VIRTUE BPG Integrated Final PDF

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/275032174
Best Practice Guidelines for the application of Computational Fluid

Dynamics in Marine Hydrodynamics
Technical Report · April 2009

DOI: 10.13140/RG.2.1.4642.7680
CITATIONS READS
3 1,145
15 authors, including:
Richard Marcer Christian Berhault

Principia - La Ciotat France Ecole Centrale de Nantes
83 PUBLICATIONS 350 CITATIONS 30 PUBLICATIONS 41 CITATIONS
SEE PROFILE SEE PROFILE
Luís Eça Leif Broberg

Technical University of Lisbon 7 PUBLICATIONS 18 CITATIONS
86 PUBLICATIONS 749 CITATIONS
SEE PROFILE
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
corrosion in shipbuilding View project
Tsunamis in the Atlantic and the English ChaNnel Definition of the Effects through numerical Modeling View project
All content following this page was uploaded by Richard Marcer on 16 April 2015.
The user has requested enhancement of the downloaded file.

BEST PRACTICE GUIDELINES
FOR THE APPLICATION OF
COMPUTATIONAL FLUID DYNAMICS
IN MARINE HYDRODYNAMICS
Prepared by
ATKINS
&
VIRTUE members
VIRTUE – The Virtual Tank Utility in Europe

Best Practice Guidelines
for the application of Computational Fluid Dynamics
in Marine hydrodynamics
Acknowledgements
First, we would like to thank all the different VIRTUE Work Package members for their
significant effort in contributing and improving Best Practice Guidelines for the application of
Computational Fluid Dynamics in marine hydrodynamics such as Resistance, Sea-keeping,
Manoeuvring and Cavitation.
Second, we are very grateful to the Work Package Best Practise leaders for their coordination
and enormous contribution to the elaboration of this integrated document, namely:
 Dr. Ir. Hoyte C. Raven from MARIN, Dr. Luís Eça from IST, Dr. Leif Broberg from
FLOWTECH and Dr. Carl-Erik Janson from Chalmers University on Resistance
applications.
 Ir. Christian Berhault, Dr. Richard Marcer, Dr. Christine De Jouette from PRINCIPIA
and Dr. Paul Gallagher from Atkins on Sea keeping applications.
 Ir. QiuXin GAO from the University of Strathclyde, Ir. Serge Toxopeus from MARIN
and Dr. Bertrand Alessandrini from Ecole Centrale Nantes on Manoeuvring
applications.
 Prof. Dr. Ir. Tom van Terwisga from MARIN, Dr. Martin Hoekstra from MARIN, Ir.
Heinrich Streckwall from HSVA and Dr. Francesco Salvatore from INSEAN on
Propeller and Cavitation applications.
We would also like to thank Ir. Antoine Pages from Sirehna for providing a very good
overview on optimisation techniques.
Third, we would like to acknowledge authors of MARNET-CFD, which provided a first Best
Practise Guidelines document for the application of Computational Fluid Dynamics in Marine
hydrodynamics. The MARNET-CFD have served as a useful starting point to form the present
document by updating and further enhancing Best Practices with recent exploitable results
from the VIRTUE project.
Finally, we would like to address a special thanks to Prof. Dr. Ir. Charles Hirsch and reviewers
of the European commission who challenged members of the VIRTUE project on this
deliverable.
Ir. Benoît Post from Atkins.
Issue Date: 28 April 2009 Page 1

Glossary
AP The aft perpendicular, AP, is defined as a vertical line passing through the
rear face of a rudder post of the ship or the centre of a rudder stock of the
ship having no rudder post
CFD Computational Fluid Dynamics
CFL The Courant Friedrichs Lewy number for incompressible flow is defined as
CFL= U.t / x, where t is the time step, x the local cell size and U the
local velocity
DLWL Designed Load Water Line
DNS Direct Numerical Simulation
DOE Design Of Experiment
EFFORT European funded project on Full-Scale Flow Research and Technology
focused on validating and introducing innovative Computational Fluid
Dynamics (CFD) prediction methods for the performance of the ship/propeller
combination at full scale, instead of the usual model scale
ERCOFTAC ERCOFTAC is a scientific association of research, education and industry
groups in the technology of flow, turbulence and combustion
EXPRO-CFD European funded project aiming at making substantial improvements to
techniques for the prediction of non-linear and extreme wave loads and
responses of offshore floating production systems
FP The fore perpendicular, FP, is defined as a vertical line passing through a
point at which the designed load waterline, DLWL, crosses the front face of
the bow
GM Meta-centric height
QNET-CFD European funded project aiming at improving the quality of industrial CFD
calculations, and increasing the level of trust in the results obtained by CFD.
Lpp It represents the length between perpendiculars, i.e., the horizontal distance
from the fore perpendicular, FP, of the ship to the aft perpendicular, AP, of
the ship
LES Large Eddy Simulation
MARNET-CFD European funded project aiming at covering all matters relating to the
development and exploitation of Computational Fluid Dynamics (CFD) in the
European maritime and offshore industries.
PMM Planar Motion Mechanism
RANSE Reynolds Averaged Navier Stokes Equations
RSM Reynolds Stress Model
SPH Smoothed Particle Hydrodynamics
SWENSE Spectral Wave Explicit Navier Stokes Equations
VIRTUE European funded project “The Virtual Towing Tank Utility in Europe” aiming
at creating virtual basin by improving available CFD tools and integrating
them in a comprehensive simulation environment of ship behaviour at sea.
6 DOF 6 Degrees Of Freedom

Table of Contents
1. INTRODUCTION .................................................................................... 7
1.1. Background ................................................................................................................... 7
1.2. Scope ............................................................................................................................. 7
1.3. Notes for readers .......................................................................................................... 8
2. MODELLING THEORY FOR MARINE APPLICATIONS ....................... 9

2.1. Fluid equations of motion ............................................................................................ 9
2.1.1. General Fluid Dynamic Equations ........................................................................... 9
2.1.2. Incompressible and Isothermal assumption .......................................................... 10
2.2. Inviscid and irrotational hypotheses ........................................................................ 11
2.2.1. Potential Flow Equations ....................................................................................... 11
2.2.2. Steady ship flow application .................................................................................. 12
2.2.3. Free surface modelling .......................................................................................... 14
2.2.4. Integration of viscous effects ................................................................................. 15
2.3. Turbulence Modelling ................................................................................................. 15
2.3.1. RANSE .................................................................................................................. 17
2.3.2. LES ........................................................................................................................ 22
2.3.3. Hybrid RANSE/LES ............................................................................................... 23
2.3.4. Near wall treatment ............................................................................................... 23
2.4. Free surface Modelling............................................................................................... 25
2.4.1. Interface tracking ................................................................................................... 25
2.4.2. