth).
Thus some process as triad wave-wave interactions,
bottom friction, diffraction, breaking, etc, that were
not accounted in the global simulations may become
rather relevant in this area. A solution to improve
the model predictions in the north of the Caspian Sea
would be to define a higher resolution computational
domain, nested in the area that covers the entire basin,
which would cover the northern sector of the sea and
where the main shallow water processes available in
SWAN would be also activated (performing eventually
a calibration process).
In relationship with some other factors that may
improve the performances of such a wave prediction system based on spectral phase averaged models,
on the first place should be probably considered the
quality of the wind fields. Many studies have been performed on the effects of the wind fields when modeling
waves in enclosed, semi-enclosed or relatively small
basins (as for example Ardhuin et al., 2007 or BolanosSanchez et al., 2007). The most obvious conclusion
would be that the accuracy of the results provided by
the wave models is highly dependent on the accuracy
of the meteorological models that were used to force
them.
Looking at the wind comparisons presented in
Table 5 and Figure 6 it can be noticed that the
ECMWF (acronym for European Centre for MediumRange Weather Forecasts) wind field with 1.5 degrees
spatial resolution and six-hour temporal resolution
(Berrisford et al., 2009) that was used for forcing
SWAN in the present study is in general reasonably
close with the satellite data, but in the same time it is
far to provide a perfect match. Thus a higher accuracy
and resolution (in both space and time) of the wind
field would definitively lead to an improvement of the
wave predictions.
CONCLUDING REMARKS
As regards the average energetic conditions, the maximum values encountered for the wave power were
between 2050 kW/m in the Black Sea and a little over
20 kW/m for the Caspian Sea, while for example in the
Iberian coast these values are 50100 kW/m. On the
other hand, in relation with the highest energetic conditions the differences are even more significant for
the three environments considered. In the Black Sea,
it was around 300 kW/m, while in the Caspian Sea
less than 100 kW/m, and in the Iberian nearshore
the maximum wave power had sometimes values
exceeding 600 kW/m.
Nevertheless, it should be also highlighted that is
expected in the near future a quite dynamic development of the wave energy devices for small amplitude
waves. In this perspective, the problem of renewable
energy extraction in the Black and the Caspian Seas
might become of actuality at least as regards the hybrid
projects wind-waves energy farms.
REFERENCES
Ardhuin, F., Bertotti, L., Bidlot, J.-R., Cavaleri, L.,
Filipetto, V., Lefevre, J.-M. & Wittmann, P., 2007. Comparison of wind and wave measurements and models in the
Western Mediterranean Sea. Ocean Eng. 34, 526541.
Berrisford, P., D. Dee, K. Fielding, M. Fuentes, P. Kallberg, S.
Kobayashi & S. Uppala, 2009. The ERA-Interim archive,
Publisher Shinfield Park, Reading.
Bolanos-Sanchez, R., Sanchez-Arcilla, A. & Cateura, J.,
2007. Evaluation of two atmospheric models for wind
wave modelling in the NW Mediterranean. J. Mar. Syst.
65, 336353.
Dykes, J. D., Wang, D. W. & Book, J. W., 2009. An evaluation
of a high-resolution operational wave forecasting system
in the Adriatic Sea, J. Mar. Syst. 78, 255271.
620
Gaspar, P. & Florens, J.-P. , 1998. Estimation of the sea state

bias in radar altimeter measurements of sea level: Results
from a new nonparametric method, J. Geophys. Res.,103,
15,80315,814.
Guedes Soares, C. & Rusu, E., 2005. SWAN Hindcast in the
Black Sea, Fifth International Symposium WAVES 2005,
3rd7th July 2005, Madrid, Spain, Paper number 155, CD
edition, 11p.
Labroue, F. Ogor, G. Lafitte, L. Marchal, & Rafanel, M.,
2002. Improving nonparametric estimates of the sea state
bias in radar altimetry measurements of sea level, J.Atmos.
Oceanic Technol., 19, 16901707.
Rusu, E. & Guedes Soares C., 2008. Wave Energy Assessments in the Coastal Environment of Portugal Continental, Proc. 27th International Conference on Offshore
Mechanics and Arctic Engineering (OMAE 2008), Estoril,
Portugal, Paper n.o : OMAE2008- 57820.
Rusu, E. & Guedes Soares C., 2009. Numerical modeling
to estimate the spatial distribution of the wave energy
in the Portuguese nearshore, Renewable Energy, Vol. 34,

pp. 15011516.
Rusu, E., 2009. Wave energy assessments in the Black
Sea, Journal of Marine Science and Technology, Vol. 14,
pp. 359372.
Rusu, E, 2010b. Modeling of wave-current interactions at
the Danubes mouths, Journal of Marine Science and
Technology, Vol. 15, pp. 143159.
Rusu, L., 2010a. A system to evaluate the oil spills propagation in the coastal environmental of the Black Sea,
Journal of Environmental Engineering and Landscape
Management, 18 (4), 288295.
Rusu, L., Ivan, A., 2010. Modelling Wind Waves in the Romanian Coastal Environment. Environmental Engineering
and Management Journal, 9(4), 547552.
Vandemark, D., Gourrion, J., Bailey, S. & Chapron, B.,
2000. An Altimeter Wind Speed Model Using Both
Radar Backscatter and Significant Wave Height, NASA
Technical Report: 20010020922.
621

Multi-purpose Offshore Platform (MpOP)

Nuno Santana
Gabinete de Estudos e Projectos, Arsenal do Alfeite, S. A., Almada, Portugal
ABSTRACT: The possibility of colonizing the seas and oceans is an important step in the exploitation and
sustainable management of marine resources of our planet. With the increasing need for renewable offshore
energy and the need to relocate the existing shoreline fish farms due to local pollution associated, it is imperative
to create infrastructures that support these exploitations in offshore environment with particular concern on
spatial planning, environmental sustainability and economic viability. Based on these assumptions, this paper
describes a new concept of a Multi-purpose Offshore Platform as headquarters of wind farm/wave farm and
fish farm. To be sited in locations where water depth exceeds 50 m, the proposed solution is a semi-submersible
platform, framed type, with a payload of nearly 600 metric tons. Can be fully assembled onshore and towed
to its final location with significant commissioning cost savings. The platform may have multiple layouts with
dedicated areas for work on deck, cranes, accesses for equipments and personnel, ration silos and helipad.
This new concept enables an efficient development of different business areas coexisting on offshore environment, minimizing environmental impact with large economic benefits in infrastructures investment and
management.
INTRODUCTION
Wind farms already exist in shallow waters based in

fixed structures and fish farms are designed to operate
in coastline conditions. These exploitation activities
in deepwater metocean conditions have new and more
demanding requirements that are not yet established
due to pioneer character they still have.
With the oil and gas rigs as reference and considering a scenario of requirements for wind/wave farm and
fish farm, this paper analyses oil offshore platforms
state of the art, suitable location conditions, activities requirements and accommodation type. Based on
that, a technical and economic approach is established and after deciding the suitable offshore platform
type, the payload and layout area necessary to cover
the functional requirements are estimated as well
preliminary main particulars.
Then, to assure that ballast tanks have enough
capacity to cover stability in transport, operation and
survival situations are evaluated loads and ballast
conditions for these main particulars. The results are
acceptable values of air gap and GM, considering small
angle stability.
Logistics and Transports related with Operation
and Maintenance (O&M) are identified for both activities and are considered equipment transfer solutions
and accesses for personnel to support external and
internal logistics.As MpOP is a floater platform, mooring configurations are identified and suggested. In
order to optimize maritime spatial planning, this study
is extended to include others business activities such
as environmental monitoring, R&D, tourism, national
sovereignty and others that may coexist, contributing

to increased knowledge of the seas and oceans.
This paper is structured into 12 chapters, which
shows the methodology used to define preliminary
design of the MpOP. The last chapter summarizes the
conclusions of this study.
STATE-OF-THE-ART
Around 1887, H.L. Williams built a 91.5 meters

long pier out from the beach, near Summerland in
California and built a drilling rig on it. In 1891 the
first submerged oil wells were drilled from platforms
built on piles on the Grand Lake St. Marys in Ohio. In
the early 1930s the Texas Company developed the first
mobile steel barge for drilling in the Gulf of Mexico
coast. The first fixed oil platform beyond the horizon was built in 1947, about 11nm off the Louisiana
shore. In 1996, Shell Oil launched Troll in the North
Sea, Norway at 457 meters high and in 305 meters of
deep. The Troll Offshore Gas Platform is the heaviest man made mobile object ever made, with a dry
weight of the 656,000 tons. In June 2010, there were
over 620 mobile offshore drilling rigs (Jackups, semisubmersibles, drillships, barges) available for service
in the competitive rig fleet. Currently, the worlds
deepest platform is the Perdido in the Gulf of
Mexico, a spar platform floating in 2,438 meters of
water since August 2010.
Nowadays, fixed platforms, compliant towers,
Jack-up drilling rigs, SPAR platforms, TLP and mini
623
TLP platforms, semi-submersible platforms, NUI and

others floating production, storage and offloading
systems are being used in water depths exceeding
450 m. All of these systems are proven technology and
in use in offshore production all around the world.
2.1
Platform Types
2.1.1 Fixed Platform (FP)

Consists of a jacket (a tall vertical section made
of tubular steel members supported by piles driven
into the seabed) with a deck or more placed on top,
providing space for crew quarters, a drilling rig, and
production facilities. The fixed platform is economically feasible for installation in water depths up to
450 m.
2.1.2 Compliant Tower (CT)
Consists of a narrow, flexible tower and a piled foundation that can support a conventional deck for drilling
and production operations. Unlike the fixed platform,
the compliant tower withstands large lateral forces by
sustaining significant lateral deflections, and is usually
used in water depths between 300 and 600 m.
2.1.3 Jack-up drilling rigs
Jack-up Mobile Drilling Units (or jack-ups) are rigs
that can be jacked up above the sea using legs that
can be lowered, much like jacks. They are designed to
move from place to place, and then anchor themselves
by deploying the legs to the ocean bottom using a rack
and a pinion gear system on each leg. These MODUsMobile Offshore Drilling Units are typically used in
water depths up to 120 m, although some designs can
go to 170 m depth.
2.1.4 SPAR Platform (SPAR)
Consists of a large diameter single vertical cylinder
supporting a deck. It has a typical fixed platform
topside (surface deck with drilling and production
equipment), three types of risers (production, drilling,
and export), and a hull which is moored using a taut
caternary system of six to twenty lines anchored into
the seafloor. SPARs are presently used in water depths
up to 900 meters, although existing technology can
extend its use to water depths as great as 2250 meters.
2.1.5 Tension Leg Platform (TLP) and
Mini-Tension Leg Platform (Mini-TLP)
TPL consists of a floating structure held in place by
vertical, tensioned tendons connected to the sea floor
by pile-secured templates. Tensioned tendons provide
for the use of a TLP in a broad water depth range with
limited vertical motion. The larger TLPs have been
successfully deployed in water depths approaching
1200 m.
Mini TPL is a floating mini-tension leg platform of
relatively low cost developed for production of smaller
deepwater reserves which would be uneconomic to
produce using more conventional deepwater production systems. It can also be used as a utility, satellite,
or early production platform for larger deepwater

discoveries. The worlds first Mini-TLP was installed
in the Gulf of Mexico in 1998.
2.1.6 Semi-submersible platform
These platforms have hulls (columns and pontoons)
of sufficient buoyancy to cause the structure to float,
but of weight sufficient to keep the structure upright.
Semi-submersible platforms can be moved from place
to place; can be ballasted up or down by altering the
amount of flooding in buoyancy tanks; they are generally anchored by combinations of chain, wire rope
or polyester rope, or both, during drilling or production operations, or both, though they can also be kept
in place by the use of dynamic positioning. Semisubmersibles can be used in water depths from 60 to
3,000 m.
2.1.7 Normally Unmanned Installation (NUI)
This is a type of offshore facility that is designed to
be operated remotely without the constant presence of
personnel. They are characterized by their small size,
often consisting of just a well bay with a helipad on
top. They are often a compromise of providing the
convenience of surface wellheads, which are easier to
build and maintain, while avoiding the high operating
costs of a full production platform. They are generally only used in shallower water, where constructing
many small NUIs is a relatively easy and cheap option
as compared to the cost of using subsea wells. These
installations will include an emergency shelter with
essential food and water in order to provide a safe
refuge in the event that weather or other considerations prevent a visiting crew from returning to base.
Regular visits may be made for routine maintenance
and for smaller well work such as wireline operations.
3
LOCATION CONDITIONS
There are many aspects that can identify a suitable

location of a wind farm, a wave farm or a fish farm
and it depends on the specific requirements of the
project. Nowadays, a very important factor is minimizing visual impact when the location is chosen.
The landscape from the coast must not change and
the location of the farms must have a minimum distance (D) far away from shore to be out of sight,
including maritime aids to navigation, depending on
the maximum height of structure in water (H).
Considering for wind farms, 250 meters the height to

the tip of the rotor blade and 20 meters the height of the
platform for wave farms and fish farms, the minimum
distances are 30.5 nm and 8.6 nm from shore (for a sea
level observer).
The distance to a deep draft port with capacity to
support heavy and large equipment offloading is a
factor that must not be forgotten in the global project.
624
Figure 1. Some types of offshore platforms.
This will affect directly the cost of all logistics for

installation, O&M and decommission of farms. A
nearby land connection point for transmission of
energy to the grid and the proximity of a fish harbor to
put the harvest fish in the market at low cost are also
an important aspect to be considered.
There are some restricted and forbidden areas to
have in consideration as protected areas, submarine
cables areas, military exercises area, access to ports
and craft fisheries and trawls fisheries.
The western coast of Portugal have good conditions for development and exploitation of wind/wave
farm and fish farm due to the depth water of around
50 to 80 m depth, good weather conditions most
time of the year, water temperature and well developed infra-structures (deepwater ports, shipyards and
grid-points).
The best site depends of type of wave/wind technology and equipments used as well the fish species to
harvest. The environmental conditions are defined by
the wind, wave, current, water depth, soil and seabed
characteristics. For a fish farm we must include water
conditions like temperature, pH, oxygen saturation,
local fauna and seasonal influxes of predators.
4 ACTIVITIES REQUIREMENTS
4.1 Wind/wave farms
An offshore wind/wave farm electrical system consists
of six key elements: Wind and wave turbine generators; offshore inter-turbine cables (electrical collection
system); offshore substation; transmission cables to
shore; onshore substation; and connection to the grid.
Normally, a substation is installed if the farm is large
(100 MW or more); it is far from shore (15 km or
more); or the connection to the grid isnt at collection
voltage (e.g. above 36 kV). Most future offshore wind
farms will be large and/or located far from shore, and
so will require at least one offshore substation. Offshore substations are used to reduce electrical losses
by increasing the voltage and then exporting the power
to shore. Typically serve to step-up the voltage from

the site distribution voltage (30 to 36 kV) to a higher
voltage (100 to 220 kV), which will usually be the
connection voltage. This step-up dramatically reduces
the number of export circuits (subsea cables) between
the offshore substation and the shore. Typically, each
export circuit may be rated in the range 150 to 200 MW.
For projects located far from the grid connection
point, or of several hundred megawatts in capacity,
AC transmission becomes costly or impossible, due to
cable-generated reactive power using up much of the
transmission capacity. In such cases, high voltage DC
(HVDC) transmission is becoming an option. Such
a system requires an AC/DC converter station both
offshore and onshore.
As there isnt any wind farm based in a semisubmersible solution, it was considered a 30.5 nm
offshore wind farm with only 30 floating 3.6 MW
wind turbines. In order to optimize maritime spatial
planning, between wind turbines was considered 20
devices with 1 MW wave turbines. With a power-rating
of 128 MW it will be needed a 130 MVA transformer,
switchgears and other equipment. Cables shall be
installed on a dedicated cable deck and will terminate
at structure through a vertical guide from seabed to
above water level and into conventional switchgear.
The development and design of the substation is
highly dependent on the actual location and the operation and maintenance philosophy for the substation
and for the farm. Several workers will do the daily
O&M of wind/wave farm and their permanent presence will be needed to assure the farm operation
and security. The crew will need dedicated work deck
with crane to do the handling of spare parts as well
as workshop containers and accesses to wind farm
workboats to move inside the farm. Helicopter could
be an economic and fast way to travel and therefore
the platform shall have a helipad.
4.2 Fish farms
The need to work with seven/eight different pellets or
rations over a fish cycle where the fish is placed in a
cage with low weight and are grown to market size,
e.g. 20 g up to 2500 g. In the case of market size limit
reduced to about 600 g, the number of different pellets
could be limited to 4 or 5. Pellets are distinguished by
their composition and size with a density of 0.4 to 0.65.
The diets of hatcheries and juveniles have very low
consumption and thus may be stored in small compartments. The highest caliber and increased consumption
may have to be stored in more than one silo. It will be
considered a number of silos between 6 and 8.
The daily amount of feed to produce 1 ton/day of
fish depends on several factors including: the species,
feed used (0.4% to 1.6% of fish weight), the water
temperature throughout the year and the size of fish to
bring to market. Considering a permanent fish stock
of 2000 tons with a feed used of 1% of fish weight
(different species and market sizes) results in an
average daily ration of about 20 ton.
625
Whereas the conditions of a storm impediment to

supply will last for more than 10 days, the overall stock
value would be 200 tons, value that we add a safety
margin because not only is it difficult to ensure that
the stock is full when the storm starts, as can be seen to
rupture in one or another type of feed, so a global stock
of 200 tons seems a reasonable value for the maximum
capacity of ration storage. Managing the logistic and
weather conditions makes possible the fish farm only
be supplied 3 or 4 times per month.
The feed system is composed of feed blowers that
generates air to the feed lines, silos, feed dosers and
feed selectors that distributes feed for the correct cages
and ensures the feed handling. All the feed procedure
can be remotely controlled with support of cameras
and sensors and operated from the control room.
Would be necessary a laboratory with tanks to control fish health and fish growing as well a container to
store maintenance tools, equipment and harvest tools.
The local management and maintenance of fish
farm will be done by 4 persons minimum and their
permanent presence will assure the O&M and security of the farm. The crew will need a workboat to
move inside the farm.
5 ACCOMMODATION
As multi purpose concept, MpOP must have the ability
to be adaptable to different requirements of different activities and their business models throughout its
useful life.
Assuming this condition, machinery room is an
integrated structure on MpOP and is located in the center of deck. Above the machinery room is the control
room also in the center in a higher position with a 360
view of all farms. The position in the center of main
equipments and controls allows, in damage condition,
the operability of main system of the platform.
The accommodations facilities of the MpOP will
be based on a 20 ISO standard container which allow
a wide range of solutions and the possibility to be
upgraded in the future. This non-structural solution
also ensures an economical transportation by road, rail
and sea as any freight container. It exists in the market high-quality products, manufactured and certified
to the relevant standards and regulations for offshore
service container that include several equipment and
outfitting as telecommunications, electrical and water
installations, HVAC, fire protection and safety as well
as conditions for installation plug and play on board.
Crew accommodations, kitchen, freezer, WC, waste
management, several storage rooms, workspaces, laboratories, offices, control rooms and laboratory are
the minimum standardized containers needed for the
MpOP operation.
and some can be operated in ultra deepwater. A suitable location will be sited over 30.5 nm far from shore,
and therefore, in Portugal the depth for this distance is
over 50 m. For this distance, the installation and commissioning costs of fixed structures are much more
expensive than in shallow waters. This means that a
floater type will be the suitable type due to it can be
sited in a no depth limit location or seabed type and, as a
vessel, can be towed for installation, relocation, maintenance/repair and decommissioning with no need of
offshore barges and offshore cranes.
When comparing floater types, wave and wind
induced motions are not the only elements of performance to consider. Economic viability plays a
significant role and is important to carefully analyze
construction, installation, commissioning, decommissioning and ease of access and suitable conditions
for O&M.
In general terms, spar type has better heave performance than semi submersibles due to its deep draft and
reduced vertical wave-exciting forces, but it has more
pitch and roll motions since the water plane area contribution to stability is reduced. TLP has very good
heave and angular motions, but has a complex and
expensive mooring system related to installation and
operation. Semi-submersible can offer good heave, roll
and pitch performances with simple mooring configurations at a relative low cost. NUI have an interesting
design concept due to small size, be operated remotely
and be easier to build and maintain, avoiding the high
operating costs.
Combining a semi-submersible with a NUI concept
the result is a small platform with good response to
wave and wind induced motions, easier to build with
low cost of installation and commissioning and low
cost of O&M (most of the task operated remotely)
with a simple but reliable mooring system. Deep-water
semi-submersibles state of the art is actually in 6th generation with 4, 6 or 8 columns and 2 pontoons (braced)
or ring pontoon (framed). The conceptual approach
6 APPROACH
Nowadays, all types of state of the art oil and gas offshore platforms are developed to operate in deepwater
Figure 2. 3D preview of MpOP concept approach.
626
is a semi-submersible platform with 4 columns and a

ring pontoon.
The MpOP payload in operation condition will be
the necessary to satisfy the sum of wind/wave farm,
fish farm equipment, transport and logistic considered:
Wind/Wave farm 220.0 tons; Fish farm 205.7 tons;
Accommodations 65.5 tons and Logistics 108.8 tons.
The MpOP has a payload of 600 tons.
MAIN PARTICULARS
Important aspects to be considered on preliminary

design of the main particulars are the economic aspect
related to a near shipyard with production capacity onshore, the possibility of transport and type of
transport to offshore location and control at this stage
the stability of platform on transport, operational and
survival conditions.
There was developed a program in order to establish
main particulars with control of different variables and
results. Introducing, as input, overall dimensions of
columns and pontoon, topside weight, fresh water, fuel
and fish ration capacities, draft, ballast and mooring
lines weights, the program returns as output deadweight, air gap (freeboard), center of gravity to keel
(KG) and metacentric height at small angles (GM).
Naturally, the platform must have the needed
dimensions to cover assumed requirements: machinery room and control room at center, substation,
dedicated work decks, 13 containers and walkways
(with the possibility to upgrade), helipad, ration silos
integrated in platform columns and approximately 600
tons of payload. The first layout area approach was
considering a deck area of approx. 1000 m2 . As the
platform will be framed type (ring pontoon), a square
deck of 35 35 m overall is considered.
The fact that the platform has a square deck means
a cost saving in project and production of the platform
due to the symmetry in YZ and in XZ plane. Platform
can be projected and be built with a block strategy
considering equal columns and pontoons elements,
like a production line.
Two different configurations of columns have
been considered: Rectangular cross-section or circular
cross-section. Circular cross-section drag coefficient
(CD ) is lower than rectangular cross-section CD but
assuming same mooring system, same water plane
area and same draft for same deadweight and comparing both configurations the metacentric height for
the square column configuration is higher than that
of the circular column vessel. In addition, rectangular
columns are much easier to manufacture than circular columns. Furthermore, it will reduce global costs
of production due to optimisation of nesting, easier
welding and production control as well as production
tasks becoming more effective. These factors led to the
decision to choose a square column configuration.
Assuming an air gap of 10 m, structure weight was
estimated as well as equipment weights needed. It
will be necessary a volume displaced of columns of
5 5 6 m and pontoon of 25 5 3 m to cover estimated 2000 ton of operation deadweight. It will be

necessary to store 625 m3 of fish ration, which corresponds to 200 tons, distributed for 8 different silos.
Fish ration will be stored inside columns above operation water line. The CG of this load is below deck line
and reload operations are much easier to do through
a hatch on deck level than by a hole on topside of
silos on deck. The column volume above water line is
1000 m3 that cover the 625 m3 needed for fish ration
plus volume for structural stiffeners.
8
STABILITY
A stable and upright position in the water is important for the safety of the crew, equipment and for a
proper transport, operational and survival conditions.
To keep the CG as close to z-axis as possible, primary
and secondary ballast tanks are located in pontoons.
All tanks have vertical divisions to reduce free surface
effects on stability.
8.1 Transport condition (wet towing)
The platform could be 100% build in a shipyard and
then wet towed to offshore location, saving cost in
installation and commissioning with offshore barges
and offshore cranes. One typical hazard that may affect
semi-submersible platform installation during towing
is loss of stability in bad weather conditions. The water
plane area can drop suddenly, GM drops consequently
and the platform can capsize. As the depth of pontoon is 3 m, towing the platform with less of 3 m
draft condition, even if possible, could be dangerous.
Considering 5m draft, corresponding 30% ballast, GM
for this condition is 4.79 m against 68.10 m of 0%
ballast. The drag of Towing 2 condition is higher than
Towing 1 condition but this one is a safer condition to
tow the platform, see Table 1.
8.2 Operation condition
After installation, primary ballast tanks will be partial
filled with the necessary ballast to reach the operation
draft. During normal operation, platform CG will shift
due to varying distribution of the daily consumed and
restored fish ration, fresh water, fuel and equipment/
spares/helicopter as well as external effects from wind
Table 1.
Preliminary main particulars of MpOP.
Main particulars
LOA
Beam Overall
Operating draft
Air gap
Areas:
Main deck
Work deck
627
Capacities
35
35
6
10
m
m
m
m
1225 m2
300 m2
Fish ration
Fresh water
Fuel
Helideck
Crew
Spare parts
Payload:
200
30
70
7
20
15
600
ton
ton
ton
ton
person
ton
ton
(wind/wave farm and fish farm) needs to O&M,

i.e. personnel and supplies, spare parts and tools, fish
ration, fresh water and fuel. The following ways could
do the transport associated to this logistic. Personnel could travel by an offshore vessel supply (OVS),
dedicated fast boat or by helicopter. Supplies, equipment and spare parts, fresh water and fuel will be
supplied by OSV. Fish ration could be transport by
OSV or fish boat. It depends of the logistic strategy
assumed for this exploration.
Internal logistic is related to the O&M of wind/
wave farm and fish farm. It will be done by smaller and
faster workboats and will transport personnel, tools
and spare parts. The pellets to feed fish are transferred
by appropriate airflow system (hoses) and ocean cables
transfer the electric energy.
Table 2. Transport conditions, Towing 1 and Towing 2.
Load (ton)
Ballast (ton)
Deadweight (ton)
Draft (m)
Air Gap (m)
KG (m)
GM (m)
Table 3.
0% ballast
Towing 1
30% ballast
Towing 2
0
0
1.650
2
14
10.00
68.10
0
380
2.050
5
11
8.31
4.79
Operation conditions, Operation 1 and Operation 2.

100% load
Operation 1
10% load
Operation 2
300
194
2.155
6
10
8.83
3.91
30
473
2.155
6
10
7.87
4.87
10 ACCESSES
Load (ton)
Ballast (ton)
Deadweight (ton)
Draft (m)
Air Gap (m)
KG (m)
GM (m)
This condition has only been considered in a

perspective to reserve some ballast volume in tanks
to balance the platform in that situation and so, other
considerations must be done. During survival conditions (if damaged, with flooded column), the platform
is evacuated, ocean cables are unplugged and some
loads could be loosing free (fish ration). In order to
compensate for the loss of weight and balance the
platform, the secondary ballast tanks can be completely filled with a total capacity of 300 ton.
Transferring personnel and equipment in an offshore

environment could be a dangerous operation due to
the proximity of vessel and platform necessary to
accomplish the operation, the sea and weather conditions and the different heave and roll motions of vessel
and semi-submersible platform.
An offshore crane usually accomplishes transferring equipment from vessels to and from an offshore
platform. The MpOP has two cranes on opposite corners in order to assist all deck platform perimeter.
Both corners have a dedicate work deck area of 150 m2
each to works and maneuverings. The vessel is docking to columns of the platform that are protected with
fenders. Normally in oil industry, to transfer personnel
from platform to and from a vessel is used the equipment crane to lift a personnel basket. Some baskets
can transfer up to 4 persons.
Transferring personnel from workboat or vessels
to and from a platform is usually accomplished by a
swing rope or a personnel basket. These two devices
have been used by the oil industry and they can be used
very safely. The MpOP has piers attached to columns
positioned 3 meters above operation water line where
workboats can dock and the personnel can be transferred by swing rope or by vertical ladder, depending
of sea condition. As appendage structures, the piers
are build from tubular elements and in that way are the
most transparent to sea state possible.
Transfer personnel by helicopter is faster than by
boat and for this distances could be less expensive too.
The heli-deck will be in a high position far away from
control room.
11
and waves must not compromise the stability of the

platform. Primary ballast tanks will balance and/or
compensate the platform to assure this condition.
Maximum load condition of consumables goods
(Operation 1) comprises fresh water 30 ton; fuel
70 ton; fish ration 200 ton. Considered minimal limit
condition of consumables to keep wind/wave and
fish farms in operation (Operation 2) comprises fresh
water 3 ton; fuel 7 ton; fish ration 20 ton, as shown in
Table 2.
8.3
Survival condition
LOGISTIC AND TRANSPORTS
It is possible to identify two distinct types of logistic

that are common to wind/wave farm and fish farm;
external logistic and internal logistic.
External logistic is related with onshore supply
of personnel, equipments and goods that platform
MOORING CONFIGURATION
The moorings lines, which apply forces to the system

(self-weight and traction), have to be considered to
calculate the center of gravity and any ocean cable
or hose/pipe connected to the platform should be
considered.
628
There are some suitable configurations of mooring systems that can be applied to a framed semisubmersible platform. We can divide in two categories:
spread mooring (sm) and dynamic positioning (dp).
Catenary Mooring (sm) The mooring lines of a
free hanging catenary mooring arrive horizontal to the
seabed so that the anchor point is only subjected to
horizontal forces. The restoring forces are mainly generated by the weight of the mooring lines returning the
system to equilibrium.
Multi-Catenary Mooring (sm) The catenaries
mooring lines incorporate weights and buoys to form
S or Wave type configurations.
Taut Spread Mooring (sm) The mooring lines of
a Taut Spread Mooring arrive, typically at an angle
to the seabed with the anchor point capable of resisting horizontal and vertical forces. The restoring forces
are mainly generated by the elasticity of the mooring
line. The mooring lines of a TLP are orthogonal to the
seabed, with the restoring force mainly generated by
the change in buoyancy of the topside structure.
Active Mooring (dp) Active Mooring consists
of mooring lines which are spread around the floating structure, where the inboard end of each mooring
line is held by a servo controlled winch. A central
computer tensions or loosens the mooring lines in
order to keep a fixed seabed position.
Propulsion (dp) The technique consists of
positioning a floating structure above a fixed seabed
point by the use of propellers or thrusters that are
controlled from a central computer.
As a semi-submersible the mooring configuration
will be with catenary spread with large motions vertically and large motions horizontally.Any options could
be a suitable configuration depending of equipment
technical requirements for heave, sway, roll, cost and
seabed type. At this preliminary design stage we cant
define what is the best mooring configurations to use
but, for now, dynamic positioning is excluded due to
the cost of equipments and O&M associated.
12
The platform can be a base where navy ships can

refuel and continue missions of National sovereignty.
Could be used as an Offshore Dive and ROV training center or as an Eco-tourism hotel allowing long
offshore stays with very good conditions for diving
and other recreational activities.
13
CONCLUSIONS
13.1 Location
Minimum 30.5 nm far from shore in order to mitigate
visual impact. For this distance, in Portugal, water deep
is around 50 m to 80 m.
13.2 Wind/wave farm_requirements
130 MVA transformer, switchgear and other
equipment 200 tons, offshore crane, workboat (RIB),
4 permanent members crew accommodation facilities,
spare parts store container, helipad.
13.3 Fish farm requirements
Silos for 250 tons of ration, 625 m3 of capacity; 10
days of storage; 8 silos; fish feeding system; offshore
crane; workboat (RIB); and 4 permanent members
crew accommodation facilities, spare parts container,
helipad.
13.4 Accommodation
Machinery room and control room are structural
and accommodations facilities are based on 20 ISO
standard containers. There are Multi layout possibilities to meet different business model.
13.5 Payload
MpOP has a total payload of 600 tons (Wind/Wave
farm 220.0 tons; Fish farm 205.7 tons; Accommodations 65.5 tons and Logistics 108.8 tons).
OTHERS ACTIVITIES
13.6 Stability
Others business activities could be present in the

platform in order to optimize the offshore infrastructure investment and increase knowledge of oceans
minimizing costs and environmental impact. Site
environmental monitoring is an important activity to
support wind/wave and fish farms O&M. The sitecollected data of a weather station and a hydrograph
station could be integrated in related national institutes
increasing that way the accuracy of weather forecast
and sea knowledge of that area.
R&D could be present too in several fields of
study: meteorology, hydrology, aquaculture, biology,
chemistry, geology, ecology, marine biotechnology,
engineering and many others. A small science team
on board, equipment and a laboratory make possible
to develop R&D in site, at low cost, what could only
be possible by an appropriated boat.
Platform has positive GM (at small angles intact

stability) to be towed to offshore location (100% build
in shipyard saving installation and commissioning
costs) and for operation conditions.
13.7 Logistics and transport
There are two distinct types of logistic associate with
activities. External Logistic is related with onshore
supply of MpOP and Internal Logistic is related with
O&M of farms. Fuel and fresh water will be need.
13.8 Accesses
2 Offshore cranes to transfer equipments and goods.
Tubular piers attached to columns only to transfer
personnel.
629
with a deck area of around 1000 m2 free to use,

a payload of 600 tons and multiple layout possibilities.
REFERENCES
Figure 3. 3D preview of a possible layout of MpOP.
13.9
Mooring configuration
Platform will be moored with a catenary or a taut

spread lines.
13.10
Others activities
Environmental monitoring (weather/hydrograph

station); R&D; National sovereignty; Eco-tourism,
Dive and ROV training center and others.
The MpOP as a semi-submersible offshore platform
can be towed to location, which represents low cost
of installation, commissioning and decommissioning.
Columns and pontoons have rectangular cross-section
in order to minimize production costs. Platform can
be projected and be built with a block strategy considering equal columns and pontoons elements due
to symmetries as well deck due to standard accommodations facilities. Fish ration will be stored inside
columns. Rations CG is below deck line and reload
operation are easier to do at deck level. The MpOP
is a solution to step forward into offshore deep-water
multi exploitation, a Multi-purpose Offshore Platform
AkvaGroup Fish Farms (http://www.akvagroup.com)

AWS Truewind, Offshore wind technology overview
NYSERDA PON 995, Task Order no.2, agreement no.
9998 Sep 17, 2009
Basic Ship Theory K.J. Rawson and E.C Tupper
Longman Scientific&Technical
Digenes Modules Offshore Containers (http://www.
diogenes-modules.com )
ImarE/RINA Joint Offshore Group International
Conference Offshore Safety: Protection of Life and
The Environment, London 2021 May 1993
Conference Installation of Major Offhore Structures
and Equipments, London 1718 Feb 1993
Conference Offshore Operations Post Piper Alpha,
London 68 Feb 1991
Instituto Hidrogrfico Manual de Navegao (Clculos
Nuticos) Lisboa Ministrio da Defesa, Marinha 1989
OffshoreTechnology 7. Electrical (http://www.offshorewind
energy.org/ca-owee/indexpages/Offshore_technology.php)
RWE npower renewables (http://www.rwe.com/web/cms/
en/324354/rwe- npower - renewables/sites/projects-inconstruction/wind/greater-gabbard-offshore-wind-farm/
construction/)
Siemens Wind farms References (http://www.energy.
siemens.com/us/en/power-transmission/grid-accesssolutions/references.htm)
The Institute Of Marine Engineers Conference Proceedings
Design and Safety Assessment for Floating Installations
Volume 107, n. 1 1995
The Society of Naval Architects & Marine Engineers
Principles of Naval Architecture John P. Comstock 1667
Wave Energy Centre Potencial e Estratgia de desenvolvimento da energia das ondas em Portugal, Relatrio
(2004)
Wind Energy The Facts (http://www.wind-energy-thefacts.org)
630

Development of an integrated system for personnel and

equipment transfer to offshore wind turbines
T.A. Santos & E. Carichas
Rinave, Registro Internacional Naval, Lisbon, Portugal
N. Fonseca & J. Pessoa

F. Duarte
Associao das Indstrias Navais, Lisbon, Portugal
J. Abreu Valente
Euroshide, Lisbon, Portugal
L. Baptista
Veranavis, Lisbon, Portugal
J. Cruz & M. Leal

Ocean, Lisbon, Portugal
ABSTRACT: The British organization Carbon Trust promoted a competition within the Offshore Wind
Accelerator framework for solving one of the critical issues concerning the operation of far offshore wind
farms. The maintenance of 200 turbines in the new offshore wind farms which will be built around the United
Kingdom poses a considerable challenge, as turbine access and work platforms will be located 20 m above the
water in open sea conditions, up to 300 km from the coast line. This paper describes an integrated system for
personnel and equipment transfer (ISPET) to the offshore wind turbines.
The system consists of a swath ship with an ample aft deck where up to 6 capsules, inverted cone shaped, are
stowed in berths. In seastates between 1 m and 3 m wave height, each of these capsules is to be moved to the wind
turbine platform, carrying up to 6 persons and 1t of equipment, using a shipboard crane when the ship is suitably
located alongside the wind turbine at a safety distance. The wind turbine work platform is fitted with a 360 ring
where the capsule is to be deposited using the crane and left during the duration of the maintenance operations.
The main benefits of this system are that it is capable of providing access to wind turbines in seastates up to 3 m
and personnel and cargo are transferred protected inside a capsule. Costs are relatively low as a wind farm with
200 turbines can be serviced by 23 ships.
INTRODUCTION
The British organization Carbon Trust promoted a

competition within the Offshore Wind Accelerator
framework for solving one of the critical issues concerning the operation of far offshore wind farms. The
maintenance of 200 turbines in the new offshore wind
farms which will be built around the United Kingdom
poses a considerable challenge, as turbine access and
work platforms will be located 20 m above the water in
open sea conditions, up to 300 km from the coast line.
In order to increase the availability of the turbines
and therefore maximize energy production, maintenance teams should be able to access these platforms
in up to 3.0 m significant wave height. Existing access
techniques, that is, small boats or helicopters, allow

access in up to 1.5 m significant wave height. The
challenge put forward by the Carbon Trust is to extend
access at least to 3.0 m waves, 30 knots wind and
3 knots current.
The Portuguese Maritime Industries Association
(AIN) brought this competition to the knowledge
of its associates and a number of companies and
organizations, organized in the AIN Ship Design Working Group, accepted the challenge to develop an
access system for maintenance teams to wind turbine
work platforms and submit this Portuguese bid to the
Carbon Trust competition.
The proposed solution was named Integrated
System for Personnel and EquipmentTransfer (ISPET)
631
and consists of a 57 m length overall swath ship with

an ample aft deck where up to 6 capsules, inverted
cone shaped, are stowed in suitable berths. In seastates between 1.0 m and 3.0 m wave height, each of
these capsules is to be moved to the wind turbine
platform, carrying up to 6 technicians and 1 t of equipment, using a shipboard crane when the ship is suitably
located alongside the wind turbine at a safety distance
(transfer position).
The wind turbine work platform is fitted with a
360 ring where the capsule is to be deposited using
the crane and left during the duration of the maintenance operations. Recovering the capsule is carried out
using the same principle. For seastates below 1.0 m
wave height, the ship can deploy a gangway from
the top of bridge directly to the turbine platform or
rigid inflatable boats can be launched easily (in those
sea conditions) and access of maintenance team
carried out through the conventional ladder fitted on
the turbine foundation.
The fundamental issue determining the success of
the system is the capability to place the capsule in
the turbine work platform in a 3.0 m seastate. The
reason for choosing a swath hull is because it is possible to control its vertical motions. The very small
waterplane area results in small vertical wave exciting
forces (as opposed to conventional ships) which can
be controlled by active systems (thrusters and/or stabilizing tanks). It is also possible to introduce passive
systems (hydrodynamic stabilizers) to significantly
increase the damping of the vertical motions and in
this way reduce dynamic amplifications.
Seakeeping calculations demonstrated that one or
more of these equipments need to be fitted in the swath
ship as motions at the tip of the crane are still significant due to the very high position of the work platform.
In this respect, the large size of the area inside the
rings of the platform and comparatively small size of
the bottom of the capsule (cone shaped) facilitate the
positioning of the capsule, but this may not be sufficient for easy and smooth operation and therefore
some ship stabilization is requested.
The main benefits of this system are that it is capable
of providing access to wind turbines in seastates up to
3 m as the ship can be located at a safety distance from
the turbine head on to the incoming waves, limiting the
ship motions to heave and pitch. Furthermore, personnel and cargo are transferred protected inside a capsule. Cost is relatively low as a 200 turbines wind farm
would require 2/3 ships, 18 capsules and a ring fitted in
the work platform of each tower. Each ship, including
the 6 capsules, may cost up to 21 million euros.
The ISPET system has been submitted for patenting by the companies and organizations involved
in its development. The steps necessary to bring
this system to commercial use are: detailed analysis
of maintenance requirements for fleet dimensioning;
research into seakeeping, aerodynamics and structural
problems (including numerical analysis and model
testing); detail engineering design of ship, crane,
capsule and platform; manufacturing and testing of
prototype.
Figure 1. Most common Access methods.
2 ACCESS SYSTEMS
2.1 Transfer of personnel
Current access methods to offshore wind turbines
include mostly the use of small craft of single hull or
catamaran types. Helicopters are also to used to access
the nacelle directly when weather conditions prevent
access by boat, but are only capable of accommodating 5 technicians. Typically wind farm operators will
contract the helicopter operator for a number of days
of use a year. However, this access method is dangerous and a number of accidents have been registered
including fatalities. In the oil/gas offshore industry,
access by means of baskets is also common.
In the North Sea, early experience of offshore wind
has shown that access to a wind turbine by boat is
challenging in waves of around 1 m height or more.
Currently most standard boat transfers cannot be performed in sea states where the significant wave height
is greater than 1.5 m and wind conditions are in excess
of 12 m/s. In exposed locations, such as in UK and Irish
waters, the average number of days where the wave
height is greater than 1.5 m is considerably greater.
Operating in concert with the wave height restrictions are restrictions from the water depth, swell and
underwater currents. Shallow waters amplify the local
wave height and can significantly change the wave
form characteristics. Generally, where a turbine in a
wind farm is located in the shallow water, this turbine
will present the most access problems.
Wave data (UK offshore wind farm sites) shows
that access using a standard boat and ladder (significant wave heights up to 1.5 m) is possible 80 per
cent of the available time. However this accessibility
rate is too low for good overall wind farm availability. Accessibility is defined as the percentage of total
available time a turbine can be accessed.
In winter, accessibility is typically worst when there
is the greatest likelihood of turbine failures; yet at
these times there are higher winds and hence potentially higher levels of production loss. Accessibility
can be improved to above 90 per cent if access is
made possible in significant wave heights between 2.0
and 2.5 m.
Providing access in yet more extreme conditions is
probably too challenging considering cost, technical
difficulty and safety. A safety limit on sea conditions has to be set and rigidly adhered to by the wind
farm operator. This implies that 100 per cent accessibility to offshore wind plant will not be achievable
632
Table 1. Reference wind farm and environmental

conditions.
Figure 2. Vessels currently operated to access offshore

wind turbines.
Number of wind turbines per service area

Wind turbines
Distance between each wind turbine
Average Water Depth
Distance from port to wind farm
Area (foot-print) of wind farm (N-S, E-W)
Shape of the wind farm
200
5 MW
1 km
45 m
up to 300 km
20 km 8 km
See Figure 4
Tidal range
Current speed
Wind speed
Wind direction
Wave period
Wave direction
up to 5 m
up to 3 knots
up to 30 knots
variable
up to 25 s
variable
These craft are deployed next to the turbine foundations and ladders are used by the technicians to climb
to the turbine work platform. Figure 2 shows some
examples of these crafts.
2.3 Innovative access systems
A number of innovative systems have been proposed
in recent years aiming at extending the accessibility limits as shown in Figure 3. Innovations fall in a
number of different categories: articulated gangways,
stairways, elevators, articulated arms. Most innovative
concepts are still at a development stage except for
the Amplemann system which is already undertaking
trials.
3 WIND FARM MAINTENANCE
REQUIREMENTS
3.1 Reference wind farm and wind turbine
Figure 3. Innovative access methods.
and 90 per cent accessibility seems a reasonable target. Improvements in availability thereafter must be
achieved through improved system reliability.
2.2 Vessels
A key requirement for the vessels employed in access
to offshore wind turbines is that these vessels are
robust and can operate in adverse weather conditions.
In many cases, aluminium catamarans approximately
20 m in length, with capacity for 12 technicians are
being used. Vessels speed can be in excess of 20 knots.
The main characteristics of a reference wind farm

where given in the technical specification of the
Carbon Trust competition. The environmental conditions and wind turbine main characteristics were also
specified and are shown in Table 1.
Figure 4 shows the lay-out of the reference wind
farm. Note that any ship for service in this farm is to
have autonomy for going from the operational base to
the wind farm, sustaining a number of days of continuous operation in the wind farm and returning to the
operational base.
Figure 5 shows the main particulars of wind
turbine, being worth noting that the height from the
waterline (lowest astronomical tide) to the work platform is 20 m, and this is the height to be reached by
any proposed innovative access system.
3.2 Maintenance requirements
For calculating the maintenance requirements, it will
be considered that each wind turbine suffers one fault
per month and needs two planned maintenances a
633
Figure 4. Lay-out of reference wind farm.
year. This means, for the reference wind farm, 2800

unplanned and planned maintenances a year. Of the
unplanned maintenances (2400), 70% require one
access (hours), 27% require one access (a day) and
3% need 4 accesses as these are serious faults. Each
planned maintenance means 2 days and therefore 2
accesses and there are 2 planned maintenances a year.
The total number of accesses per year will be 3416.
It is assumed that maintenance teams work during
day and have to access and be recovered from the
turbine each day. The reference wind farm will therefore require 11 accesses/day in each day with less than
3.0 m significant wave height (310 days).
The number of accesses to be carried out in each
day varies during the year, being in fact a stochastic
variable, dependent on weather and number of turbine faults, which are more frequent and more severe
in winter. It can be estimated that in winter with Hs
between 1.5 m and 3.0 m, there are on average 16
accesses a day to be carried out. In summer, with Hs
less than 1.5 m there will be on average 12 accesses
are to be carried out per day.
Taking into account that in winter there will be
approximately 8 hours of light and in summer 16 hours,
the required fleet can be estimated at 4 ships in winter, each performing 4 accesses a day. In summer,
2 ships are sufficient, performing 6 accesses a day.
If the ships can perform transfers at night, 2 ships
might be sufficient for servicing the reference wind
farm in winter. Possibly a third one will be required for
covering maintenance and repair times and allow for
crew rest.
INTEGRATED SYSTEM FOR PERSONNEL

AND EQUIPMENT TRANSFER
4.1 Concept design
Figure 5. Wind turbine characteristics and lay-out.
Taking in consideration the reference problem

described in the previous section, the conceptual
design of a vessel capable of performing the mission
was undertaken. The new design was named integrated system for personnel and equipment transfer
(ISPET). The system consists of a swath ship with an
ample aft deck where up to 6 capsules, inverted cone
shaped, are stowed in berths. In seastates between 1 m
and 3 m wave height, each of these capsules is to be
moved to the wind turbine platform, carrying up to 6
persons and 1t of equipment, using a shipboard crane
when the ship is suitably located alongside the wind
turbine at a safety distance. The wind turbine work
platform is fitted with a 360 ring where the capsule
is to be deposited using the crane and left during the
duration of the maintenance operations. Recovering
the capsule is carried out using the same principle. For
seastates below 1 m wave height, the ship can deploy a
gangway from the top of bridge directly to the turbine
platform or rigid inflatable boats can be launched and
access carried out through ladder fitted in the wind
turbine foundation.
634
Table 2.
Figure 6. Integrated system for personnel and equipment

transfer.
Main particulars.
Length overall
Beam
Draught
Depth main deck
Displacement
Work deck area
Maximum speed
Economical speed
Electrical plant
Propulsion plant
Bow thrusters
Fuel oil
Fresh water
Water ballast
57.0 m
25.0 m
7.0 m
14.5 m
1770 t
300 m2
16 kn
12 kn
5 800 kW
2 1800 kW
2 300 kW
170 m3
90 m3
770 m3
provided with dynamic positioning system (class II),

integrating the two bow thrusters and the two pods.
Ship shall be fitted with a crane for capsule handling, a gangway for direct access to wind turbine work
platform and 2 Ribs for wind turbine boarding using
wind turbine ladder. Powerful searchlights to be provided allowing the ship to operate without daylight
in order to extend working time in the winter time.
There will be an emergency helicopter winching area
located forward.
The main particulars of the ship are shown in
Table 2. Figure 8 shows the general arrangement of
the ship.
These main particulars have been checked against a
database of some 50 swath vessels built since 1969 to
2010. The proposed ship is fitting well in the general
trends, as shown in Figure 9, where the black losangue
indicates the vessel. Heickel coefficient is defined as
(standard units in SI system):
Figure 7. Side and bow profiles of ship.
4.2
Preliminary design
Critical for the success of the design was that the vessel
has as low as possible motions in waves. This requirement leads to the selection of a swath type of hull form,
as shown in Figures 6 and 7. The ship is to have capacity to perform transfer operations in the wind farm
during one month. She will carry up to 6 capsules in
an ample aft deck on the port side (300 m2 ) equipped
with special rails, similar to those of the platform, for
storing and fixing the capsules. The accommodation
will be for a crew of 40 members which will be composed by 16 seamen, with double cabins for sailors
and single cabins for officers and 24 technicians (in
double cabins) belonging to the maintenance teams.
The ship is capable of carrying 20 t of spare parts
and equipment and has consumables for an endurance
of 30 days. The ship shall be capable of making 16
knots for transit to and from the wind farm (10 hours).
Within the wind farm the ship will operate at a reduced
safety speed of 12 knots. She shall have diesel electric propulsion with 5 gensets and 2 pods. Ship to be
Furthermore, the air gap between the bottom of the

wet deck and the full load waterline was found to be
around 3 m, for the larger ships, and this was kept in
mind in the current project.
4.3 Regulatory and class requirements
The application of a swath hull form in a offshore
support vessel is innovative in many respects. Furthermore, since it is larger than ships currently employed
in offshore turbine maintenance, rules and regulations
to be complied with are largely unkown or unsuitable
for this particular vessel.
Smaller vessels in the UK typically comply with
the MCA Small Work Boat Code. However, this vessel is above the 24 m and 500 GT thresholds and
should fall in the scope of Load Line, Solas and
Marpol conventions.
Given that the ship will perform offshore duties,
another option would be to use the Guidelines for
635
Figure 8. General arrangement.
Offshore Supply Vessel, IMO-MSC (2006). However,

these guidelines are not applicable to ships carrying
more than 12 industrial personnel and this might be the
case if the maintenance teams are taken as industrial
personnel. Regarding intact stability, these guidelines
prescribes the Intact Stability Code. Damage stability

is according to SOLAS for cargo ships with a one
compartment standard.
Other option could be to comply with the Special Purpose Vessel Code, IMO-MSC (2008), with
636
Taking in consideration that the ship has a swath

hull, intact and damage stability would probably have
to be calculated using the High Speed Craft Code,
which contains criteria appropriate for catamaran
and swath type of vessels, notwithstanding the fact
that this ship will not be a high speed craft.
Clearly, in the future, with an expanding fleet of
ships of this type, the need will arise for developing rules and regulations tailored to this type of ship.
At the moment, rules and regulations would have to
be concerted with classification societies and national
authorities in detail.
In what concerns class notations for this type of
vessel, an example which might be offered by a
classification society would be that shown below.
In this case the hull, machinery, electrical installation

and fire protection would be designed to the standard
rules for steel ships, Bureau Veritas (2009). The stability and wet deck structure would be calculated using
the rules for high speed craft, Bureau Veritas (2002),
and seakeeping could be evaluated, with adaptations,
using the relevant part of the rules for the classification
of naval vessels, Bureau Veritas (2006).
Furthermore, ship is to comply with vibration and
noise rules and regulations and the crane and capsule
are to be certified by a Classification Society.
4.4 Access to offshore wind turbine in
good weather
Figure 9. Main particulars of swath type vessels.
the special personnel being the 24 technicians. Damage stability would then be according with Solas as a
passenger vessel but without need for applying safe
return to port requirements.
4.4.1 Gangway
The ship has on the bridges top structure a specially designed gangway with 3 degrees of freedom,
consisting on two articulated ramps. The gangway is
hydraulically operated and once positioned, it deploys
personnel and small equipments directly at platform
level (20 meters high at L.A.T.). This gangway is
operable in sea states up to 1 meter wave height.
The combined weight of the gangway structure and
auxiliary parts is about 7 tonnes. The dimension of
the gangway fully extended is about 12 meters plus
additional 2 meters of the stand by platform.
The gangway tower is located on the portside of the
bridge top, above a reinforced bridge wing structure.
The gangway can easily reach the tower ring platforms
landing areas with a 7 meters clearance between the
637
Figure 10. Access using the gangway (in good weather

conditions).
ships side and the platform end and with a clearance

of 15 meters between ships side and the tower itself.
The estimated time of deployment of the gangway (up to 1 meter wave height) is about 40 seconds. The time of recovery is about 15 seconds. The
transfer time of personal and equipment is variable
but it can be as fast as 40 seconds (2 crew team). The
Gangway operates in the 180 arc range of the portside and has freedom of movements until a certain
degree of motion. The gangway works with all type
of foundations and even with different types of tower
platforms.
4.4.2 Service Rigid Inflatable Boats (R.I.B.)
The ship is equipped with two service R.I.B. (and
associated davits). Each of these small crafts are 7
meters long, weighting 3 tonnes and transporting up
to 7 crew and their respective equipments. The R.I.B.
will operate in sea states of up to about 1 meter wave
heights and have a range of about 50 miles.
These service crafts will be lowered by their davits
to the water and proceed, with personal and small
equipments, towards the towers requiring maintenance. Once arriving the tower structure the personal
is disembarked and will climb the tower ladders until
reaching the work platform. It might be required the
use of safety lines by the climbers in order to prevent
falls. These hook lines should be attached to a cable
that runs alongside the ladders.
Figure 11. Access using the crane and capsule (in bad
weather conditions).
The crane then lifts the capsule and rotates from

ship to 90 , unfolds and lifts its extreme. Capsule
is then positioned over the landing area (area comprised between the two rings in the work platform). The
capsule is then lowered slowly towards the landing
area. Several attempts may be necessary in very bad
weather. The capsule is finally berthed in landing area
and the cable is payed out.
Maintenance team member gets out of capsule
through walkway and opens passage to tower platform. Other team members disengage cargo brakes.
The crane pays-out cable. The maintenance team pulls
cargo from capsule to tower platform. At the same
time the crane rotates to ship, pays-out cable. The ship
is then ready to depart from transfer position and the
maintenance team is left in the turbine to start working.
The recovery of the maintenance team is expected
to take about half an hour. The ship approaches the
tower heading towards waves and wind. Maintenance
team members stow cargo on capsule and take seats
in capsule. As the ship is in the transfer position, the
crane is unfolded and deployed. The hook is attached
to the capsule and the crane lifts the capsule and hoists
it fully clear of landing area. The crane then rotates
towards ship and unfolds, positioning the capsule over
its assigned berth. The crane then lowers the capsule
to the assigned berth. The ship can then depart from
the transfer position. Maintenance team comes out of
capsule and with aid of deck team brings cargo to deck
of ship.
4.5 Access to offshore wind turbine in bad weather

4.5.1 Launching and recovery operations
The launching of a maintenance team comprises the
operations described below and should take about
40 minutes. These operations in the vicinity of the
turbine tower need to comply with the Code of Safe
Practice for the Carriage of Cargoes and Persons by
Offshore Supply Vessels, IMO-MSC (1997).
The ship approaches the tower heading towards
waves and wind (from 100 m to the tower). Maintenance team members enter into capsule and deck team
prepares capsule for transfer. The crane is positioned
over the capsule berth. Ship is then in the so-called
transfer position.
4.5.2 Capsules and crane

This ship is equipped with a hydraulically operated AHC crane (Active Heave Compensation) fitted
with a dynamic controller that compensates the ship
motions in order to stabilize the cranes extremity. The
crane weights an estimated 2025 t and has a maximum reach of 20 m radius and 15 m height. From the
preliminary seakeeping study (see attachment) it was
found that the most important motions to be controlled
were the heave and pitch. The calculated motions
reveal that it is fairly easy to implement a motion
controller in the crane in order to make the required
hydraulic adjustments to maintain the capsule almost
638
Figure 14. Capsule plan.

Figure 12. General lay-out of capsule.
Figure 13. Capsule profile.
static, relatively to the platform, even with the ship in

motion.
The capsule structure consists on a two level cabin
made of GRP. This GRP cabin is then supported by
bearings located around a steel ring at the middle
body. This steel ring is the structural part that supports the cabins weight when berthed in the vessel
or in the tower. There are eight steel tubes departing from the ring converging to the capsule lower end
(detailed engineering studies required in the development stage). These tubes are encased in special rubber
fenders in order to protect the inside GRP structure
from collisions and improve sliding when docking on
the platform or ship. There are also four steel tubes
departing from the middle steel ring converging to the
capsules top end where the suspension hook device is
located.
The protective steel cage (consisting on the tubes

and metal rings) is separated from the GRP cabin.
This separation is needed to allow the cabins rotation inside the cage. The rotation is carried out by a
small electric engine, battery operated, (or manually
in the event of electric failure) and rotates the cabin as
much as 90 , in order to direct one of the capsule doors
towards the platform. The ring as also a 360 range rail
protection to prevent crew falls from the capsule. The
rail is divided into several parts and all of them are
removable in order to allow a free passage in way of
the capsules entry.
The capsule will be able to carry a team of 5
elements, plus their tool packs and a cargo load of
up to 1 t. The capsule has five comfortable and safe
seats, fitted with safety belts. The capsules lightweight
is approximately 4 t. With full capacity it will have
the weight of 5.5 t maximum. Cargo is to be inserted
and removed from the capsules cabin using a simple pallet jack. Inside the capsule cabin the cargo is
safely secured with special lashing devices, both on
floor and on top. Theres a medical stretcher conveniently stored at the top of the capsule upper deck
cabin. This stretcher might be used to transport injured
crew members from the platform to the capsule. Inside
the capsule the stretcher is firmly fixed to the floor.
The stabilized injured team element is then transported
safely inside the capsule to the ship, through the use
of the crane as emergency removal device.
Under the capsule seats there are first aid kits. On
the lower deck there is a survival kit. The lower deck
of the capsule is accessible by a watertight hatch. The
lower deck is an essential part of the capsule in terms
of capsule stability on water, in the unlikely event
of a fall to the sea. The lower deck provides additional flotation to the capsule and space for tool
storage and electric batteries. If required the bottom
of the lower deck maybe fitted with fixed ballast (to
improve stability). Preliminary stability studies were
carried out.
The deployment time is estimated within the range
of 1525 minutes (depending on the sea conditions and
number of landing attempts). This time includes: the
639
Figure 16. Plan of tower work platform.
Figure 15. Tower work platform.
moment when the capsule, fully loaded, is lifted from

the ships berth, the extension of the cranes arms until
reaching the platform, the lowering of the capsule into
the platform landing ring, and finally the suspension
hook disengaging and returning of crane to initial position. The recovery time is estimated within the range
of 1015 minutes (depending on the sea conditions).
This time includes: the lowering of cranes hook to
the platform, the hook engaging to the capsule (done
by the capsules team), the team fully prepared and
secured inside the capsules cabin, the lifting of the
capsule from the platform and finally the successful
lowering down into the ships capsule berth.
The concept of the capsule is highly versatile, allowing the operators to customize them according to
their specific objective. There could be capsules fitted
with workshops, dormitories, kitchen/mess, offices,
W.C.s, etc. This could be useful to provide important
infrastructures to personnel involved in long term
repairs in one specific tower without having the
support of a mother ship.
4.5.3 Tower work platform
The platform consists on a simple round steel structure supported by eight main beams, with eight pillars
connected to the tower (detailed engineering studies

required in the development stage). There will be a
two ring layout at the far distance part of the round
structure. The capsules will fit between the external
and internal steel rings. These rings will have fenders
in order to prevent damage to the capsule and to the
platform structure itself.
It is estimated that the complete platform will
weight approximately 50 t. The maximum radius of
the platform will be 10.9 m (including the 3 m tower
radius). The platforms have surrounding rails in order
to prevent falls. The rails are removable in way of
capsules exit, in order to facilitate the procedures
of transference of personnel and cargo between the
capsules and platform. Beneath the rings that berths
the capsules theres a permanently fitted safety net
to increase the security of personnel. The platform
rings provide an almost 360 capsule deployment area
around the tower. There are also four gangway landing
zones.
PRELIMINARY ANALYSIS OF SWATH SHIP

PERFORMANCE
5.1 Seakeeping
The availability of the turbines depends in first place on
the number of faults actually occurring, however this
assessment is beyond the scope of this work. Availability also depends on the maintenance vessels being
available for servicing the faulty turbines and this
aspect was covered above, being shown that maintenance operations might be covered with two vessels.
A third factor comes into play because the service vessels may be available, but unable to provide access to
the turbines due to bad weather. It is the aim of this new
concept to extend the access from the current 1.5 m to
3.0 m significant wave height.
In order to demonstrate that the proposed swath ship
and transfer system is able to provide access in 3.0 m
640
Figure 17. Geometric model with higher order panels for

the hydrodynamic calculations.
wave height, a seakeeping study was undertaken in

head seas with the stationary ship positioned besides
a vertical fixed cylinder with 6 m diameter, fixed at
the 45 m depth bottom and crossing the free surface.
The results of this study are summarized here, including statistical results of the motions at the tip of the
crane for a set of realistic seastates with significant
wave height between 1.0 m and 3.0 m and mean wave
period between 5 s and 11 s.
The hydrodynamic problem is solved with 3D linear radiation-diffraction higher order panel method.
The bodies are discretized into a set of higher order
panels with varying potential on each panel. Figure
shows the geometric model for the hydrodynamic
calculations. The regular and the irregular incident are
two-dimensional, meaning that, far away from the ship,
they propagate in one direction only. However, due to
the presence of the vertical column, the ship will be
subjected to a three dimensional wave field.
The solution of the hydrodynamic problem results
on the transfer functions of the absolute motions. The
translational motions at the end of the cranes arm
are calculated by the linear combination of the six
degrees of freedom absolute motions. The assumption
is that the angular motions are of small amplitude.
Figure 18 presents the transfer function amplitudes of
the longitudinal and vertical motions at the end of the
cranes arm. This point as the following coordinates
with respect to midship at the waterline: P (8.5 m;
20.0 m; 32.0 m). The lateral motion is not presented
since the related amplitudes are small.
The longitudinal and vertical motions have two
important dynamic amplifications related to the natural periods of heave and pitch. These motions need to
be controlled in order to obtain a stable platform.
In fact, the reason for choosing a swath hull is
because it is possible to control the vertical motions
with this configuration. The very small waterplane
area results in small vertical wave exciting forces (as
opposed to conventional ships) which can be controlled by active systems (thrusters and/or stabilizing
tanks). It is also possible to introduce passive systems
Figure 18. Transfer function amplitudes of the longitudinal

and vertical motions at the top of the cranes arm.
(hydrodynamic stabilizers) to significantly increase

the damping of the vertical motions and in this way
reduce dynamic amplifications.
These measures were artificially included in the
seakeeping calculations, effectively reducing the
dynamic amplifications to a maximum value of 1.
The motions of the cranes arm in irregular waves
are calculated using linear spectral techniques. The
JONSWAP wave spectrum is applied to represent
stationary seastates as function of the significant wave
height (Hs), peak period (Tp) and the peakness parameter assumed as 3.3. Combining the wave spectrum
with the motions transfer functions one obtain the
response spectra in irregular waves. The Rayleigh
distribution is used to represent the probability density
function of the motion responses maxima. Finally the
calculations were carried out for a range of Hs between
1.0 m and 3.0 m and Tp between 6.5 and 14 s.
Figure 19 presents the significant motions at the end
of the cranes arm as function of the mean wave period
Tz and for Hs = 3.0 m. Significant motion means the
average of the 1/3 largest motion peaks.The significant
longitudinal motion for the worst seastate is 1.65 m
while the vertical motion is 2.05 m. This corresponds
to Hs of 3.0 m and a mean wave period of 6 s. This is
a seastate with a very low probability of occurrence,
because the wave period is low. The motions reduce as
the mean wave period increases. More realistic wave
periods for the same Hs of 3.0 m result on a significant longitudinal motion of 1.36 m and vertical motion
of 1.60 m. This level of motions can then be easily
641
Figure 20. Effective power of the hull, brake power at the

propeller shaft and power required from the motors.
Figure 19. Significant motions at the end of the cranes arm

as function of the mean wave period Tz and for Hs = 3.0 m.
compensated by the active heave compensation system of the crane and small rotations of the crane.
The result is minimal motions at the tip of the crane,
and also of the suspended capsule.
A criterion for safe operation is defined as a 95%
probability that the motions at the end of the crane
arm are smaller than 0.4 m. This objective is achieved
using a ship with excellent seakeeping behavior
together with a crane with oscillatory motion compensation. The objective is that the operation can be
carried out, complying with the criterion, in seastates
with significant wave height of up to 3.0 m.
It is concluded that the transfer operation complies
with the operational criterion if:
(a) The heave and pitch motions of the ship are controlled with a purpose designed active system so
that the dynamic amplification response around
the natural frequencies is reduced;
(b) The crane oscillatory motion compensation system is able to compensate both the vertical and
the longitudinal motions with amplitude of up to
2.5 m and a minimum period of 7 seconds.
Given the former premises, the proposed system
successfully extends the operability of the transfer
operation from the current 1.5 m to 3.0 m of significant wave height.
5.2
Powering
Once the hull shape is defined and the total weight

and weight distribution estimated to obtain the ship
draft and trim, the next step is to calculate the ship
resistance to the advance. The hydrodynamic resistance can be decomposed basically in two components
assumed independent, namely, the viscous resistance
and the wave resistance. The estimate of the viscous resistance assumes that the related effects are
dominated by the hull friction component, where
the friction coefficient is estimated by the ITTC
1957 line (ITTC International Towing Tank Conference). The viscous resistance is then proportional
to the hull speed squared and to the hull wetted area.

An empirical method is used to account for a small
viscous contribution related to the hull shape.
The wave resistance is related to the energy spent to
generate a complex wave system as the hull advances
through the free surface. The method to calculate this
component is based on the potential flow theory and
basically applies a distribution of singularities satisfying the free surface boundary condition and with
unknown strength to be determined by the application of the kinematic boundary condition. Once the
velocity potential is determined, the Bernoulli equation gives the pressures which integrated over the
hull wetted surface result on the hydrodynamic forces,
including the wave resistance force.
The wind resistance on the hull and structure above
the water also needs to be considered but this component is typically much smaller than hydrodynamic
resistance and will not be considered at this stage.
Figure 20 shows the effective power of the hull,
brake power at the propeller shaft and power required
by the engine. Total power required at 12 knots is
1470 kW and at 16 knots is 3637 kW.
5.3 Intact and damage stability

The main objective of intact and damage stability
studies at this stage of the preliminary project is to
demonstrate the technical feasibility of the ship in
terms of floatation and stability. The results obtained
for a full load and ballast condition allow the conclusion that the ship has sufficient stability with the tank
arrangement currently provided. Ballast capacity is
more than sufficient for ballasting the ship down to an
appropriate trim, heel and stability condition. Tables 3
and 4 show summaries of the departure and arrival
conditions. The arrival condition (after extended
period servicing the wind turbine farm) is a ballast
condition, as the ship requires ballast to ensure mainly
appropriate drafts.
Metacentric height (2.25 m and 2.38 m) and righting
arms appear sufficient for compliance with suitable
stability criteria. Ballast capacity is also more than
sufficient for compensating heeling when the crane is
being used at its maximum reach with a suspended
capsule.
642
Table 3. Summary
consumables).
Lightship
Diesel Oil
Fresh water
Provisions
Stores
Capsules
Crew and effects
Freefall
Displacement
KG (solid)
Draft fwd
Draft aft
GM (fluid)
KMt
TPC
of
departure
condition
(100%
1378.00 t
169.70 t
95.10 t
60.00 t
20.00 t
32.00 t
4.00 t
10.00 t
1768.80 t
9.05 m
6.96 m
6.98 m
2.25 m
11.29 m
1.47 t/cm
Figure 21. Hydrostatics.
Table 4. Summary of arrival ballast condition (10%

consumables).
Lightship
Diesel Oil
Fresh water
Ballast water
Provisions
Stores
Capsules
Crew and effects
Freefall
Displacement
KG (solid)
Draft fwd
Draft aft
GM (fluid)
KMt
TPC
1378.00 t
16.97 t
9.55 t
104.13 t
6.00 t
20.00 t
32.00 t
4.00 t
10.00 t
1580.65 t
9.46 m
5.40 m
5.92 m
2.38 m
11.88 m
1.47 t/cm
Figure 22. Components of lightship weight.
Society will have to be ascertained. Damage stability

will certainly pose a difficult problem for this type
of ship.
6
It is also important to note that draught in full

load condition is approximately 7.0 m and in ballast
condition with 10% consumables it is approximately
5.65 m. These draughts mean that a minimum clearance of 3.0 m is maintained between the wetdeck
and the waterline, reducing the probability of slamming, and a minimum clearance of 2.0 m is maintained
between the top of the submerged pontoons and the
waterline, precluding the coming to the surface of the
pontoons. Both these aspects are quite important for
the seakeeping performance of the ship.
Figure 21 shows the ships hydrostatics, being
worth noting the sudden change in hydrostatic characteristics occurring at the transition from the submerged hulls to the struts.
It is also interesting to analyse the decomposition
of the lightship shown in Figure 22. The structure
of the hull (steel) amounts to nearly 2/3 of the total
lightship. Outfit corresponds to 19% of the total
lightship weight, while machinery represents 17%.
It is clear that comprehensive studies will have
to be undertaken and compliance with the criteria
agreed with the Maritime Authority and Classification
QUALITATIVE RISK ASSESSMENT
The main advantage of the proposed system is personnel and equipment being transferred protected inside
a capsule, which is lifted to and from the work platform. The dangers associated with helicopters (wind,
collisions), suspension (failing of ropes), gangways
(failure of connection to tower), stairs (climbing) are
not incurred.
A quantitative risk assessment of the system is to
be carried out in order to assess the most significant
risks. However, below, a brief qualitative review of
some hazards and safety measures aiming at mitigating
the consequences is given.
Failure of suspension cable leading to capsule
falling to water: capsule floats and is stable; design
considers impact in water; personnel and equipment
to be appropriately secured, immersion suits used,
ribs provided to recover personnel.
Capsule suspended collides with tower: capsule
structure designed to sustain collisions with tower
using design principles for lifeboats.
Capsule sustains damage when deploying and
cannot be recovered: other capsule may be deployed in
643
platform at opposite side and used to recover personnel

and equipment.
Crane collides with tower or platform due to ship
motions: crane structure designed to sustain collisions
with tower; the design is to limit damage so that at least
the crane can lower capsule to water and rib recovers
personnel.
Failure of dynamic positioning while in transfer position: ship to be fitted with a certified DP2
system; ship is swath with two engine rooms so
redundancy from flooding or fire hazards is inherent.
Failure of dynamic positioning or crane impede
recovering capsules: other ship in wind farm is called
to recover capsules; in good weather the personnel can
be recovered using ribs or gangway.
Ship collides with tower foundation: foundation
tower and ships hull are to be designed for sustaining collisions; crane to have maximum possible reach
in order to keep ship away from tower; ships tank
arrangement is designed to prevent oil spills following
collision.
Personnel falling from platform: protection nets to
be fitted inside the platforms ring.
Large accelerations sustained by personnel when
in suspended capsule: capsule to be suspended as
near to the tip of crane as possible; aerodynamics to
be studied for low rotation; fin to be fitted to align
capsule with wind direction.
Personnel sustain injury in turbine and medical
evacuation is required: capsule is sufficiently large
to accommodate two persons in stretchers.
Figure 23. Building cost.
Figure 24. Operational and maintenance costs.
8
7
7.1
BUILDING AND OPERATING COSTS

Building costs
The building costs can be split between ship, capsules and platform. The values (in percentage) given in
Figure 23 are estimates. It is anticipated that
the dynamic positioning system and active heave
compensated crane, while using technology currently
in use in offshore applications, will have to be adapted
to this specific type of operation. The total cost of the
ship is expected to be around 21 million euro.
The capsules cost is expected to be, in total, for
six units, around 300000 euros. The cost of the work
platform to be installed in each wind turbine is
estimated to be around 250000 euros.
7.2
Operating and maintenance costs
The operational and maintenance yearly costs given

below consider a typical northern European scenario and are merely estimates. It is considered that
platform maintenance and insurance costs are diluted
in wind turbine costs.
The operational and maintenance costs amount
to 5100000 euros a year. As shown in Figure 24,
paramount among these are the personnel costs, which
make up 50% of the total. Fuel costs are also important,
as well as the maintenance costs of the vessel.
CONCLUSIONS
This paper has described the conceptual design of a

swath ship to be used to provide access to far offshore wind turbines in seastates up to 3 m significant
wave height. The dimensions and characteristics of this
ship, when located at a safety distance from the turbine head on to the incoming waves, certainly allow
the operation in such sea conditions.
A significant advantage of the system is that
personnel and cargo are transferred protected inside
a capsule, which is lifted to and from the work platform. The dangers associated with helicopters (wind,
collisions), suspension (failing of ropes), gangways
(failure of connection to tower), stairs (climbing) are
not incurred. Wind and current will be compensated by
the dynamic positioning system. The ship will be operated next to the platform in head seas and wind is also
expected to come predominantly from that direction.
Overall building, operating and maintenance costs
are relatively low as a 200 turbines wind farm would
require 2 or 3 ships, 12 to 18 capsules and a structure
(ring) in the work platform of each tower. A patent is
pending for the integrated system.
This preliminar design requires still considerable
research and development, as well as design effort,
until finally reaching the maturity necessary for producing a prototype.
Main issues to be researched would include the
numerical evaluation of seakeeping behaviour of
644
different hull designs, with passive systems to minimize vertical motions, the development and assessment of an active control system based on vertical
thrusters
to minimize the ship vertical motions, development
and assessment of an active control system based on
a longitudinal tank partially filled with water to control the pitch motion, assessment of aerodynamics of
different capsule shapes.
Finally, some model testing of ship-capsule-tower
with waves and wind in tank facilities would be needed
before the detailed design of the ship, integration
of crane and dynamic positioning system and detail
design of capsule and platform.
ACKNOWLEDGEMENTS
The authors would like to acknowledge Engo scar
Mota and Engo Ventura de Sousa, on behalf of
AIN, for bringing to our attention the Carbon
Trust Competition and for their encouragement to
this work.
REFERENCES
Bureau Veritas (2002), Rules for the Classification of High
Speed Craft.
Bureau Veritas (2006), Rules for the Classification of Naval
Ships.
Bureau Veritas (2009), Rules for the Classification of Steel

Ships.
Carbon Trust 2010. Offshore Wind Accelerator (OWA)
Access Competition Competition Overview and
Technical Specification.
Dinsmore, R. 2004. Small SWATH Research Vessels, in
Small Research Vessel Compendium, University National
Oceanographic Laboratory System (UNOLS).
Dockter, M.E., Schmidt, K. 1996. SWATH Research Vessel:
The Building of RV Western Flyer, Marine Technology
and SNAME News, July.
Hansa 2010. Swath ship Elbe by A&R, Hansa International
Maritime Journal, Supplement Ships 2009.
IMO-MSC (2006), Resolution MSC.235(82) Adoption
of the Guidelines for the Design and Construction of
Offshore Supply Vessels.
IMO-Assembly (1997), A.863(20) - Code of Safe Practice for
the Carriage of Cargoes and Persons by Offshore Supply
Vessels (OSV Code).
The Crown Estate 2010, A Guide to an Offshore Wind Farm.
IMO-MSC (2008), Resolution MSC.266(88) Code of
Safety for Special Purpose Ships.
Van Slyke, R. J., Morgan, J.M., Leach, T.S., Etchemendy, S.A.
2001, R/V Western Flyer Hull-Strength Upgrade, Marine
Technology.
Vrugt, R., Noordenbos, P.H., Dudson, E. 2002. Rough
Seas and Small Passenger Ferries the Damen 3717
Swath Solution, RINA conference on High Speed Craft,
November 1112.
Wetterling, P., Richerdt, T. 2006. OptiSWATH A new way
towards optimum SWATH Design, HANSA International
Maritime Journal, 143, Nr. 9.
Yoshida, T., Fujita, Y., Fujino, M. 2000. A proposal of the
CS-Swath as a Transocean High Speed Ferry, RINA.
645

Modeling wave energy for the Portuguese coast

D. Silva, P. Martinho & C. Guedes Soares
ABSTRACT: The models presented here aim at evaluating the wave energy in Portugals continental coast.
The system implemented for this effect is based in SWAN (Simulating WAves Nearshore), a coastal prediction
model, using as input wind fields from the atmospheric model MM5 (Mesoscale Meteorological Model) and
wave boundary conditions from WAVEWATCH III model, which is forced by re-analysis data of NCEP/NCAR
on the scale of the North Atlantic. SWANs simulations are validated with data from two buoys, Sines and Leixes,
which belong to the Portuguese Hydrographic Institute. A statistical analysis of the errors in the predictions of
significant wave height and mean wave period is carried out and their time evolution is presented. Since previous
studies had already been carried out for this area, using the WAM model for wave generation in the Atlantic
Ocean and SWAN for wave transformation in the coastal area, a comparison between those results and the ones
obtained in this study is also made.
INTRODUCTION
The prospects of depletion of non-renewable energy

sources and the need to decrease the impact caused to
the environment, led to the present tendency to turn
into renewable energy resources. Oceans have a large
energetic potential in different forms.The wave energy,
which is considered in this work, is due to the wind
action on the ocean surface that generates the waves.
This type of energy is abundant and predictable
allowing devices to extract power from a given location. Wave power depends on the energy density spectrum and the group velocities of waves. This means
that, although the significant wave height is a relevant
parameter, when assessing the wave power in a specific
site, a location having in general higher wave heights
is not necessarily also the richest one in wave power
(Rusu & Guedes Soares 2008, 2009). The waves, since
their generation, can travel for kilometers with small
loss of energy. Their interaction with the ocean bottom will cause the dissipation in higher scale due to
phenomena like reflection, refraction and whitecapping, so in offshore is where more energetic potential
exists.
Because the wind blows mainly from west to east,
the highest ocean energy is concentrated in the west
coasts, in latitudes between 40 60 of north and
south. The annual average power in the wave fronts
varies in these areas between 30 and 70 kW/m, with
peaks up to 100 kW/m in the Atlantic SW of Ireland,
in the southern Ocean and close to the Cape Horn.
Portugal has substantial wave power potential, because
of its location in relatively high latitude and a long
stretch of ocean immediately to the west (Rusu &
Guedes Soares 2008, 2009), making it a target for the

operation of energy devices near to its coast.
Energy converters can be split into three categories
depending on their location: shoreline, nearshore and
offshore (Guides Soares et al. 2012). To plan the installation of these devices it is necessary to study the wave
conditions, which can be determined with numerical
models.
The development of a system that is able to predict
the wave conditions in various coastal locations is very
important. Numerical models can be forced by atmospheric models and produce long term hindcasts, of
about 4050 years, allowing creating a solid statistical
base in both ocean (Pilar et al. 2008) and costal environments (Rusu & Guedes Soares 2008). WAM and
WW III are third generation models for processes of
wave generation and deep water propagation. SWAN
is probably the model most used for coastal areas, to
calculate the directional spectrum evolution, allowing
obtain estimates of wave parameters in coastal areas
and lakes.
In this work the results are validated with Sines and
Leixes buoys from the Hydrographic Institute.
IMPLEMENTATION OF THE WAVE

PREDICTION SYSTEM
This work has the aim to evaluate the wave conditions for Portugals continental coast and to study its
potential wave energy.
The MM5 model (Dudhia et al. 2000) is used
for meteorological forecast, of limited area, not
647
hydrostatic, with a system of sigma coordinates, appropriate to simulate or predict atmospheric circulation of
mesoscale and microscale.
The WW III (Tolman 2009) model has been developed at the Marine Modeling and Analysis Branch
(MMAB) of the Environmental Modeling Center
(EMC) of the National Centers for Environmental Prediction (NCEP). This is a third generation model that
solves the spectral action density balance equation,
using as input wind and ice fields and bathymetry.
The SWAN model allows predictions of the wave
conditions in limited water depth, like the case of
estuaries, lakes and coastal zones, using as input
boundary conditions, bathymetric data, wind and currents fields. It is based on the energy balance equation
that describes the evolution of the wave spectrum
in time, geographical and spectral spaces. In most
third generation wave models, the action density spectrum (N(, )) is considered, instead of the energy
density spectrum (E(, )), since in the presence of
ambient currents action density is conserved whereas
energy density is not. The action density is equal to
the energy density divided by relative frequency ().
SWAN considers effects of spatial propagation, refraction, dissipation, generation, and non-linear wavewave interactions represented in Hasselmann equation
(The SWAN team, 2006):
This equation can be related to spherical coordinates defined by longitude and latitude , and get
the form:
In the left-hand side of equation it is represented

the local rate of change of action density in time (first
term), propagation of action density in geographical
space (second and third term), shifting of the relative frequency due to variations in depths and currents
(fourth term) and refraction due to depth-induced and
current-induced (fifth term) (Booij et al. 1999). At the
right-hand side of the equation there is the sum of
energy sources and sinks (Stot ). They correspond to
the atmospheric input, whiteccaping dissipation and
nonlinear quadruplet interaction, for deep water, and
bottom friction, depth induced wave breaking and nonlinear triad interaction, for shallow water (The SWAN
team, 2006):
The difference between the system now implemented and the one described in Rusu and Guedes
Soares (2009) is that the present one uses WW III
wave model as boundary conditions for SWAN while
the other used WAM. Furthermore in the work of Rusu
Figure 1. Bathymetry of Portugal. Areas of study: 1)

Aguadoura and Porto, 2) Figueira da Foz, 3) S. Pedro de
Moel, 4) Peniche and Nazar, 5) Cascais, 6) Sines and
7) Sagres.
and Guedes Soares (2009), the wind fields used were

the ones of the project HIPOCAS (Guedes Soares,
2008), which were produced from reanalysis atmospheric data (Weisse & Feser, 2003), while the present
one uses wind fields produced by MM5, which in the
meanwhile is running operationally (Guedes Soares
et al. 2011). The use of MM5 in a fine grid is shown
to improve predictions close to the coast (Rusu et al.
2009).
SWAN first test run is made for the entire coast
of Portugal (without nestings). The purpose is to validate results for continental Portugal before simulating
with nested grids. The next step will be to study the
nested runs that are represented in the Figure 1.
The system implemented in this study consists on
coupling the SWAN model, forced with wind fields
produced by the atmospheric model MM5, with WW
III model for wave information, which is forced by
re-analysis data of NCEP/NCAR. The bathymetry is
obtained from GEBCO (General Bathymetric Chart of
the Ocean).
The next scheme illustrates the input data used in
SWAN, and data that forced WW III and MM5.
The WW III model covers almost the entire North

Atlantic basin and SWAN the coast of Portugal as
shown in Figure 2.
648
Figure 2. Implementation areas for WW III and SWAN.
Figure 4. Position of the two Portuguese buoys.
Table 2.
Buoy 1
Buoy 2
Coordinates and depth of the buoys.

Latitude
Longitude
Depth
41.2033 N
37.9211 N
9.0883 W
8.9289 W
83 m
97 m
Figure 3. Domain of model SWAN implementation.

Table 1.
Computational grid.
Latitude
North Atlantic
(WWIII)
North Atlantic
(NCEPs
Reanalysis 2)
Portugal
(SWAN)
Portugal
(MM5)
Longitude
Resolution
15 N/72 N 66 W/7 E
1 1
15 N/72 N 66 W/7 E
1.875 1.905
35 N/45 N 11 W/6 W 0.05 0.1

35 N/45 N 11 W/6 W 0.345 0.345
The area where the WW III boundary conditions and

MM5 wind fields are applied is shown in Figure 3.
The main characteristics of the computational grid
are presented in Table 1. The SWAN simulations are
performed in the non-stationary mode, for 25 directions and 24 frequencies. The formulations used are
Komen for wave generation, which describes the transfer of the wind energy to the waves, and Hasselmann
for whitecapping dissipation.
3 VALIDATIONS IN THE TIME DOMAIN
To evaluate the model performance a comparison is
made between the measurements of the two buoys
Figure 5. Time series of Hs (Buoy 1).
and the simulations results. The buoys are deployed in

Leixes (Buoy 1) and Sines (Buoy 2), as represented
in Figure 4.
Their specific location is displayed in table 2.
Time series with the buoys data and SWAN simulations are shown below, in Figures 5 to 8. The
wave parameters used to compare are significant wave
height (Hs) and mean wave period (Tm). The period
considered is from October 1st to November 20th,
2009. It is not possible to have all points validated
649
in this time interval due to buoy failure and for some

days that MM5 had no results (20th to 25th October).
A statistical evaluation using RMSE (Root Mean
Square Error), Bias, SI (Scatter Index), r (correlation coefficient) analyses is made and their results
presented in Table 3. The formulas to obtain those
statistical values are the following:
Figure 6. Time series of Tm (Buoy 1).
Figure 7. Time series of Hs (Buoy 2).
where Xi corresponds to measurements values, Yi to

simulated values and n the number of points used.
Analyzing Table 3, it can be seen for Hs that:
the Bias in case of buoy 1 is positive showing that
the model underestimates the buoy data, but in case
of buoy 2 is negative which means that the model
overestimates the buoy.
the correlation coefficients for both buoys are
higher than 0.87, i.e. the data correlate well, specially in buoy 1, with 0.91;
both buoys reveals good results for RMSE, with
values lower then 0.45, being better for buoy 1 with
0.43 m.
Figure 8. Time series of Tm (Buoy 2).

Table 3.
Buoy 1
Buoy 2
Statistical results.
Hs (m)
Tm (s)
Hs (m)
Tm (s)
BMED
SMED
BIAS
RMSE
SI
1.988
7.649
1.725
7.749
1.959
5.783
1.882
5.309
0.038
1.865
0.16
2.439
0.425
2.328
0.449
2.829
0.213
0.304
0.260
0.365
0.908
0.603
0.875
0.545
207
207
217
217
650
Figure 9. Hs spatial distribution. In background significant

wave height scalar fields and in foreground wave vectors.
For the case of mean wave period the results are not
as good.
4 ANALYSIS OF THE ENERGETIC POTENTIAL
The energy in SWAN is computed with transport
components (expressed in W/m, i.e. energy transport per unit length of wave front) given by the
relationships:
where x, y are the problem coordinate system (for the

spherical coordinates x axis corresponds to longitude
and y axis to latitude), and cx , cy are the propagation
velocities of wave energy in the geographical space
defined as:
Hence the absolute value of the energy transport

(denoted also as wave power) will be (Rusu & Guedes
Soares 2009):
A case study is made for 2009/11/09 at 00h, when a

wave power peak is registered. For the energetic assessment an evaluation of significant wave heights spatial
distribution is also done and can be seen in Figure 9. Its
wave power result can also be observed in Figure 10.
The incoming waves are from northwest, but when
they reach the coast, most precisely in prominent
zones, like Setubal and Sagres cape, their direction
change, due to refraction. Because of that, these
Figure 10. Computed wave power (W/m). In background

wave power scalar fields and in foreground energy transport
vectors.
regions are sheltered areas, as significant wave height

reduces as the waves reach the coast. The highest values for this parameter are in the western nearshore
(near Peniche) with Hs of around 4 m, while the lowest
are in the south.
For wave power the scenario is similar as for significant wave height, with the highest values concentrated
in west coast and the lowest ones in the south.
The average theoretical wave energy for the western
coast is about 60 kW/m and the highest value is also
near Peniche (80 kW/m).
5 ANALYSES OF THE RESULTS OBTAINED
IN RELATION WITH THE OUTCOME
OF PREVIOUS STUDIES
The aim of this comparison is to understand if implementing the WW III model can get better results or at
least results with the same order of magnitude, as the
ones from systems using WAM.
The previous studies used for this analysis are Rusu
et al. (2008) and Bento et al. (2010), both used WAM
model to provide the boundary conditions.
In case of Rusu et al. (2008) the wind field used is
produced by the regional atmospheric model REMO
coupled with NCEP for areas away from the limits of
REMO (Pilar et al. 2008), from the HIPOCAS (Hindcast of Dynamic Processes of the Ocean and Coastal
Areas of Europe) project. The period is from December 3rd, 1993, until February 28th, 1994, and one of
the targets areas is represented in Figure 11.
The domains of SWAN implementation when using
WW III as boundary conditions are exactly the same.
The same period and the same wind (in SWAN implementation) are considered. The bathymetry for the
nested area in this case is from GEBCO, which have
different resolution that one used by Rusu et al. (2008).
Both bathymetry resolutions are presented in table 4.
651
Figure 11. Domains of model SWAN implementation.

Figure 12. Position of the two buoys.
Table 4.
Bathymetry resolution for A2.
Table 6. Statistical results using WW III as boundary

conditions.
Resolution
Table 5.
Bmed Smed Bias
0.01 0.02
0.0083 0.0083
A2 (Rusu et al. 2008)

A2 (GEBCO)
RMSE SI
B1 Hs 3.318 3.209 0.11 0.707

(m)
B2 Hs 3.322 3.00 0.322 0.793
(m)
0.213 0.835 678

0.239 0.825 676
Statistical results (Rusu et al. 2008).

Bmed Smed Bias
B1 Hs 3.31
(m)
B2 Hs 3.32
(m)
RMSE SI
N
Table 7.
4.09
0.77 0.97
0.29 0.91 678
3.84
0.52 0.75
0.23 0.91 676
Statistical results (Bento et al. 2010).

Bmed Smed Bias
RMSE SI
Buoy 2 Hs (m) 2.032 2.514 0.48 0.522 0.257 0.822 106
For SWAN simulations the same formulations

are used.
The buoy data used to validate the simulations
results in both studies are from the WAVEMOD project
(Paillard et al. 2000). The ones considered to compare
are denoted by B1 (9.24W, 40.26 N) and B2 (9.07W,
40.26 N) (Paillard et al., 2000) being their location in
Figure 12.
The statistical results for both cases are displayed in
tables 5 and 6. The wave parameter is significant wave
height.
Looking at the results it can be seen that for B1 the
statistic parameter RMSE presents better results using
WW III model as boundary conditions, but in case of
B2 this statistic parameter is similar for both cases. For
the correlation coefficient the results are better for the
study of Rusu et al. (2008).
Table 8. Statistical results using WW III as boundary

conditions.
Bmed Smed Bias
RMSE SI
Buoy 2 Hs (m) 2.085 2.362 0.28 0.546 0.262 0.848 105
The next study made by Bento et al. (2010), considered the period from October 30th to November 21st,
2009, and the wind fields were from MM5. Taking into
account that this period is within the time frame used
in this work, the domain of model SWAN implementation (without nestings) and the wind field are the same,
it was not necessary to run the SWAN again. Therefore only a validation is made for the same period. The
652
buoy used for compare is the Buoy 2 (Sines) and the

wave parameter is Hs .
The results for both systems, with WW III and
WAM, are close. If for one hand using WAM give
better results for RMSE and SI, for another r and Bias
are better using WW III.
6
CONCLUSIONS
The present work describes a prediction system implemented in Portuguese nearshore, which uses three
models, MM5 (for wind fields), WAVEWATCH III
(for wave generation) and SWAN (for nearshore simulations), where data from the first two models forced
the third.
The validation results show better results for significant wave height than for mean wave period, with
better correlation coefficient, as expected.
Comparing the results obtained in this work with the
outcome of previous studies, it can be seen that, when
coupling WWII with SWAN, the results are similar to
systems that couple WAM with SWAN.
In terms of wave energy analyses, the west coast of
Portugal has higher potential with an average about of
60 kW/m.
ACKNOWLEDGEMENTS
MAREN Marine Renewable Energy, Energy Extraction and Hydro-environmental Sustainability, which is
partially funded by the Atlantic Area Programme.
REFERENCES
Bento, A. R., Silva, D., Rusu, L. & Guedes Soares, C. 2010.
Evaluation of sea waves offshore the Faro and Setubal harbours (in Portuguese), Proceedings of 1st Hydrographic
Engineering Journeys, Lisbon, 2122 of June 2010;
Bernardino, M., Rusu, L.; Silva, D., Bento, R., Pilar, P. &
Guedes Soares, C. 2009. Performance evaluation of the
forecasting system implemented in the Portuguese harbours (in Portuguese), Proceedings of 6th Portuguese
Journeys of Coastal and Portuary Engineering; Funchal,
89 of October 2009;
Booij, N., Ris, R.C., Holthuijsen, L. H. 1999. A third generation wave model for coastal regions. Part 1: Model
description and validation, J. Geophys. Res. 104, C4,
pp. 76497666;
Dudhia, J., Gill, D., Kuo, Y.R., Bourgeois, A., Wang, W.,
Bruyere, C., Wilson, J. & Kelly, S. 2000. PSU/NCAR
Mesoscale Modeling System. MM5 Modeling System
Version 3. NCAR Tech. Notes;
Guedes Soares, C., Bhattacharjee J.,Tello, M., Pietra, L. 2012.
Review and classification of wave energy converters. Maritime Technology and Engineering, C. Guedes Soares et al.
(Eds.), Taylor & Francis Group, London (this volume).
Guedes Soares, C., Rusu, L., Bernardino, M., Pilar, P. 2011.
An operational wave forecasting system for the Portuguese continental coastal area. Journal of Operational
Oceanography 4 (2), 1626.
Paillard, M., Prevosto, M., Barstow, S. F. & Guedes Soares, C.
2000. Field Measurements of Coastal Waves and Currents
in Portugal and Greece. Coastal Engineering, Vol. 40,
N. 4, pp. 285296.
Pilar, P.; Guedes Soares, C. & Carretero, J.C. 2008. 44-year
wave hindcast for the North East Atlantic European coast.
Coastal Engineering, 55, pp. 861871;
Rusu, E. & Guedes Soares, C. 2009. Numerical modeling to estimate the spatial distribution of the wave
energy in the Portuguese nearshore. Renewable Energy 34,
pp. 15011516.
Rusu, E. & Guedes Soares, C. 2008. Wave energy assessments in the coastal environment of Portugal continental. Proceedings of the 27th International Conference
on Offshore Mechanics and Arctic Engineering (OMAE
2008) Estoril, Portugal. New York, USA: ASME; Paper
OMAE2008-57820.
Rusu, L., Pilar, P. & Guedes Soares, C. 2008. Hindcast of
the wave conditions along the west Iberian coast. Coastal
Engineering, 55, pp. 906919.
Rusu, L., Bernardino, M. & Guedes Soares, C. 2009. Influence of wind modelling on the predictions of waves
generated in an estuary. Journal of Coastal Research,
N. 56, pp. 14191423.
The Swan team 2006. SWAN Technical documentation cycle
III version 40.51.
Tolman, H., (2009), User manual and system documentation
of WAVEWATCH III version 3.14.
WAMDI Group, 1988. The WAM model-a third generation ocean wave prediction model, Journal of Physical
Oceanography, vol. 18, pp. 17751810.
Weisse, R. & F. Feser, 2003: Evaluation of a method to
reduce uncertainty in wind hindcasts performed with
regional atmosphere models. Coastal Engineering, 48(4),
211225.
653

Dynamics and hydrodynamics of a ship like wave energy converter in roll

M. Tello, J. Bhattacharjee & C. Guedes Soares
ABSTRACT: The hydrodynamic and dynamic performance of a wave energy device designed to convert ocean
wave energy by roll motion is analysed. To perform the study, the coastal zone of Leixes, Portugal, is chosen
as the installation site; the main hull characteristics are designed aiming to match the predominant wave period
of the operational zone. Frequency domain transfer functions are obtained, viscous roll damping corrections are
introduced, and the power take off influence is also taken into account. Moreover, analysis based on irregular
seas is made aiming to estimate the responses in a real sea state. The results show that the designed geometrical
shapes guaranties amplitudes in roll higher than 3 for the most common zero crossing period and the observed
significant wave height. Further, a static analysis of the mooring system reveals that less heavy materials for the
mooring line reduces the vertical loads on the fairlead and the anchor point. In addition, subsurface floaters may
be added to the mooring system in order to further reduce the load, especially the vertical load at the fairlead.
1
INTRODUCTION
Ocean wave energy has several advantages over other

forms of marine renewable energy, some of which are:
more constant, predictable, and with higher energy
densities. These enable wave energy devices to extract
more power from smaller volumes which reduce the
rate of energy costs.
Many different techniques to convert wave energy
into other usable form have been proposed and studied
during the last years. Thus, a number of possibilities
are available to date but no one seems to be predominant and no clearly superior engineering solution has
yet been established. On the other hand, the majority of
the designed devices have been built to convert energy
by its oscillating motions at the frequency of waves,
floating on the sea surface or below the sea level.
Since wave energy converters (WECs) have to operate continuously harvesting the energy from ocean
waves, they have to fulfil general requirements such
as: survivability, serviceability and practical installation. Hence, WECs can be designed to survive 1 to 50
year storms or to move to the fail safe thus avoiding
extreme loading, in storms. Moreover, there is a general understanding (Budal and Falnes 1980, Price et al.
2009) that WECs also need to meet certain requirements such as: (a) good power capture over the range
of most commonly incident wave frequencies, considering the phase or complex conjugate control and (b)
ensure harmony between device and power take off
(PTO) to achieve maximum efficiency.
Most of the WECs are oscillating systems and
some are based on the pitch motion as the Pelamis
(Henderson 2006, Retzler 2006) that aligns perpendicular to the wave direction and has articulated segments
that pitch with respect to each other and the PS Frog

MK.5, which is a converter that has a vertical floater
with a ballast tank hanging below (McCabe et al.
2005). Other WECs are based on the heave motion
(Vantorre et al. 2004), such as Aquabuoy (Weinstein
et al. 2004), and Wavebob (Weber et al. 2009), which
are floating buoys that have parts moving vertically.
These WECs convert energy while simultaneously
generating waves, thus their hydrodynamic problem is
a combination of diffraction and radiation problems
as mentioned by Budal and Falnes (1980). Potential
flow codes can be applied to solve the hydrodynamic
problem for WECs.
On the other hand, non-oscillating devices as the
Wave Dragon presented by Kofoed et al. (2006) and the
Seawave Slot-Cone Generator (SSG), (Margheritini
et al. 2009) convert ocean energy by capturing the
water volume of overtopped waves in a basin and
creating a hydraulic head.
For a better understanding of the devices, they have
been classified in different groups according to distance from shore and relative to the technological type.
According to distance from shoreline, this is grouped
as: shoreline, near shore and offshore WECs, while
according to technological type WECS can be divided
in six principal groups: attenuators, overtopping, oscillating water column (OWC), oscillating wave surge
converters (OWSC), point absorbers and pressure differential devices. Although a few of them are presented
in Table 1, a discussion of the different types is omitted
here, but an extended discussion of the devices can be
found in Thorpe (1999), Falnes (2007), Falco (2010)
and Guedes Soares et al. (2012).
Looking into the early designs to convert ocean
wave energy, it can be observed that the majority
655
Table 1. Classification of WECs according to distance

from shore and technological type.
Classification
Sub Classification
Examples
Distance from
shoreline
Shoreline
Near shore
Offshore
Attenuators
Overtopping
Oscillating water
column
Oscillating wave
surge converter
Point absorber
Pressure differential
Pico plant
Waveroller, Oyster
Pelamis, MacCabe
Pelamis, MacCabe
Wave dragon
Azores pico plant
Technological
type
Others
Waveroller, Oyster
Powerbuoy, PsFrog
CETO
Anaconda, Tetron
of them are concentrated on small, modular devices,

yielding high power output when deployed in arrays.
In comparison to the previous multi-megawatt designs,
these small size devices were rated at a few tens of
kilowatts each. More recent designs for offshore
devices have also concentrated on small designs,
all classified as point absorber due to their small
dimension in comparison to the wave length.
Typically, the point absorbers have been designed as
axisymmetric floaters to have the same response independently of the wave direction. However, Buchner
(2011) had the interesting idea of using the experience
and technology from ship dynamics to devise a shiplike WEC that was capturing energy from waves while
performing a pitch motion.
The present paper has been motivated by attempts to
develop a PTO of a spherical shape where the mechanism of energy conversion depends on the relative
motion of an external sphere attached to the floater
and an internal one kept relatively stationary based on
gyroscopic principles.
Therefore, the objective was to provide large roll
motions to such a sphere and this led to the idea of a
ship like floater in roll as these amplitudes are larger
than the pitch motion considered by Buchner (2011).
In the present work, the design of a hull form and its
parameters that govern roll motion are studied. Moreover, the effect of the PTO will be accounted by a
constant damping, as in the work of Buchner (2011).
The hydrodynamic properties will be studied and the
final designed hull forms will be optimized to guaranty
larger roll amplitudes. For comparison purposes, the
performance of the ship-like WEC is also compared
with the one of a sphere. The effect of the mooring lines
is also studied by means of static analysis and considering only variation of the mooring configurations and
components as the material properties, adding buoys
and clamp weights.
2 THEORIES OF FLOATER DYNAMICS

The response of a vessel at sea is a complicated phenomenon involving the interactions between the vessel
Figure 1. The coordinate system and definition of the

heading angle used by WAMIT (See WAMIT User Manual).
dynamics and several distinct hydrodynamic forces.

All responses are nonlinear to some extent but in many
cases when nonlinearities are small a linear theory will
yield good predictions. In the case of WEC, ignoring interactions between mooring systems and power
cables, the floating device can be assumed as a freefloating body. Thus, rigid body oscillatory motions
are solved in the frequency domain to obtain the six
degrees of freedom by solving:
These excitation forces Fk and ship motions j can be

conveniently represented on a right handed Cartesian
coordinate system, X = (x, y, z), fixed with respect to
the mean position of the vessel and origin in the plane
of the undisturbed free surface. As shown in Figure 1,
the translatory displacements in the x, y and z directions are respectively surge 1 , sway 2 , and heave 3 ,
while the rotational displacements about the same axis
are respectively roll 4 , pitch 5 , and yaw 6 , where
subscripts k, j indicate force in the k-direction due to
motion in the j-mode, Mkj are the components of the
mass matrix for the ship, Akj and Bkj are the added
mass and damping coefficients, Ckj are the hydrostatic restoring coefficients and Fk are the complex
amplitudes of the exciting forces.
This investigation uses the WAMIT software to calculate the potential flow hydrodynamic coefficients
and harmonic wave exciting forces.
WAMIT is based on the linear and second-order
potential theory for analysing floating or submerged
bodies, in the presence of ocean waves. The panel
method is used to solve for the velocity potential and
fluid pressure on the submerged surfaces of the bodies.
WAMIT evaluates separate solutions for the diffraction
and the radiation problems for each of the prescribed
modes of motion. Then the relevant hydrodynamic
parameters including added-mass and damping coefficients, exciting forces, RAOs, the pressure and fluid
velocity, and the mean drift forces and moments are
656
evaluated. More details can be found in the WAMIT

user manual, while the theoretical formulation for the
responses at different locations of the body can be
found in Tello et al. (2011).
The need of a small metacentric height (GMt), in

order to have less restoring moments, contributing to the existence of roll angles at periods even
different than the resonance period.
In addition, it is also important to consider the heave
performance in order to avoid higher acceleration in
the vertical direction.
3 A WEC BASED ON ENERGY CAPTURE BY

ROLL MOTION
3.1 Influence of the hull form in roll motion

Many WECs have been proposed and a remarkable
example is the work of Buchner (2011), where a device
harnessing wave energy by pitch motion is presented.
This has motivated the study of a device harnessing ocean energy by roll motion since it experiences
bigger amplitudes than the pitch motion and this is
the requirements of the PTO considered here. Thus,
the present study focuses on the hydrodynamics and
dynamics for a floating WEC exploiting roll motion.
To this purpose the WEC is assumed to operate
in the coastal zone of the Leixes, Portugal, and
following are the requirements to maximize energy
capture:
The resonance period should meet the most common

period in the Portuguese coastal area of Leixes,
which is in the range of 59 [s] (74.33% of the registered waves) (Rusu and Guedes Soares, 2009, Silva,
et al. 2012). For the purpose of the study a period
of T 7.5 [s] has been chosen in order to aid to a
better understanding of the main hull parameters.
The centre of gravity (C.G). should be placed at the
centre of the water plane determined by the waterline (hereafter water plane), so as to give larger space
for the power take off (PTO) equipment and better
safety of installation on board. Moreover, this will
simplify the moments applied to the PTO since it is
a rotational converter and not a translational one.
Figure 2. Panel discretised geometries for the analysis, from

left to right: triangular, elliptical and rectangular shape.
Table 2.
Shape
Roll motion is related to hull parameters by: (a) the

water plane area, (b) hull form (c) the vertical centre of buoyancy and (d) the vertical centre of gravity.
However, the analysis is carried out only for items a,
b and c, while the centre of gravity is held at the water
plane area.
To understand the behaviour of the hull form in
roll motion, three different geometries are analysed:
triangular, elliptical and a rectangular geometry (See
Fig. 2).
The analysis is carried out with WAMIT and the
respective response amplitude operators (RAOs) are
obtained considering a viscous roll damping coefficient around 0.14, obtained by the formulation of
Journee et al. (2003). Their main characteristics and
the hydrodynamic coefficients obtained are presented
in Table 2.
From Table 2 it can be observed that the elliptical
shape has smaller percentage of the added moment of
inertia, followed by the triangular and the rectangular
shape. The advantage of this smaller added moment
of inertia can be seen when comparison between the
elliptical and the triangular geometry is made. In this
case, the smaller amount of added mass experienced by
the elliptical shape explains why the elliptical design
has smaller resonance period even when both have
approximately the same GMt. When comparing to
the rectangular design the advantage of the elliptical
geometry is more evident, since the rectangular shape
needs smaller GMt values and larger displacements to
match the same resonance period and this can only be
achievable by moving the centre of gravity or increasing the displacement which is impractical (Cb = 0.96).
Therefore, the future discussion is focused on the
triangular and the elliptical shapes.
In Fig. 3, roll and heave motions are plotted for the
elliptical and the triangular design. It can be observed
that the resonance roll period for the triangular shape
is 7.75 [s] with amplitude of 5.7 [deg/m] and the elliptical shape experiences a resonance period of 7.5 [s]
with roll amplitude of 12.5 [deg/m]. It is also seen
Main hull parameters and hydrodynamic coefficients for the triangular, elliptical and rectangular geometries.
L B T
[m] [m] [m] [tons] L/B B/T Cb Ixx
Triangular 12 6
Ellipse
12 6
Rectangular 12 6
2
2
2
43.49 2
92.29 2
137.95 2
3
3
3
Vcb
T
Vcg kxx kyy kzz BMl BMt GMt roll [s] B44
0.3 54.1 0.81 0

0.6 127.2 0.93 0
1.0 216 0.98 0
657
A44
2.4 4.8 4.8 4.96 1.24 0.44 5.7 7.8

4.3 12%
2.4 4.8 4.8 4.58 1.38 0.45 12.5 7.5 197.3 7%
2.4 4.8 4.8 6.26 1.57 0.59 19.7 7.0 246.6 20%
Figure 3. Heave and roll motion for the triangular and the elliptical geometries in following, beam and bow quartering waves.
Figure 4. Exciting moment on the vessel hull.
that the roll amplitudes for the elliptical shape is bigger than for the triangular geometry for a wave period
ranging from 4 [s] to 11 [s] for quartering bow and
bean seas.
This variation can be understood from the exciting
moment applied to the vessel. Figure 4 shows how the
exciting roll moment applied to the vessel vary with
respect to the cross sections of the floating structure.
In the case of rectangular cross section, the moments
are due to areas (1 + 2 + 3), for the round cross section
areas (2 + 3) and for the V cross section the exciting
moment is only due to area (3).
Then to maximize energy production and match the
wave period of the Leixes area, the floating devices
should match the following requirements:
A relatively wider cross sectional area closer to the

waterline in order to increase the exciting moment.
A smaller second moment of area and therefore
a smaller hull displacement in order to keep the
same resonance period with smaller GMt values,
which will guaranty bigger roll motions for wave
frequencies different than the resonance.
The elliptical shape seems to be the best choice since
the roll motion is bigger compared to the triangular shape. The small percentage of added moment
of inertia permits to match the resonance period at
smaller GM values, and the bigger cross sectional
area, closer to the water plane, increases the exciting moment. Its slenderness is also important since
it decreases heave motion avoiding vertical movements and therefore improving the operability of the
device (see Fig. 3).
Figure 5. Roll motion in bow quartering and beam waves,

with the PTO, considering Iz r = 2.0 k and 30 k [Nms/rad].
3.2 Hull form analysis including the PTO system

Considering what has been stated before, the next step
is to modify the hull shape enhancing its performance
for roll motion. In this section, modelling of the PTO
and its influences in roll motion is considered for a
floating structure carrying multiple and a single PTO.
To accomplish this work, the physical model of an
apparatus for gyroscope propulsion is considered (see
US patent No US 6,705,174 B2), Thus according to its
physical model, the PTO can be addressed similarly
to the mechanical principle of the well known gyroscopic stabilizer. According to Bhattacharya (1978)
this type of stabilizers are very effective and can reduce
around 60% to 80% of the roll motion but were not
used for reduction of roll motion due to large spare
requirements.
The moment Mz comes from the conservation of
angular momentum and is given by:
where Iz is the mass moment of inertia of the spinning element about the axis of spin, r is the angular
velocity of spin, and 4 the velocity of rolling motion.
Figure 5 plots the RAO for roll motion for a given
hull geometry with and without the influence of the
PTO. It can be observed that a significant reduction of
the roll amplitude occurs when applying the PTO.
658
Table 3.
Shape
Main hull parameters and hydrodynamic coefficients of OPTMAL01 geometry.

L B T

[m] [m] [m] [tons] L/B B/T Cb
OPTMAL01 12
4.5 1.8 52.2
roll
T
Vcb Vcg kxx kzz BMt GMt 90 135 [s] B43
Ixx
2.7 2.5 0.54 0.7 0
1.8 4.8 4.8 1.05 0.33 16
A43
11.6 6.3 189.3 3.10%
Figure 6. Heave and roll motion for OPTMAL01 geometry in following, beam and bow quartering waves, respectively,
and a profile view of elliptical geometry.
3.2.1 Hull form analysis for three spheres

The hull form designed here is denoted as OPTMAL01. This is designed with a water plane area as
a result of modification of the elliptical shape (less
breadth at the fore and aft) with the length of 12 [m],
beam 4.5 [m] and draught 1.8 [m]. The main hull
parameters and the hydrodynamic coefficients are presented in Table 3, while a perspective view is shown
in Fig. 6.
Moreover, a rounded shape (with nearly elliptical
water plane areas) has been chosen for the design,
because the contribution of viscous damping and other
damping sources to the total roll damping are less than
other shapes, consequently leading to increased roll
amplitudes. Other advantage of the rounded shape is
the smaller values of block coefficient adding to match
the required resonance period at smaller GMt.
OPTMAl01 has been analysed with WAMIT considering an external damping of 30 k [N-m-s] for the
three spherical PTOs. Viscous damping factor of 0.07
has been taken from the experimental work of Pessoa
et al. (2010). RAOs are presented in Fig. 5 for heave,
pitch and roll motion.
Table 3 presents the hydrodynamic coefficients for
the proposed design. It is seen that it experiences
smaller amount of added moment of inertia, and this

is even less than the one obtained for the elliptical
geometry (see Tables 3 and 2). The reason of this
hydrodynamic behaviour is mainly due to the rounded
shape of OPTMAL01.
From Fig. 6, it can be seen that the RAO for heave
motions is less than 1 for all wave periods. Hence,
the device will not experience high vertical motion.
This is convenient since it has to be moored, therefore,
simplifying the station keeping requirements.
Figure 6 also presents the plots corresponding to the
RAO for the pitch motion. This shows that the device
experiences bigger pitch amplitudes which could limit
its operation, but this does not represent a big problem
since the resonance for pitch is around 4 [s] which is
normally associated to small wave amplitudes.
In the case of roll the proposed geometry presents
a broader shape for bow quartering and beam waves.
This can be explained due to its relative smaller GMt
achieved for this geometry which a result of the smaller
Ixx and the smaller vessel displacement, consequently
decreasing the GMt. However, any change of GMt
changes the resonance period, but it does not happen for OPTMAL01 due to the rounded cross sections
which have small added moment of inertial permitting
659
Table 4.
Shape
Main hull and hydrodynamic parameters UFO02 and the sphere.

roll
L
B T
[m] [m] [m] [tons] L/B B/T Cb Ixx Vcb
T
Vcg kxx kyy kzz BMt GMt 90 135 [s] B43 A43
OPTMAL01 6.00 3.6 1.8 16.2 1.67 2.00 0.42 13.7 0.63 0
OPTMAL01 3.98 3.98 1.8 14.2 1.00 2.21 0.50 12.3 0.66 0
1.4 2.4 2.4 0.85 0.22 16.8 11.7 6.3 6.1 3.1%
1.6 1.6 1.6 0.87 0.2 14.9 10.5 7.0 184.1 1.0%
Figure 7. Heave and roll motion for UFO02 and the sphere geometries in following, beam and bow quartering waves,
respectively, and a profile view of geometries.
to hold the resonance period at even smaller values of

GMt (see Table 3).
3.2.2 Hull form analysis for one sphere
In this section the configuration and the hull shape
for a floating device designed to carry only one PTO
are analysed. Table 4 and Figure 7 present the main
hull parameters and the hull form of the geometries
denoted as UFO02 and sphere.
To analyse the design of UFO02 and Sphere, their
hull form have been analyzed with WAMIT considering a damping of 10k [N-m-s] due to the spherical
PTO system and a viscous damping factor of 0.009 has
been chosen. The viscous factor has been taken from
the experimental work of Pessoa et al. (2010).
From Figure 7, it can be seen that for UFO02, the
RAOs for heave motions is less than 1 for all wave
periods which is useful since the device is supposed to
be moored, thus avoiding high loading on the mooring system. In the case of the sphere, it experiences
amplification at its resonance period which can result
in bigger loading at wave frequency for the mooring
system.
Figure 7 also plots the RAO for pitch motion for
both designs. Amplitudes developed by the UFO02
design are bigger than the spherical form and also their
respective resonance period are different. However,
pitch amplitudes presented by UFO02 are subjected
to smaller wave periods and consequently smaller
wave amplitudes than that experienced by the spherical
design.
Roll motions are plotted in Fig. 7 and a broader
shape is presented for both designs which are approximately the same for bow quartering and beam waves.
The similarity of the roll motion can be understood
when looking to Table 4 where the resonance period,
the added mass and the potential damping are found
similar for both. Finally, from the analysis of these
designs to carry one PTO, it seems that the UFO02
shape is better than the spherical shape.
3.2.3 Motion responses in irregular seas
The present analysis estimates the responses of the
floating device in a given irregular sea. This analysis is
made considering the same wave spectrum approaching the device for incident wave angles from 180 to 0 .
However, the discussion is only restricted to incident
angles from 120 to 30 , which represent the area of
operability of the device, since it is designed to operate
near shore. Vertical acceleration and other longitudinal
660
Table 5.
Sea states most probable of occurrence in the area 16, of Global Wave Statistics (Hogben et al. 1986).
Sea state 1
Sea state 2
Sea state 3
Sea state 4
H [m]
Tz [s]
H [m]
Tz [s]
H [m]
Tz [s]
H [m]
Tz [s]
1
2
3
4
5
6
7
8
9
10
0.1
0.1
0.2
0.2
0.3
0.3
0.4
0.4
0.5
0.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
7.5
1.6
1.7
1.8
1.9
2.0
2.1
2.2
2.3
2.4
2.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
2.6
2.7
2.8
2.9
3.0
3.1
3.2
3.3
3.4
3.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
8.5
Figure 8. Motion in roll and lateral accelerations calculated at the bow for the geometries OPTMAL01, UFO02 and sphere.
The first three polar plots are for roll motion while the remaining three are for lateral acceleration.
responses are not presented due to small responses

experienced in quartering and beam seas.
Table 5 shows the chosen scatter diagram of wave
period and height adopted for the analysis. The chosen values represent the most probable occurrences for
area 16, which were retrieved from Global Wave Statistics (Hogben et al. 1986). Hereafter denominations of
sea state refer to table 5.
Figure 8 presents the results of the roll angle and
the lateral acceleration for the OPTMAL01, UFO02
and the sphere geometries. In the polar plots the radio
represents the sea state (according to Table 5), while
angles describe the angle of incidence of the waves (see
Fig. 1). As an example, the response in sea state 1 is
plotted in the polar diagram by the radio 0 to 1, which is
related to significant wave heights Hs = 0.1 0.5 and
zero crossing wave period of Ts = 7.5 respectively.
The result shows that OPTMAL01 and UFO02
experience approximately the same responses while
the sphere presents a slight improvement. It can be
seen as well that the devices operate between 3 and
5 for sea states 3 [Hs = 1.6 2.5; Ts = 7.5], between
8 and 12 for sea state 4 [Hs = 2.6 3.5; Ts = 8.5],
which are the most probable occurrence for Area 16.
The importance of this analysis relies on how much the
device will roll while operating in irregular seas, since
the device is targeted to generate energy by the roll
motion. Thus, from the analysis made for the three

geometries presented, it is highly probable that the
proposed hull forms including the effect of the PTO
will roll at amplitudes higher than 3 at the chosen
operational area.
Figure 8 also plots the lateral acceleration felt
onboard (at the forward part of the hull device). The
devices experience different responses. The spherical
shape presents higher values for this criterion, followed by OPTMAL01 and then by UFO02. However,
the acceleration for all shapes does not represent any
hazard for the devices, since the value of the lateral
acceleration is relatively small, half of the acceleration
gravity (0.5g).
The small variation in the GMt and therefore the
similarity in the resonance period, permits for all
devices to experiences similar responses in irregular
seas. Therefore, no significant differences can be seen
in the performance of these hull shapes. However, the
lateral acceleration can be used as a selecting criterion
when looking for the best configuration and hull form.
3.3
Mooring line analysis
Mooring lines are subject to environmental forces from

currents and ocean waves. Current forces might be
divided into mean and fluctuation forces, and vortex
661
Figure 9. Coordinate systems and forces action on the cable (Wang et al. (2010)).
induced vibrations, while wave forces can be divided

into the 1st order forces at wave frequencies (WF); and
the 2nd order forces and higher order forces. The 2nd
order forces comprise mean wave drift forces, forces at
sum frequencies (HF) and forces at difference frequencies (LF), while higher order forces comprises wetted
surface effects, ringing and viscous (non-potential)
effects.
Those environmental forces are considered according to the analysis proposed. Generally, the analysis of
mooring cables is divided in two stages: during pretension and in service. When the cable is addressed during
pretension a static analysis can be performed neglecting the environmental factors (such as current and
wind when the mooring cable is considered a heavy
cable). When analysis is made for in service condition
different approaches can be used to solve the dynamic
and hydrodynamic problem (Johanning et al. 2006).
Mooring lines can be made of chain, wire rope,
chain/wire rope combination or synthetic mooring
lines. Other components can be used as a clump
weight, spring buoy. Clump weights improve the performance and reduce costs. They provide concentrated
weight at the point close to the seabed thus replacing a
portion of chain and increasing the restoring force of
the mooring leg. While spring buoys can be applied to
reduce weight of mooring lines that must be supported
by the vessel hull, reduce effects of line dynamics in
deep water and reduce vessel offset for a given line
size.
Anchor systems are available in a variety, some
of the most common are: drag embedment anchors,
pile anchors (driven, jetted, drilled and grouted), caisson foundations (suction anchors), gravity anchors and
propeller embedment anchors.
The forces acting on the cable segments (see Fig. 9)
in water are: the tension: T
and (T + dT )(
+ d );

tangential drag force;
effective weight; GW dl k;
= Hdl and the binorF = Fdl ;
normal drag force; H

mal drag force; Q = Qdl ,
where T is the tension along
the cable, GW is the effective weight per unit length
in water, F, H , and Q are the tangential, normal and
binormal drag forces caused by ocean currents per
unit length, respectively. The equilibrium of a cable

suspended in water results in the following differential
equations:
The tangential, normal and binormal drag forces per

unit length in water can be acquired through Morisons
equation (Morison et al. 1950). Moreover the constitutive relation of the mooring cable is assumed
to be continuous and extensible with a linear elastic
stressstrain relationship given by Hookes Law:
where E is the Youngs Modulus, A the cross-sectional

area of the unstrained cable, is the unit elongation
and l0 is the unstrained cable length. Under axial tension the unit effective weights of cable in water, GW
and the cross-sectional area, A must be modified as
GW = GW 0 /A0 /(1 + ) A = A0 /(1 + ).
3.3.1 Numerical model
Herein the mooring line is analyzed during pretension to determinate its influence and to select the
best suitable one for the WEC, from a static point
of view. Discussion of designs is presented by means
of: one or two systems, different mooring radius and
cables properties, and the use of buoys for water depth
of 50 [m].
When pretension analysis is considered, a static
approach is chosen to study the mooring problem. In
this process, the mooring cable is discretized into
segments with the same dy = D/, where D is the
anchoring radius in the horizontal direction and is
the segment division number of the mooring cable.
662
The end point of each segment is numbered by the

index i, which runs from 1 at beginning of the system
till + 1 at the end of the system. The coordinate of
the anchor point (x1 , z1 ) is assumed to be (0, 0). Here,
the following parameters should be known prior to
calculation: design pretension load T+1 ; water depth
H ; anchor point; anchoring radius D; cable properties, includingYoungs modulus E, cross-sectional area
A0 , unit effective weight of cable in water GW 0 ; and
buoyancy force and position, if buoy is applied. It is
clear that the coordinate of the top point (x+1 , z+1 ) is
equal to (D, H ).This analysis considers heavy mooring
cables therefore drag forces are neglect. The iterative
equations are:
The cable is discretized with the same dy = D/, so

the length of strained segment is:
Figure 10. Two dimensional profile view of mooring.
Table 6.
Other option to obtain the unstrained segment is from

Equation 10:
The geometrical equations of the strained cable are:
With the given coordinate (x+1 , z+1 ), the iterative

procedure can be initiated with an assumed elevation angle tan1 (H /D) at the top point +1 . Fig. 9
shows that in each iterative step, the calculated coordinate of the anchor point (x1 , z1 ) should be compared
with the real anchor point (0, 0) to see if the error
E1 = (x12 + z12 )2 is greater than 1 , where 1 is a specified small quantity. If E > 1 , +1 needs to be
actualized and the running process starts again until
E 1 . When two cables are used, the analysis is run
till the difference of the tension at the joint is less than
10%, thus saving larger computational time.
3.3.2 Analysis of mooring line configurations
The analysis is divided in eight combinations (see
Table 6). These combinations are chosen in order to
study the influence of the mooring parameters at the
fairlead position. The groups are denoted as: group 1
(mooring 1 and mooring 2) to compare the effect of
a single cable or if a buoy should be employed, group
2 (mooring 3 and mooring 4) to study the influence
of the mooring radius, group 3 (mooring 5 and mooring 6) to analyse the effect of including two different
material properties, and finally, group 4 (mooring 7
and mooring 8) to analyse the variation of the tension
when system of chain-steel and steel-steel are used.
Moreover the study considers two different types
of material, a chain type with GW 0 = 2.86 [kN/m],
A0 E = 12.997 106 [kN] and BS = 1.5 104 [kN];
and steel wire type with: GW 0 = 0.814 [kN/m]
A0 E = 1.150 103 [kN] and BS = 1.6 104 [kN];
where GW 0 is the weight per unit length in water, A0 E
the stiffness and BS the breaking strength.
Figure 10 shows the two dimensional profiles of the
mooring design with two cables.
Mooring lines characteristics for each design.

Buoy
System 1
System 1
Item
N
sys
H
[m]
D
[m]
x
[m]
z
[m]
Buoy
[kN]
Mat
D1
[m]
H1
[m]
Mat.
D2
[m]
H2
[m]
M. 1
M. 2
M. 3
M. 4
M. 5
M. 6
M. 7
M. 8
2
1
2
2
2
2
2
2
50
50
50
50
50
50
50
50
50
50
40
30
50
50
30
30
40
30
20
30
30
20
20
45
45
45
40
40
45
45
175
175
190
163
80
120
105
Chain
Chain
Chain
Chain
Chain
Chain
Chain
Steel w.
10
50
10
10
20
20
10
10
5
50
5
5
10
10
5
5
Chain
Chain
Chain
Chain
Steel w.
Steel w.
Steel w.
40
30
30
30
30
20
20
45
45
45
40
40
45
45
663
Table 7.
Mooring line tensions and characteristics.
Item
Group 1
Mooring 2
Tension 1
Tension1
Tension 3
Tension 4
Mooring 1
System 1
System 2
Group 2
Mooring 3
System 1
System 2
Mooring 4
System 1
System 2
Group 3
Mooring 5
System 1
System 2
Mooring 6
System 1
System 2
Group 4
Mooring 7
System 1
System 2
Mooring 8
System 1
System 2
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
n+1
1
T
[kN]
Th
[kN]
Tv
[kN]
Alpha
[deg]
Lo
[m]
L
[m]
Gw
[kN]
250.0
97.4
300.0
147.5
350.0
197.4
400.0
247.4
60.0
45.1
185.3
46.0
60.0
45.1
185.3
45.6
60.0
45.9
199.9
59.6
75.0
45.2
170.4
46.5
75.0
45.2
93.0
57.6
60.0
44.9
132.6
93.1
52.0
47.7
129.8
90.3
95.6
97.4
140.8
142.0
180.7
181.7
219.0
219.8
44.2
44.8
44.0
46.0
44.2
44.8
44.1
45.6
44.2
44.8
44.1
45.6
43.7
45.2
43.6
46.5
43.7
45.2
43.5
43.7
44.2
44.7
45.2
45.4
44.5
44.5
44.1
44.2
231.0
1.0
264.9
39.6
299.8
77.2
334.7
113.4
40.6
4.8
180.0
0.9
40.6
4.8
180.0
0.6
40.6
4.8
195.0
38.3
61.0
1.3
164.7
1.1
61.0
1.3
82.2
37.4
40.6
4.9
124.6
81.3
26.9
16.9
122.1
78.7
67.5
0.7
62.
15.6
58.9
23.0
56.8
27.3
42.6
6.1
76.3
1.2
42.6
6.1
76.3
0.8
42.6
6.1
77.3
40.0
54.4
1.6
75.2
1.4
54.4
1.6
62.1
40.6
42.6
6.2
70.0
60.8
31.2
20.8
70.6
60.7
74.26
74.27
212.4
71.91
71.92
205.6
71.17
71.19
203.6
70.78
70.80
202.4
11.33
11.33
32.4
67.45
67.45
192.9
11.33
11.33
32.4
57.55
57.55
164.9
11.33
11.33
32.4
49.97
49.97
142.9
23.53
23.53
67.3
53.72
53.72
153.6
23.53
23.53
67.3
50.23
50.24
143.7
11.33
11.33
32.4
49.20
49.21
140.7
11.15
11.15
31.9
49.24
49.25
140.8
To analyse the mooring effects at the fairlead position, the following consideration for the design of
the mooring system is made: when only one cable is
designed to restrain the WEC then system 1 is defined
between the fairlead and the anchor point, while in
the case of the design with two cables, system 1 lies
between the fairlead and the joint of the two cables (See
Fig. 10) while system 2 is placed between the joint and
the anchor point. In addition, when a buoy is employed
this will be placed at the joint of the two cables.
Design requirements for moorings 1, 3-8 have been
considered. These requirements are: the same profile
for system 1 (only valid in each group of analysis) and
designed to transmit the same horizontal tension in all
configurations conformed by two systems. Finally the
smaller vertical tension is considered as a selecting
criterion.
The study for groups 2, 3, 4 and mooring 2 in
group 1 should be designed with anchor which can
support vertical tension at the anchor, while only mooring 1 presents the smaller vertical tension thus drag
anchor could be suitable for this configuration. When
comparing mooring 1 and mooring 3 it shows that for

40 [m] of mooring radius, mooring 3 does not present
any bigger increment on the vertical tension applied
in the anchor, however the opposite holds true when
a reduction to 30 [m] is made for mooring 4 (chain
material for both systems). This is another important
design requirement for WECs due to the fact that they
are designed to operate in farms simultaneously.
In general, the results shown in Table 6 and 7 for
mooring 3 and mooring 4 as well as for mooring 6 and
mooring 7 (both with the same material properties, see
Table 6); the buoyancy requirements increases as the
mooring ratio decreases (chain material for both systems). On the other hand, when steel wire is used, a
considerable reduction of the buoyancy is found for
mooring 6 in group 3, and this can be confirmed comparing mooring 7 and mooring 8 in group 4. The results
for mooring 1 and mooring 5 show that when the
buoy is located at deeper positions (for mooring 5)
less buoyancy is needed.
Finally from the analysis made in group 1, the need
of including a buoy is pointed out since considerable
664
reduction of the vertical load is observed. The importance of considering materials with less weight is
evident as it decreases mooring loads on the WECs.
Less heavy materials are also needed due to the reduction of the net buoyancy required (see Table 6). On
the other hand, there might be an extra loading on the
mooring due to the buoy near the sea surface. Thus the
depth of the buoy position should be studied in detail.
CONCLUSIONS
From the analysis made for a WEC exploiting roll

motion, the dependence of the resonance period and
the amplitudes of roll motions on the metacentric
height, hull form and the structural inertia of the floating structure are observed. Taking into account those
parameters, three different hull forms are studied.
The analysis for the designed hulls shows no significant variation in their free floating body performance.
Further, when they are analyzed taking into account
the effects of PTO, modelled by a constant damping,
this results in a small variation of the resonance period
but not significant, while the motion amplitude seems
to decrease considerably.
Furthermore, analysis in irregular seas is performed, obtaining roll motion higher than 3 for the
most commonly observed significant wave heights
and zero crossing wave periods. Additionally, smaller
accelerations are observed which is important for the
device performance.
Finally, a set of proposed mooring designs are studied. The analysis reveals the relevance of using less
heavy materials for the mooring cable and the importance of buoyancy to reduce vertical loads at the
fairlead. Moreover, significant reduction of the vertical load at the fairlead was found when varying the
mooring radio.
ACKNOWLEDGEMENTS
Hydrodynamic analysis of wave energy converters
with different motion characteristics (WAVE CONVERTERS). The project has been financed by the
para a Cincia e a Tecnologia), from the Portuguese
Ministry of Science and Technology, under contract
PTDC/EME-MFE/105006/2008. The second author
has been funded by the Portuguese Foundation for
Science and Technology (Fundao para a Cincia e a
Tecnologia,) through a Post Doctoral Fellowship under
the contract no. SFRH/BPD/47210/2008.
REFERENCES
Bhattacharya, R. 1978. Dynamics of marine vehicles. John
Wiley and Sons, New York.
Budal, K. and Falnes, J. 1980. Interacting point absorbers with

controlled motion. In: Power from Sea Waves (B. Count,
ed.), pp. 381399. Academic Press, London.
Buchner, B. 2011. Model tests and simulations on a wave
energy converter based on inverse offshore engineering,
Marine Technology and Engineering, C. Guedes Soares
et al. (Eds.), Taylor & Francis Group, London, pp. 349
367.
Falnes, J. 2007. A review of wave-energy extraction. Marine
Structures 20:18520.
Falco, A.F.O. 2010. Wave Energy Utilization: A Review
of the Technologies. Renewable and Sustainable Energy
Reviews, 14 (2010) 899918.
Guedes Soares, C., Bhattacharjee, J., Tello, M. and Pietra,
L. 2012. Review and classification of wave energy converters. Maritime Technology and Engineering, C. Guedes
Soares et al. (Eds.), Taylor & Francis Group, London (this
volume).
Henderson, R. 2006. Design, simulation, and testing
of a novel hydraulic power take-off system for the
Pelamis wave energy converter, Renewable Energy 31
271283.
Hogben, N., Da Cunha, L. F., and Ollivier, H. N. 1986. Global
Wave Statistics, British Marine Technology, Publishing
Urwin Brothers Limited, London.
Johanning, I., Smith, G.H. and Wolfram, J. 2006. Mooring
design approach for wave energy converters. Journal of
Engineering for the Maritime Environment 220:15974.
Journe, J.M.J. and Adegeest, L.J.M. 2003. Theoretical manual of strip theory, program SEAWAY for Windows.
Report 137, September.
Kofoed, J.P., Frigaard, P., Friis-Madsen, E. and Srensen, H.C.
2006. Prototype testing of the wave energy converter wave
dragon. Renewable Energy, pp. 31:1819.
Margheritini, L., Vicinanza, D. And Frigaard, P. 2009. SSG
wave energy converter: Design, reliability and hydraulic
performance of an innovative overtopping device, Renewable Energy, 34, 13711380.
McCabe, A.P., Bradshaw, A., Meadowcroft, J.A.C. and
Aggidis, G. 2005. Developments in the design of the PS
Frog Mk 5 wave energy converter. Renewable Energy,
31:14151.
Morison, J.R., OBrien, M.P., Johnson , J.W., and Schaaf,
S.A. 1950. The Force Exerted by Surface Waves on Piles.
Petroleum Transactions 189, pp. 149157.
Pessoa, J., Fonseca, N., and Guedes Soares, C. 2010. Experimental and numerical study of the depth effect on the first
order and slowly varying motions of a floating body in
bichromatic waves. Proc. Of the 29th Int. Conf. in Offshore
Mechanics and Arctic Engineering (OMAE). Shanghai,
China, June 611. Paper n OMAE2010-21188.
Price, A.A.E., Dent, C.J. and Wallace, A.R. 2009. On the
capture width of wave energy converters, Applied Ocean
Research, 31, 251259.
Retzler, C. 2006. Measurements of the slow drift dynamics
of a model Pelamis wave energy converter, Renewable
Energy 31, 257269.
Rusu, E. and Guedes Soares, C. 2009. Numerical Modelling to Estimate the Spatial Distribution of the Wave
Energy in the Portuguese Nearshore. Renewable Energy.
34(6):15011516.
Silva, D. Martinho, P. and Guedes Soares, C. 2012. Modelling
the Wave Energy for the Portuguese Coast. Maritime Technology and Engineering, C. Guedes Soares et al. (Eds.),
Taylor & Francis Group, London (this volume).
Tello, M., Ribeiro e Silva, S. and Guedes Soares, C. 2011.
Seakeeping Performance of Fishing Vessels in Irregular
Waves. Ocean Engineering. 38(56):763773.
665
Thorpe, T.W. 1999. A brief review of wave energy. UK

Department of Trade and Industry. Report ETSU-R120.
Vantorre, M., Banasiak, R. and Verhoeven, R. 2004. Modelling of hydraulic performance and wave energy extraction by a point absorber in heave.Applied Ocean Research,
26:672.
Wang, L., Guo, Z., and Yuan, F. 2010. Quasi-Static ThreeDimensional Analysis of Suction Anchor Mooring System. Ocean Engineering, Vol. 37, pp. 11271138.
WAMITUser Manual Version 6.4, 6.4PC, WAMIT Inc.
Weber, J., Mouwen, F., Parrish, A. and Robertson, D. 2009.

Wavebobresearch & development network and tools in
the context of systems engineering. In: Proceedings of 8th
European Wave Tidal Energy Conference; pp. 416420.
Weinstein, A., Fredrikson, G., Parks, M. J. and Nielsen,
K. 2004, Aquabuoy, the offshore wave energy converter numerical modeling and optimization Proc.
MTTS/IEEE Techno-Ocean 04 Conf., Kobe, Japan, vol. 4,
pp. 18541859.
666
Diverse

The CARCACE project deepwater platforms Modular designs

for in situ experiments
D. Ribeiro & A. Hilrio
CESAM and Biology Department, University of Aveiro, Portugal
ABSTRACT: The CARCACE project aims to study ecosystems created by large organic falls in the deepsea and it required the deployment of cow carcasses at 1000 m depth in the Setubal Canyon. The carcasses
were attached to a platform designed to enable the deployment of a variety of instruments and experiments.
A total of 5 survey dives will take place every six months using ROVs or research submersibles. Various
prototypes of instruments and innovative systems can be tested during theses dives. Sediment traps, current
meters, hydrophones, cameras and miniaturized automated labs, are some examples of the instruments that can
be attached to the platforms. It will also be possible to conduct long-term studies of materials resistance, which
will be subjected to 100 atmospheres for up to two years. Cooperation with various entities that can provide
technologies for the project, through Barter Agreements has been initiated and some examples are presented.
1 THE CARCACE PROJECT

1.1
General scientific background
Most deep-sea communities are limited by low food

availability (Gage & Tyler 1991). Even in highly productive surface waters, deep-sea animals are generally
adapted to a relatively low supply of carbon. Some
deep-sea animals, however, can alter their metabolism,
growth rate, feeding behaviour, and reproduction to
exploit episodic pulses of organic enrichment, such
as animal carcasses, plant debris, and wood (Gage &
Tyler 1991, Levin 2002). Decomposing whale carcasses, for example, deliver large pulses of organic
material to the seafloor and serve as habitat islands for
unique assemblages of deep-sea macrofauna (Smith &
Baco 2003). Varying in time and space, these discrete
resource patches are thought to contribute to habitat
complexity and increase biodiversity in the deep-sea
environments (Baco & Smith 2003, Dahlgren et al.
2004, Braby et al. 2007).
In situ studies on whale carcasses showed at least
three successional stages following the arrival of
a fresh whale carcass at the deep-sea floor. After
soft tissues removal by necrophages and scavengers,
opportunistic species take advantage of the organic
enrichment of the sediments and exposed bones. This
is followed by a sulphophilic stage where chemoautotrophy is sustained by sulphide coming from the
anaerobic breakdown of bone lipids (Smith & Baco
2003). Depending on the size of the whale, its bones
can contain enough oil to support chemosynthetic
species for as long as 80 years. Because whale falls
share a number of taxa with other chemosynthetic
habitats, such as cold seeps or hydrothermal vents,
they have been hypothesized to act as steppingstones
in the evolution and distribution of chemoautotrophic

communities (Smith et al. 1989, Distel at al. 2000).
During both the opportunistic and the sulphophilic
stages, whale falls also harbour a number of potentially endemic species (Smith & Baco 2003), the
most outstanding being perhaps the recently described
bone-eating worm Osedax (Rouse et al. 2004).
The ecology, biogeography and evolution of deepsea whale fall communities have in the last 10 years
become topics of broader interest to the oceanographic
and marine biological communities, setting the stage
for more detailed ecological and phylogenetic studies.
Surprisingly, all long-term studies of whale carcasses
and other large organic falls on the deep-sea have been
restricted to the Pacific (Smith & Baco 2003, Milessi
et al. 2005, Braby et al., 2007, Fujiwara et al. 2007)
and the role of these habitats in the Atlantic Ocean
has been overlooked. In the Atlantic, the only observations of mammal carcasses were either on shallow
waters (Glover et al. 2005) or short-term observations
(Kemp et al. 2006), which being of utmost importance
to understand these habitats, are not sufficient to evaluate the deep-sea community response to intense pulses
of organic falls nor their importance as sulphide-rich
habitat islands at the Atlantic Ocean deep-sea floor.
Present and past population connectivity between
cold seeps in both side of the Atlantic and hydrothermal vents in the Mid-Atlantic Ridge (MAR) is
currently under debate (Cordes et al. 2007) but the
importance of large organic falls has not been discussed yet. A first insight into the role that organic
decomposition play in providing habitat for chemoautotrophic invertebrate symbioses in the deep Atlantic
was given by the discovery of vestimentiferan tubeworms, at the site of a shipwreck 30 miles of the
669
1.2 Objectives and working plan
Figure 1. Deployment of five cow carcasses in the Setubal

Canyon in March of 2011.
Table 1. Cruise planning for the experimental site in the

Setubal Canyon.
Mission
Date
Vessel
Deployment
Sampling
Sampling
Sampling
Sampling
5th March 2011

November 2011
March 2012
November 2012
March 2013
NRP Alm. Gago Coutinho

To be determined (TBD)
TBD
TBD
TBD
coast of Spain at 1160 m depth (Dando et al. 1992).

Vestimentiferans are predominant constituents of
Pacific vent and both Pacific and Atlantic seep habitats
but their absence from hydrothermal vents in the Mid
Atlantic Ridge (MAR) and cold seeps in the Northeast
Atlantic remains one of the most intriguing questions for scientists studying deep-sea chemosynthetic
environments.
Only a strategic selection of new study sites
has the potential to resolve a global map of
deep-sea chemosynthetic environments biogeography and define biogeographic boundaries. The deep
Portuguese margin and the Azores area are not due
to receive large organic inputs from land but encompass an important area of the distribution of cetaceans
in the Northeast Atlantic both resident (e.g. Hyperoodon ampullatus) and migrating (e.g. Balaenoptera
physalus) (Harwood & Wilson 2001, Silva et al. 2003)
and therefore are exceptional places to study the
impact that marine mammal carcasses might have in
the deep Atlantic Ocean.
During the CARCACE (Colonization of mammal
carcasses in the deep Atlantic Ocean), mammal carcasses will be deployed at a 1000 m depth, one in the
S. Jorge channel (Azores) and another in the Setubal
canyon. These two in situ experiments will allow a
comparison between the colonization dynamics and
species composition between areas with different geological and hydrological settings and address questions
related to species distribution, dispersal strategies
and phylogeography.
Studies of any new environment generally fall into

three consecutive phases: composition, structure and
dynamics. During the CARCACE project these three
phases will be followed to address a wide range
of questions in the biology of habitats provided by
large organic falls. The objectives for this study are
1) describe deep-water mammal carcass fauna in
the Atlantic, 2) investigate functional anatomy of
organic-fall specialists, including potential bacterial
endosymbioses, 3) determine phylogenetic relationships of organic-fall specialists and their closest
relatives to evaluate evolutionary hypotheses 4)
analyse the trophic structure of colonizing metazoan
assemblages 5) elucidate the importance of large
organic falls as a stepping-stone habitat for vent and
seep species in the Atlantic.
To achieve these objectives it was initially planned
to deploy two carcasses of stranded marine mammals
in two places where marine mammals occur naturally
and possibly died. However, because of logistic constraints it was decided to use cow carcasses (Fig. 1)
as an alternative simulator of a large organic fall.
Cow bones have been used in other areas of the worlds
oceans (e.g. Monterey Canyon) and have been colonized by the same groups of animals that colonize
whale falls (Jones et al. 2008). The Setubal Canyon and
the S. Jorge Channel, both at approximately 1000 m
depth, were chosen as study sites. These sites have
been selected to maximize integration with a variety
of geological and biological data obtained in other
projects, and also because of their proximity to shore
and the laboratories involved in the project, which
will allow an efficient use of ship-time and virtually undisturbed retrieval of bones with live fauna for
laboratory experiments and observation. Each experimental site will be visited every six months during
a period of two years (Table 1), video surveyed and
sampled using a Remotely Operated Vehicle (ROV).
This approach will allow the identification of succession patterns, a subject of broad ecological interest
because succession provides insight into deep-sea
community response to extreme point-source enrichment, both natural (e.g. from whale falls) and anthropogenic. The timescales over which large quantities
of organic material might become assimilated into the
seafloor community, and the recovery time of the local
community after dissipation of enrichment are issues
of relevance to deep-sea carbon flux and to predicting
the effects of analogous anthropogenic organic enrichment in the deep-sea floor (e.g. relocation of sewage
sludge, fishery discards, and disposal of animal waste).
NEW DESIGNS FOR PLATFORMS
2.1 Specific challenges of CARCACE project

Data collection at deep-sea organic falls requires
an access to advanced manned or unmanned submersibles, reducing the possibilities for a rigorous
670
sample design. On the other hand, recurrent sampling is essential because of the dramatically different
faunal communities at different stages while the
community develops, and sampling at different spatial scales are important to address questions related to
species distribution, dispersal strategies, phylogeography, population genetics and population dynamics. On
top of these constraints, the study of artificially sunken
organic matter, as in the CARACE project, requires
the deployment and the sinking of a large quantity of
organic matter to the seafloor, which is usually done
by means of a concrete weight or railway bars. Also,
the risk that relatively large predators feeding on the
carcasses that would end up scattering the bones over
a wide area on the seafloor should be avoided.
To optimize the deployment activities we propose
to design a platform that allows to maximize the
amount of organic material deployed and at the same
time fitting positioning and environmental monitoring equipment. We developed an initial customized
deployment strategy that was intended to allow the
assembly of a particularly large lander, and the subsequent attachment of at least two tons of organic matter,
without having to use large oceanographic ships.
2.2
Innovative designs
A special platform was designed as a digital prototype using Autodesk Inventor. This floating platform
uses the JETFLOAT commercial system, together with
a proprietary new support infrastructure design from
BAROMETRICS (a company that is currently being
set up). This system allows the platform to be towed at
higher speed than it would otherwise be possible. Four
ramps are used to transport and deploy large bovines.
In the center of the platform there is a moon-pool and
a special A-Frame structure that will allow the descent
of a special partially-assembled lander. At a depth of
15 meters, a team of divers completes the assembly of
the large lander.The lander has some moving parts, and
it follows a design philosophy relatively similar to the
project intelliSTRUCT being researched at SINTEF
from Norway that apply the principles of Tensegrity
(Tensional Integrity Structuring), that are also known
as floating compression. Muscles and bones allow
various complex movements of the human body using
these mechanical phenomena.
We are also exploring other possible overlap areas
between naval engineering and biology, using the
concepts from Biomimicry research. Both the initial
and final designs of the first platform have features to
allow easy interfacing with possible future advanced
underwater robotic systems designed along the lines
of the studies of root systems for ground anchoring
developed by the biomimetics group of the Advanced
Concepts Team from the European Space Agency
(Dario et al. 2008).
Uncertainties with the time needed to fully test
this new design, together with the additional time
that would be required to obtain the necessary sea
worthiness certification, led to the postponement of
Figure 2. General view of the instrument platform, with the

central quadripod, a passive sonar reflector, and 16 mooring
attachments that were used to attach the 5 cows.
the construction and deployment of this new floating platform for the first mission. Instead, during a
design review meeting, a more simple design was chosen and built, focusing on the main task that need to
be preformed by the platform, which is, above all,
to be able to anchor in a reliable manner for three
years, all the organic material that will give rise to
the chemosynthetic ecosystem.
Since the intensity of the currents at 1000 m in
the Setubal Underwater Canyon is unknown, during
a design review meeting it was decided that it would
be desirable to have a concrete block (Fig. 2) weighing two tons to serve as a reliable anchoring system
able to cope with the possible existence of occasional
strong currents.
2.3 Innovative organizational options
One innovation of the first underwater platform from
the CARCACE project consists of the use of barter
agreements at national level in a way that enabled
a faster pace of the final design, construction, and
deployment. Barter agreements are frequently used
as tools for international cooperation. These arrangements allow for no-cost exchanges of ship-time and
major marine equipment, and promote a more efficient
and cost-effective use of each countrys resources by
giving the scientific communities access to a wider
range of marine facilities and geographical areas
in a given year than would otherwise be possible.
At the national level, and among private companies
and universities, however, these tools are not used
as frequently as they could be. The authors believe
that it is possible to deepen and broaden the use
of this enabling tool at the national level. Besides
the barter agreements, there is also the possibility to
increase the impact on the market of each mission
or research dive. This is based on the maximization
of a new parameter based on the stakeholder concept (R. Edward Freeman, 1984): stakeholder density
per mission. In space projects, due to the very high
cost of orbital launches this value is usually very high.
671
Figure 3. The reinforced concrete platform has four longer

suction conduits on the corners. In spite of the 19 embedded structures, there was still space available for more
instrumentation to be housed inside.
Figure 4. NRP Gago Coutinho from Instituto Hidrogrfico

was chosen due to the ability to use Dynamic Positioning.
An acoustic transponder was used to release the concrete
platform with several embedded experiments.
In oceanographic projects, oftentimes, it tends to be

relatively lower. For example, there are some cases
where research dives take place using tools and infrastructures that could easily accommodate additional
biological experiments and materials resistance tests.
Likewise, different universities occasionally miss out
on possible cooperation efforts. Coordination among
research groups with very different interests may be
difficult but it is possible to bypass this potential difficulty by establishing joint efforts at the private level.
New cooperation strategies are being explored through
the application of new organizational paradigms developed from studies of the evolutionary importance of
altruism at the biological level (Nowak & Highfield
2011). This also has the additional advantage to open
up new possibilities for private research to take place
alongside usual academic research. Some of the new
trends in aerospace design, towards faster, cheaper,
better missions, as described in the book The Logic
of Microspace (Fleeter, 2000) were incorporated on
our design and will continue to guide our approach to
new oceanographic instrumentation design.
using the Portuguese research vessel NRP Almirante

Gago Coutinho (Fig. 4). Before deploying the platform we surveyed the site seafloor with the vessels
multibeam echosounder.
The deployment was made on a flat area of the
Setubal Underwater Canyon, located near the Espichel
Cape.
3
3.1
FIRST MISSION
Construction of the concrete platform
The reinforced concrete platform weights 2000 kg and

has embedded structures to hold scientific instruments
and/or experiments (Fig. 3). These are: four suction
conduits; four cylindrical wells around the central area;
ten smaller experiment holders near the edges; one
larger experiment holder tube. Along the edges, there
are eight groups of markers designed to provide a
visual reference to ROV operators.
3.2
Deployment of the first platform
On the 5th of March, 2011 the first platform was

deployed at 38 16.85 N; 09 06.68W at 1004 m depth
3.3 Initial barter agreements

Although many exploratory contacts have been established with a wide variety of stakeholders, only three
barter agreements have been established up to now.
The possibility to develop and suggest a new standardized legal mechanism to support these particular
efforts is being studied, possibly with the adoption
of intellectual property management solutions using
some of the Creative Commons (Aliprandi, 2010)
principles.
3.4 Support from national veterinary authorities
The need to find a way to attach the cows to the platform, so that, on one hand they would not become
detached by predators, but that would be flexible
enough to mimic the bone layout of a whale fall, led
to the cooperation with Professor Saraiva Lima, a veterinarian from Faculdade de Medicina Veterinria.
A metal cable was introduced through the oesophagus
of the five cows in order to anchor them permanently
to the platform. Industrial straps were also used.
The national veterinary authorities (Direco
Geral de Veterinria DGV) were instrumental in
allowing access to the carcasses of cows that died of
natural causes, and provided key clarifications about
the legal framework required to enable the use of
bovines for scientific purposes. The carcasses were
collected at the farms where they died and taken
to a factory (ITS Industria Transformadora de
Subprodutos, S.A.) where the bowels were removed
and the carcasses washed and sown.
672
Figure 5. Besides providing a special type of cement that

was developed for this experiment, there are six different
samples attached to the platform inside bags that will allow
their retrieval by ROVs. There is also an additional sample
of a cement rod encased in PVC.
Figure 6. Hydrophone casing being placed on one of the

central cylindrical slots, for watertight testing.
Several cetaceans that die and wash up on the

shores create complex disposal problems to the coastal
sanitary authorities due to their large size. There is a
possible future role for marine biologists to help with
this problem by suggesting suitable locations where
the whale carcasses can be sunk with great benefit
for this research project.
3.5
Sponsorship from SECIL
In order to seize the opportunity to evaluate the

behaviour of concrete at a depth of 1000 meters,
SECIL developed a special kind of cement. After
2013, there is the possibility to recover the platform,
thus allowing an unprecedented occasion to study the
behaviour of concrete structures at greater depths than
usual. SECIL also used this opportunity to test six
different types of concretes that were placed inside
mesh bags attached to the platform (Fig. 5). These
samples will be retrieved during the sampling dives
of the CARACE project.
3.6
Cooperation with marsensing company
Marsensing is a spin-off from the Underwater

Acoustic Signal Processing Laboratory of the University of the Algarve. The goal of this collaboration is
to test the possibility to document the evolution of the
deep chemosynthetic ecosystem with new advanced
bioacoustic tools. The platform was designed to allow
the future placement of batteries that will be able to
power the hydrophones for six months (Fig. 6).
3.7
Cooperation with Adobe Engenharia S.A.
The fast fabrication of the platform was enabled by

the initiative of a private company that is currently
developing prototypes for underwater instruments and
innovative support structures (Fig. 7). The company
has the technical capability to contribute to the study
Figure 7. Adobe Engenharia has the capability to build a

very wide variety of support infrastructures quickly. This
quadripod lifted the full weight of the concrete platform
flawlessly.
of new anchoring systems, through soil deformation

analysis.
NEXT MISSIONS
The recent acquisition of a deep-sea ROV by the

Portuguese Task Group for the Extension of the
Continental Shelf (EMEPC) provides an opportunity
to the Portuguese scientific community to develop
research on the deep-sea, especially on in situ
observation and experimentation.
As an example, the concrete platform that is currently deployed at a depth of 1000 meters has four
embedded green tubes that have a diameter deliberately compatible with the suction sampler from the
Luso ROV. This feature will allow sampling under
the platform in an area that otherwise would be inaccessible. Several other aspects have been designed in
673
order to maximize accessibility and interoperability to

various underwater systems.
4.1
Four survey dives
Several different commercial ROVs are being evaluated as potential tools. The location of the instrumentation platform, being relatively near the shore, allows
a wider selection of support vessels that will be able
to make a short detour from their main mission and
therefore provide flexible opportunities to conduct the
survey dives. Each survey dive must bring back six
pushcores of sediments, as well as a variety of underwater organisms that will be collected in order to study
the evolution of the chemosynthetic ecosystem.
In order to document the gradual emergence of
new life forms on the experiment site, the dives are
scheduled to take place every six months.
4.2
Experiments that can be deployed
A special emphasis is to be given to materials research,

in part because of the relatively rare opportunity
of monitoring over time the various changes that
will occur. The variety and quantity of experiments
currently envisaged is relatively similar to the ones
that were deployed on the Long Duration Exposure
Facility LDEF payload that orbited the earth for
five years from 1984 to 1990 (Clark et al 1984).
4.3
Instruments that can be deployed
Similarly to what is happening with coastal aquatic

environments there is a vast array of instruments
that can be used to monitoring deep-sea ecosystems
(Miller et al. 2005). Sediment traps, current meters,
acoustic beacons, hydrophones, or even cameras
and miniaturized automated labs, are some of the
examples of the instruments that can be attached to
the deepwater platforms.
4.3.1 Mounted on the concrete platform
Currently there are several empty slots on the concrete platform. Along the sides, there are 10 cylindrical
support structures of 50 mm diameter for sediment
traps or other apparatus that may be available. Three
rectangular empty slots can be used to place various
boxes. There are 12 attachment points on the legs
of the quadripod. Four upper attachment points can
also be used. Three inner cylindrical cavities under the
quadripod legs are still available (the fourth is occupied
by the Marsensing Hydrophone watertight test).
4.3.2 Mounted on the ROV cage
Oftentimes ROV cages are used with several empty
areas that could easily carry various sensors and
experiments. In some cases, waterproofing tests of
small subsystems can be performed. The experiment
from Marsensing that is currently underwater, for
instance, only needed this kind of experimental dive
opportunity.
4.3.3 Piggyback payloads on ROVs and subs

In order to increase stakeholder density per dive, a
number of different ways to attach secondary instrument experiments is being designed. In some cases,
customized booms will allow a considerable increase
to the scientific yield of each dive. Lightweight
hydrodynamic fairings will be added if necessary,
to avoid any serious disruption of performance.
4.3.4 Deployed by additional mini-landers
By using PVC tubes filled with concrete, connected
by cables, it will be possible to build smaller support
structures. This will allow the placement of additional
payloads around the initial instrument platform.
FUTURE PROJECTS
The systems currently being developed can gradually

become national standards. There is a possibility to
cultivate self-sufficiency in some areas of oceanographic instrumentation. Some components could
be built under license from well established subsea
technology companies. A Portuguese system similar
to the Spanish OBSEA project (Carreras et al. 2009)
or the American Monterey Accelerated Research System cabled seafloor observatory, is currently being
envisaged for the Luis Saldanha Marine Park area.
The location of the first concrete platform that was
deployed on the 5th of March 2011 could become
the terminal node of a future network of near shore
cabled underwater observatories, linking the seafloor
around the Sesimbra region. Another system could be
located in the Nazar underwater canyon, allowing
for a wide variety of oceanographic parameters.
Scientists from CESAM and other institutions will
be invited to provide various instrumentation and
scientific support to the initial nodes of a prototype
version of a cabled underwater observatory. All the
systems and infrastructures developed and tested during the CARCACE project can be used to support
theses developments. Scientists together with partners
in the telecom industry can contribute to the efforts
currently underway for the harmonization of Ocean
Observing Systems (Del Rio & Delory 2010).
Some of the foreseeable large data output from
all theses instruments and experiments can be analyzed with the involvement of the general public under
Citizen Science projects relatively similar to the
Stardust@home supported by NASA, the University
of California in Berkeley and the Planetary Society.
CONCLUSIONS
The scientific goals of the CARCACE project are

demanding and require the systematic development
and continued deployment of advanced deepwater
platforms. Several new designs are being developed.
The existing barter agreements, together with additional cooperative ventures, using new organizational
674
paradigms, may help overcome several obstacles that

limit the scope and the depth of joint efforts among
different stakeholders at the national level. The capabilities currently being developed may enable the
launch of a new project to create an innovative underwater cabled observatory network that can allow the
creation of new products and services.
ACKNOWLEDGEMENTS
The CARCACE project is financed by the European
Regional Development Fund (ERDF) through the
COMPETE programme and by national funds
through FCT, project contract number PTDC/MAR/
099656/2008.
The second author is financed by the FCT grant
SFRH/BPD/22383/2005.
REFERENCES
Aliprandi, S (2010) Creative commons: a user guide.
Lulu.com Publishers. 84 pp.
Baco AR, Smith CR (2003) High species richness in
deep-sea chemoautotrophic whale skeleton communities.
Marine Ecology Progress Series 260:109114.
Braby CE, Rouse GW, Johnson SB, Jones WJ, Vrijenhoek
RC (2007) Bathymetric and temporal variation among
Osedax boneworms and associated megafauna on whalefalls in Monterey Bay, California. Deep-Sea Research
Part I-Oceanographic Research Papers 54:17731791.
Carreras N, Nogueras M, Arbs A, Antoni M (2009)
Watertight tests for OBSEA equipments in the hyperbaric chamber. Martech 2009 Proceedings Instrumentation Viewpoint Number 8. Third international workshop
on marine technology, November 19th 20th, Vilanova I
la Geltr.
Clark LG, Kinar WH, Carter Jr DJ, Jones Jr JL (1984)
The long duration exposure facility (LDEF). National
Aeronautics and Space Administration SP-473.
Cordes EE, Carney SL, Hourdez S, Carney RS, Brooks
JM, Fisher CR (2007) Cold seeps of the deep Gulf of
Mexico: Community structure and biogeographic
comparisons to Atlantic equatorial belt seep communities.
Deep-Sea Research Part I-Oceanographic Research
Papers 54:637653.
Dahlgren TG, Glover AG, Baco A, Smith CR (2004) Fauna
of whale falls:
systematics and ecology of a new polychaete (Annelida: Chrysopetalidae) from the deep Pacific Ocean. Deep
Sea Research Part I: Oceanographic Research Papers
51:18731887.
Dando PR, Southward AJ, Southward EC, Dixon DR,
Crawford A, Crawford M (1992) Shipwreck tube worm.
Nature 356:667.
Dario P, Laschi C, Mazzolai B, Corradi P, Mattoli V,
Mondini A, mancuso S, Mugnai S, Masi E, Azzarello E,
Hlavacka A, Pandolci C, Seidl T (2008) Bio-inspiration
from Plants Roots. Final Report. Final Report. European
Space Agency.
Del Rio J, Delory E, (2010) From ocean sensors to

traceable knowledge by harmonizing ocean observing
systems. Earthzine Fostering Earth Observation &
Global Awareness ICEO IEEE.
Distel DL, Baco AR, Chuang E, Morrill W, Cavanaugh
C, Smith CR (2000) Do mussels take wooden steps to
deep-sea vents? Nature 403:725726.
Fleeter R (2000) The logic of microspace. The Space Technology Library. Microcosm Press, Torrance, California.
447pp.
Fujiwara Y, Kawato M, Yamamoto T, Yamanaka T,
Sato-Okoshi W, Noda C, Tsuchida S, Komai T,
Cubelio SS, Sasaki T, Jacobsen K, Kubokawa K,
Fujikura K, Maruyama T, FurushimaY, Okoshi K, Miyake
H, Miyazaki M, NogiY,Yatabe A, Okutani T (2007) Threeyear investigations into sperm whale-fall ecosystems in
Japan. Marine Ecology 28:219-232.
Gage JD, Tyler PA (1991) Deep-sea biology: a natural
history of organisms at the deep-sea floor. Cambridge
University Press.
Glover AG, Kallstrom B, Smith CR, Dahlgren TG (2005)
World-wide whale worms? A new species of Osedax
from the shallow north Atlantic. Proceedings of the
Royal Society B-Biological Sciences 275:387391.
Harwood J, Wilson B (2001) The implications of developments on the Atlantic Frontier for marine mammals.
Continental Shelf Research Volume 21, Issues 810:
10731093.
Jones WJ, Johnson SB, Rouse GW, Vrijenhoek RC (2008)
Marine worms (genus Osedax) colonize cow bones.
Proceedings of the Royal Society B-Biological Sciences
272:25872592.
Kemp KM, Jamieson AJ, Bagley PM, McGrath H, Bailey
DM, Collins MA, Priede IG (2006) Consumption of large
bathyal food fall, a sixmonth study in the NE Atlantic.
Marine Ecology Progress Series 310: 6576.
Levin LA (2002) Environmental and regional trends in deepsea diversity: Complexity amidst pattern. AAAS Annual
Meeting and Science Innovation Exposition 168:A30.
Miller R, Del Castillo C, McKee B (2005) Remote sensing of
coastal aquatic environments technologies, techniques
and applications. Springer, Berlin. 347 pp.
Milessi AC, Sellanes J, Gallardo VA, Lange CB (2005)
Osseous skeletal material and fish scales in marine
sediments under the oxygen minimum zone off northern
and central Chile. Estuarine Coastal and Shelf Science
64:185190.
Novak MA, Highfield R (2011) Super Cooperators. Altruism,
evolution, and why we need each other to succeed. Free
Press, New York. 352 pp.
Rouse GW, Goffredi SK, Vrijenhoek RC (2004) Osedax:
bone-eating marine worms with dwarf males. Science
305:668671.
Silva MA, Prieto R, Magalhes S, Cabecinhas R, Cruz
A, Gonalves JM, Santos RS (2003). Occurrence and
distribution of cetaceans in waters around the Azores
(Portugal), summer and autumn 19992000. Aquatic
Mammals 29(1): 7783.
Smith CR, Baco AR (2003) The ecology of whale falls at the
deep-sea floor. Oceanography and Marine BiologyAnnual
Review: 311354.
Smith CR, Kukert H, Wheatcroft RA, Jumars PA, Deming JW
(1989) Vent fauna on whale remains. Nature 341:2728.
675

MAR International Shipping Register of Madeira

Miguel Sequeira
International Shipping Register of Madeira, Funchal, Portugal
ABSTRACT: MAR The International Shipping Register of Madeira, created within the framework of the
International Business Centre of Madeira (IBCM), provides Portugal with an important role in the in shipping
world. All entities undertaking the maritime transportation of persons and goods may register vessels in MAR,
flying the Portuguese flag. In addition to commercial vessels and oil-rig platforms, both commercial and pleasure yachts may be registered in MAR. MAR, as a Portuguese register, has become a credible and competitive
alternative compared to other international registers, maintaining the quality levels and the safety culture of
an E.U. register. All international conventions ratified by Portugal are fully applicable to and respected and,
as a consequence, MAR has never been regarded as a flag of convenience. MAR has a Technical Commission (TC) composed of a President, directly appointed by the Ministry of Public Works, Transportation and
Communications, a representative of the Madeira Autonomous Region and a representative of the Portuguese
Institute for Ports and Maritime Transportation (IPTM). The duties of the Technical Commission of MAR include
the management of the registration process, as well as the establishment of the crew composition, being also
responsible, in close articulation with IPTM, for the full compliance of international conventions ratified by
Portugal on safety and the preservation of the quality of life on board and at sea.
1
1.1
INTRODUCTION
Foreword
The objective of this paper is to provide ship-owners,

marine and ship consultants and any other entity
interested in the International Shipping Register of
Madeira MAR with the most relevant information
regarding all the procedures and steps for the registration of commercial vessels and yachts, as well
as the incorporation of shipping companies within
the ambit of the International Business Centre of
Madeira (IBCM). All the relevant operational issues
are addressed, such as the registration procedure and
required documentation, tax regime, applicable fees
and laws and other general information.
1.2 The International Business Centre of Madeira
The International Shipping Register of Madeira
MAR was created under the framework of the International Business Centre of Madeira (IBCM), aimed
at giving the island of Madeira a predominant role in
shipping and, as Portugals second and international
shipping register, to contribute to reduce the countrys
flagging out process, attracting new ship-owners
and vessels.
The International Business Centre of Madeira has
been created to attract foreign direct investment in
order to diversify and modernize the islands economy.
It has been conceived as a programme of regional
development and as a State Aid regime is fully
approved and supported by Portugal and the European
Union. The licensing process for the incorporation of

companies within the legal framework of the IBC of
Madeira is carried out through S.D.M. Sociedade de
Desenvolvimento da Madeira, S.A., a mixed capital
company appointed by the Regional Government of
Madeira to promote and administer the International
Business Centre of Madeira, including the collecting
of fees due by vessels registered in MAR.
1.3 The Technical Commission of MAR

MAR has a Technical Commission (TC) composed
of a President, directly appointed by the Portuguese
Ministry of Public Works,Transportation and Communications, a representative of the Madeira Autonomous
Region and a representative of the IPTM The
Portuguese Institute for Ports and Maritime Transportation. The maritime background and technical
expertise of all members of the TC assures the full
understanding of the maritime transportation business and the implementation of fast and adequate
procedures.
The duties of the Technical Commission of MAR
include the management of the registration process,
as well as the establishment of the crew composition, being also responsible, in close articulation
with IPTM, for full the compliance of international
conventions ratified by Portugal on safety and the
preservation of the quality of life on board and at sea.
MAR may delegate on Classification Societies,
acting as Recognized Organization on behalf of
Portugal, the undertaking of certain tasks which lead
677
to the issuance of international statutory certificates

required by the relevant instruments of IMO and ILO.
1.4
IPTM Instituto Porturio e dos Transportes

Martimos
IPTM The Portuguese Institute for Ports and

Maritime Transportation is the Portuguese authority responsible for the supervision, regulation and
inspection of the maritime sector, as well as for the
management and administration of the ports under its
jurisdiction.
The Technical Commission of MAR works in articulation with IPTM in what regards the recognition
of crew competency certificates, as well as on the
carrying out of surveys to vessels registered in MAR.
2 THE INTERNATIONAL SHIPPING REGISTER
OF MADEIRA MAR
2.1 About MAR
MAR is a credible and competitive alternative to
other international registers, maintaining the quality
levels and the safety culture of an E.U. register. In
fact, MAR has implemented high quality standards
regarding the vessels that are accepted for registration
and, as a consequence, MAR is not regarded by the
ITF (International Transport Workers Federation) as a
flag of convenience and has contributed significantly
for the inclusion of the Portuguese flag in the Paris
MOU White List.
This international recognition of the quality associated to MAR contributes to less frequent inspections
carried out by Port State Control authorities to vessels
registered in MAR.
All entities undertaking the maritime transportation of persons and goods may register vessels in
MAR, flying the Portuguese flag and having therefore full access to continental and domestic cabotage. Such entities may be companies, branches,
agencies, or legal representations licensed or not to
operate within the framework of Madeiras International Business Centre. Commercial vessels, oil-rig
platforms and commercial and pleasure yachts may
be registered in MAR, benefiting from the services
of the Private Commercial Register and Notary of
Madeiras IBC.
Entities licensed within the framework of the
International Business Centre of Madeira and whose
activity is that of maritime transportation will be part,
for all due effects, of Madeiras IBC. Such entities
with vessels registered in MAR will fully benefit from
the tax incentives provided by Madeiras IBC.
are not obliged to contribute to the Portuguese social

security regime.
However, the crew may choose the Portuguese
voluntary social security regime or any other type of
protection scheme including the P&I Club.
In addition, the wages of the crew of commercial vessels and yachts registered in MAR are also
exempt from personal income tax, according to the
Tax Incentives Statute.
2.2.2 Shipping companies
It is not mandatory to incorporate a company in
Madeiras IBC in order to proceed with the registration of a ship in MAR, although some form of legal
representation must be nominated with all necessary
powers to assure full representation before the State
and Regional authorities. In the case a company is
formed, the tax regime of the IBC of Madeira is fully
applicable.
The present regime of tax benefits allows the incorporation of new entities within the ambit of the IBC
until the end of 2013, granting significantly reduced
corporate tax rates until 2020, as evidenced in the
following table:
APPLICABLE TAX
YEARS
3%
4%
5%
From 2007 to 2009

From 2010 to 2012
From 2013 to 2020
The distribution of dividends to shareholders

of shipping companies operating within the legal
framework of the IBC of Madeira is exempt from
withholding taxes, regardless of nationality or place
of residence of the shareholder.
All companies licensed to operate in the International Business Centre may also benefit from the large
network of international treaties to avoid double taxation ratified by Portugal. The list of treaties may be
consulted at http://www.ibc-madeira.com/.
2.2.3 VAT
Yachts registered in MAR will benefit from the application of a 15% VAT rate. In addition, in case the yacht
is engaged in charter activities there will be a total
reimbursement of the VAT paid on registration and
no VAT will be applicable to ocean-going commercial
yachts on repair and maintenance operations, as well
as fuel and oil supply in open sea, according to the
Portuguese VAT code.
3
REGISTRATION OF COMMERCIAL
VESSELS AND OIL-RIG PLATFORMS
3.1 Formalities
2.2 Tax regime

2.2.1 Vessels and Yachts
Crew aboard commercial vessels and yachts registered in MAR benefit from a very competitive social
security regime. In fact ship-owning companies with
vessels registered in MAR and their respective crew
Various documents and information on the vessel are

submitted to the Technical Commission of MAR for
analysis and initiation of the registration process:
678
Owner and/or operators contracts, mortgages or

other liens relating to the vessel;
Purchase certificate of the vessel (bill of sale);

Vessels name and two other alternative names;
Application for the attribution of a call sign, as well
as description of communication equipment;
Certified copy of the register Tonnage Certificate;
Name of classification society;
Vessels characteristics and propulsion system;
Shipyard and year of construction;
Copies of the vessels certificates including those of
the classification society;
Copy of the ship Station Licence;
Tonnage measurement data.
The Technical Commission of MAR will decide,

according to the ships records, age and detention history, if a survey will be carried out before registration.
A survey is also carried out in the absence of any of
the following recognised certificates:
Transcript of registry (from any previous registration);

Deletion certificate.
3.2
Permanent registration
Vessels must deliver proof of the deletion of the previous registration, if applicable. However, a document
issued by the competent maritime authorities, stating that the deletion of the previous registration was
requested, may be initially accepted by MAR.
3.2.1 Legal Documentation

The following legal documentation is required for the
permanent registration of a vessel in MAR:
Power of attorney granted by the applicant to a local

legal representative (applicable when the domicile
or the head office of the applicant is located outside
the Autonomous Region of Madeira);
Articles of association concerning the applicant
(when the applicant is a collective person not
licensed to operate within the ambit of the International Business Centre of Madeira );
Proof of license for operations in the MIBC (when
the applicant is a collective person licensed to operate within the ambit of the International Business
Centre of Madeira);
Bill of sale;
Certificate issued by the previous registry;
Deletion certificate;
Permission granted by the mortgagee(s).
All the above documentation must be translated

into Portuguese (certified translation).
3.3 Temporary registration

3.3.1 Bareboat charter in
Vessels may be registered temporarily in MAR
(bareboat charter in), subject to authorization granted
by the owners and by the competent authorities of
the country where the vessel is permanently registered. MAR can make the registration for five (5) plus
another five (5) years (5 + 5), if agreed by the parties.
The application form shall be addressed to the
Technical Commission of the International Shipping
Register of Madeira MAR, with indication of the
charter period.
3.3.1.1 Legal documentation
The following legal documentation is required for
the temporary (flagging-in) registration of a vessel
in MAR:
3.2.2 Technical Documentations

The following technical documentation is required for
the permanent registration of a vessel in MAR:
List of Radio Equipment and to Other Auxiliary

Navigation Equipment;
Name of the Entity Responsible for the Payment of
Radio Accounts;
Cargo Safety Construction Certificate;
International Load Certificate (1996);
International Oil Pollution Prevention Certificate
(IOPP);
International Pollution Prevention Certificate for
the Carriage of Liquid Substances in Bulk;
International Certificate of Fitness for the Carriage
of Dangerous Chemicals in Bulk;
Certificate of Class (Hull and Machinery);
International Certificate of Fitness for the Carriage
of Liquefied Gases in Bulk;
Certificate of Insurance or the Financial Security
in Respect of Civil Liability for Oil Pollution
Damage (C.L.C.);
Declaration of P&I Inscription (Blue Card);
International Tonnage Certificate (1996);
Manning Certificate;
Statement of Compliance with ILO 92 and 133;
Ships General Arrangement Plan;
Fire Control And Life Saving Appliances Plan;
Approved Stability Booklet;
Mark and Model of the Automatic Pilot;
Communications System, Mark and Model
(Bridge/Prow; Bridge/Stern);
Winches, Capstans, Etc. (Mooring Devices)/Mark
and Model (Quantity Prow; Quantity Stern);
Main Engine (Place, Year of Construction and
Name of the Shipbuilder; Quantity; Type; Model;
Series Number; Maximum Power);
Auxiliary Engines/Generators (Quantity; Type;
Model; Series Number; Maximum Power).
Passenger Ship Safety Certificate;

Cargo Ship Equipment Certificate;
Cargo Ship Safety Radio Certificate;
Exemption Certificate;
Ship Station License (Radio Station);
679
Bareboat Charter;
Commercial Certificate of the Charterer;
Power of attorney granted by the charterer to a local
legal representative;
Permission of the Permanent Registry for the
registration of the vessel in MAR;
Permission granted by the mortgagee(s) in case of
any mortgage.
For the temporary registration, certified translations

are not mandatory if the documentation is delivered
in one of the languages accepted by MAR, namely
English, French and Spanish.
3.3.1.2 Deletion of the temporary registration
When the charter period is almost expiring, and if the
charterer does not renew the registration, MAR will
automatically consider such registration deleted.
On the other hand, if the charter period is still ruling,
but the parties wish to delete the registration, they can
do so by signing an addendum to the bareboat charter,
where both parties agree to delete such registration.
3.3.2 Bareboat charter out
Vessels with permanent registration in MAR may be
temporarily registered in other countries (bareboat
charter out). In such cases, authorisation for the bareboat charter out will be granted by MAR. The vessel
will fly that Registrys flag and the right to fly the
Portuguese flag will be suspended until the end of the
charter period.
The application form shall be addressed to the
Technical Commission of the International Shipping
Register of Madeira MAR requesting the temporary
registration of the vessel under another flag.
3.3.2.1 Legal documentation
The following legal documentation must be delivered
to MAR for authorization of the temporary (flaggingout) registration of a vessel:
Bareboat charter;
Certificate of the registry issued by the registry
where the vessel will be temporarily registered
confirming the temporary registration.
3.4
Other requirements
There are no direct restrictions on the age of the vessels

that may be registered, but the technical commission
will decide, on a case-by-case basis, the vessels which
will be accepted.
Vessels must be classified by one of the officially
approved classification societies:
ABS American Bureau of Shipping

BV Bureau Veritas
DNV Det Norske Veritas
GL Germanisher Lloyd
LRS Lloyds Register of Shipping
RINA Registro Italiano Navale
RINAVE Portuguesa
Other classification societies may be included

when recognised by the Portuguese Government.
All vessels registered in MAR are required to
have a manning certificate issued in order to ensure
that seafarers serving on board are qualified, fit and
sufficient in number for the performance of their
duties. The certificates of seafarers serving on board
of vessels registered in MAR shall be issued according to the provisions of the Standards of Training
Certification and Watch Keeping for Seafarers Convention of 1978 (STCW), as amended in 1995.
The ship-owner or his legal representative may
apply for the manning proposal of a vessel already
registered or to be registered in MAR. The application
must be accompanied by the following documents:
Summary identification of the vessel, including technical characteristics, equipment, area and
trade where it will operate;
General arrangement plan;
Safety plan;
Stability book;
Any other elements that the applicant may consider
necessary for such application;
Ships manning proposal duly justified.
Labour contracts shall be signed between the seamen or unions and employer, containing the following terms:
Identification of the parties;

Name of the vessel;
Nature and duration of the intended voyage or
engagement;
Period that each seaman is to be on board;
Title and function to be undertaken by each seamen;
Wages;
Agreement as to the payment of wages;
Termination of the contract.
The captain and 50% of the safe manning of the ship

must be European, of both E.U. and non-E.U. countries, and citizens of Portuguese-speaking countries,
including nationalities such as Russians, Ukrainian
and Polish, allowing therefore for a flexible manning. This requirement may be eliminated or waived
whenever duly justified.
The 95 Amendments to STCW 78 Convention,
renders mandatory for foreign seafarers who perform
services on vessels flying the Portuguese flag, to hold
the recognition of their certificates of competency.
According to the established under regulation I/10
of the referred Convention, the process of recognition of certificates to foreign seafarers is subject
to the celebration of a Protocol between Maritime
Administrations (the one who recognises and other
who issues the certificate).
In Portugal, IPTM Instituto Porturio e dos
Transportes Martimos is the responsible entity for
the compliance of the arrangements determined on
STCW 78/95 Convention. The process of recognition
shall be addressed, directed and requested directly to
IPTM, for analysis and decision.
IPTM commits itself to recognise certificates of
competency of seafarers whose nationality is as
follows:
a) European Union countries;
b) Countries with whom Portugal has an agreement
for recognition of certificates.
Countries under paragraph b) are subject to prior
verification of related system on education, training
680
and certification, considering a celebration of a

Protocol.
According to their level, certificates of competency
can be issued for the performance of management
or operational functions. In the first case theres an
obligation for demonstration of knowledge on basic
principles of the Portuguese Maritime Legislation.
REGISTRATION OF COMMERCIAL AND

PLEASURE YACHTS
4.1
Formalities
MAR accepts the registration of the following leisure

yachts:
3.5
Vessels registered and to be registered in MAR are

subject to the following fees:
For the initial registration or renewal thereof, the
following fees are due:
Commercial yachts with a hull length of between

7 and 50 meters and a maximum capacity of 12
passengers plus crew;
Pleasure yachts with a hull length above 7 meters.
Registration fees
Fixed fee of 1.800;

Variable fee:
Scale
Fee per scale
Up to 250 NT
From 250 NT up to 2,500 NT
From 2,500 NT up to 10,000 NT
Above 60,000 NT
225
0.90 per NT
0.75 per NT
0.60 per NT
0.50 per NT
0.40 per NT
0.30 per NT
0.20 per NT
0.10 per NT
It should be noted that pleasure yachts may only

be registered permanently in MAR. The application
form for the registration will be addressed to the Technical Commission of the register. For yachts with
commercial purposes, the application form shall be
followed by all the technical elements regarding safety
of the yacht, according to Portuguese legislation and
international regulations.
4.2 Fees
Depending on whether the yacht is to be used for
leisure or for chartering activities, the following fees
will apply:
NT = Net tonnage
The annual fee shall be calculated according to the

following scale:
Fixed fee in the amount of 1.400 Euros;

Variable fee:
Scale
Fee per scale
Up to 250 NT
From 250 NT up to 2,500 NT
Above 20,000 NT
200
0.80 per NT
0.40 per NT
0.25 per NT
Pleasure yachts, registered for leisure purposes, are

subject to:
An initial fee due for the registration, in the
amount of 500 Euros;
An annual fee payable to maintain the registration:
For pleasure yachts between 7 and 24 meters
long, a fee of 500 Euros;
For pleasure yachts more than 24 meters long,
a fee of 500 Euros plus 2 Euros per gross ton.
Pleasure yachts, registered for commercial purposes, are subject to:
For the initial registration or registration renewal:
Fixed fee in the amount of 1.250 Euros
Variable fee:
Scale
Fee per scale
Up to 250 GT
Above 250 GT
200
0.75 per GT
NT = Net tonnage
GT: Gross Tonnage
Ships transferred from the Portuguese traditional Registry are subject to different fees.
The fee due for passenger vessels, tug boats and
other auxiliary embarkations are those stated above,
adding thereto 15% to the initial registration or renewal
fees and 30% to the annual fee.
Other fees will be applicable on the issue, recognition or renewal by the Technical Commission of MAR
of various documents, declarations and certificates, as
well as on the undertaking of inspections on board
vessels registered in MAR.
681
An annual fee payable to maintain the registration:

Fixed fee of 1.000 Euros;
Variable fee:
Scale
Fee per scale
Up to 250 GT
Above 250 GT
200
0.75 per GT
GT: Gross Tonnage
Yachts owned by entities licensed within the legal

scope of Madeiras Free Trade Zone are entitled to
benefit from an exemption from the initial fee for
registration and to a 20% reduction in the annual fee.
Other fees will be applicable on the issue,
recognition or renewal by the Technical Commission
of MAR of various documents, declarations and certificates, as well as on the undertaking of inspections
on board vessels registered in MAR.
5
CONCLUSIONS
MAR is a high standard quality register with appropriate measures to ensure an efficient interface with all
vessels registered. All International Conventions ratified by Portugal are complied with by the International
Shipping Register. All well reputed Classification
Societies acknowledged by Portugal are entitled to
undertake surveys and other services in respect of
vessel registration. Registrations in MAR benefit
from the Private Deeds Registry Office in Madeira.

This assures a fast process of registration without
bureaucratic delays. Entities within the legal framework of the International Business Centre of Madeira
and whose business is that of maritime industry shall
for all effects and purposes benefit entirely from all
the tax and financial incentives.
MAR, with around 150 commercial vessels registered (with an average age of 16,5 years) and more than
85 commercial and pleasure yachts registered, continues to work to be recognized at international level as
a competitive and reliable actor in the shipping global
economy.
REFERENCES
Ship Registration in Madeira document downloaded from
www.ibc-madeira.com on 23 March 2011
Ship Registration Brochure document downloaded from
www.ibc-madeira.com on 23 March 2011
682

Author index
Abreu Valente, J. 631

Adak, M. 253, 285
Afonso-Dias, M. 19
Alarcin, F. 115
Alves Francisco, R.M. 3
Amdahl, J. 375
Antunes, P. 97
Anto, P. 425, 433
Azevedo, J. 261
Bagbanci, H. 269, 547, 553
Baptista, L. 631
Batista, L. 97
Baykut, N. 269
Bento, A.R. 563
Bernardino, M. 467, 513
Bettencourt, J.H. 45
Bhattacharjee, J. 571, 585, 655
Campos, A. 19
Campos, R.M. 473
Carichas, E. 631
Carvalho, J.P. 9
Carvalho, T. 277
Cerveira, F. 53
Chen, B.Q. 285, 293
Correia Guerreiro, R.M. 3
Correia Rodrigues, A. 277
Correia, F. 141
Cruz, J. 631
Gonalves de Deus, R.P. 3

Gonalves, M. 481, 525
Gotovac, H. 497
Guedes Soares, C. 9, 35, 45,
87, 133, 153, 165, 173, 181,
189, 205, 213, 223, 231, 241,
253, 269, 285, 293, 301, 309,
317, 323, 331, 349, 353, 363,
375, 381, 393, 399, 407, 415,
425, 433, 441, 451, 459, 473,
481, 487, 497, 513, 525, 533,
547, 553, 563, 571, 585, 613,
647, 655
Gulez, K. 115
Oktem, A.S. 269, 317, 323
Henriques, V. 19
Hilrio, A. 669
Huang, W. 301
Rajendran, S. 223
Ramos, A.A. 63
Requeijo, J.G. 127
Ribeiro e Silva, S. 97, 231
Ribeiro, D. 669
Robledo de Miguel, F. 607
Rodrigues, J.M. 153, 165
Roskilly, A.P. 121
Ruas, J.C. 79
Rusu, E. 481, 525, 613
Rusu, L. 467, 533
Rutgersson, A. 539
Karmakar, D. 181, 547, 553

Lampreia, S.S. 127
Leal, M. 631
Liu, B. 309, 393
Lobo, V. 73, 147
Lpez Pea, F. 579
Lpez Pieiro, A. 607
Lucas, C. 487
Luo, H. 189, 349, 399
Farias, P. 579
Fernandes, L. 505
Ferrari, V. 173
Ferreo, S. 579
Fonseca, N. 45, 53, 195, 213,
223, 595, 631
Fonseca, P. 19
Fortes, C.J. 505
Malta, E.B. 63
Mantari, J.L. 317, 323
Martinho, P. 563, 647
Martins, M.R. 433
Martins, P.T. 73, 147
Martins, R.F. 339
Mateus, A.R. 339
Matos, A.S. 127
Mendes, B. 19
Mesquita, P. 141
Mikalsen, R. 121
Monarcha, A. 195, 595
Moreira, L. 173, 205
Mota, E. 277
Moura, A. 73
Munn, A. 579
Muraleedharan, G. 487
Garbatov, Y. 293, 301, 331, 353,

363, 451
Gomes Antunes, J.M. 121
Natacci, F.B. 433

Nogueira, M.C. 63
Nuez Rivas, L.R. 607
Datta, R. 165
Dias, J.M. 127
Daz-Casas, V. 579
Dong, S. 459
Duarte, F. 631
Ekinci, S. 115
683
Padhy, C.P. 25
Parente, J. 19
Pao, A. 213
Perera, L.P. 9, 133, 153
Pessoa, J. 631
Petrov, V. 497
Pietra, L. 585
Pilar, P. 533
Pilar-Fonseca, T. 19
Pinheiro, L. 505
Queirs, J. 513
Saad-Eldeen, S. 331
Salvao, N. 467
Sampaio, C.M. 63
Santana, N. 623
Santos, F.P. 441
Santos, J. 141
Santos, J.A. 505
Santos, M.S. 339
Santos, T.A. 631
Semedo, A. 539
Sen, D. 25
Sequeira, M. 677
Silva, C.A. 87
Silva, D. 647
Silva, J.E. 451
Silva, P.P. 339
Silveira, P. 35
Somolinos Sanchez, J.A. 607
Sterl, A. 539
timac, G. 349
Sutherland, L. 53
Sutulo, S. 173, 241
Suelj, K. 539
Teixeira, A.P. 35, 441

Tekgoz, M. 353
Tello, M. 231, 585, 655
Tran Nguyen, K. 363
Varela, J.M. 107

Ventura, M. 79, 107
Villavicencio, R. 375, 381,
393
Vairinhos, V. 127
Valente, J.A. 97
Walkley, M. 505
Wang, S. 399
684
Xu, M. 407, 415

Zhang, L. 459
Zhou, X. 241

Soares C G Et Al Maritime Engineering and Technology

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Soares C G Et Al Maritime Engineering and Technology

Uploaded by

Copyright:

Available Formats

MARITIME ENGINEERING AND TECHNOLOGY

PROCEEDINGS OF MARTECH 2011, 1st INTERNATIONAL CONFERENCE ON MARITIME

Maritime Engineering and

ISBN: 978-0-415-62146-5 (Hbk + CD-ROM)

Maritime Engineering and Technology Guedes Soares et al.

Intelligent guidance in collision avoidance of maritime transportation

Analysis of maritime traffic off the coast of Portugal

Considering disabled people in sailing yacht design

Hull shape design of a compressed natural gas ship

Short sea shipping route and stowage plan optimization

Non-linear optimization applied to preliminary ship design

Planning a fleet of containerships for a given set of ports

Generation of an adaptive triangular mesh from a parametric surface

Ship propulsion & control

Conditioned maintenance predictive analysis in maritime propulsion engines

A nonlinear control approach for a vessel steering system

Bridge officer decision support tool

Ship dynamics and hydrodynamics

Influence of sea currents on manoeuvring of a surface autonomous model

Interaction of gravity waves with moored flexible floating membrane

Review of model test techniques of local slamming on ships

Wave frequency dynamics of a tanker moored inside the port

Ship hydrodynamic interaction forces in restricted waters

Analysis of the fluid-structure interaction of a composite motor yacht

Numerical investigations to study the effect of weld parameters on the temperature-time

Mechanical properties assessment of specimen subjected to random non-uniform general

Fatigue damage assessment of stiffener-frame structure

Study on ultimate torsional strength of ship hulls

FE parameters estimation and analysis of ultimate strength of a box girder

Structural behaviour of a lightweight craft

Strength of a tanker double bottom in various grounding scenarios

Review of material relations for marine structures collision simulations

Response of stiffeners with attached plate subjected to lateral impact

Explicit FE simulation of slamming load on rigid wedges with various deadrise

Risk and reliability

Modeling the risk of product spills in LNG tankers

Production regularity assessment using stochastic Petri nets with predicates

Vertical breakwater reliability analysis with direct integration method

Wind & wave Modelling

Modelling the wave conditions in the Archipelago of Azores

The global wave age climate

Review of offshore floating wind turbines concepts

Modelling wave energy for the north coast of Spain

Computational simulation of a buoy behaviour for automation and power generation in

Conceptual design of an ocean current turbine for deep waters

Assessment of the wave energy in two enclosed seas

Multi-purpose Offshore Platform (MpOP)

Modeling wave energy for the Portuguese coast

Dynamics and hydrodynamics of ship like wave energy converters in roll

MAR International Shipping Register of Madeira

Maritime Engineering and Technology Guedes Soares et al.

Maritime Engineering and Technology Guedes Soares et al.

Associao das Indstrias Navais

Maritime Engineering and Technology Guedes Soares et al.

Maritime situational awareness indicators based on AIS and

R.M. Correia Guerreiro

R.M. Alves Francisco

understanding of anything associated with the global

Maritime dimension of Portugal

Portugal has a territory with approximately

2.1 AIS data

1.2 Maritime Situational Awareness and scope of

related to the Search and Rescue Service which is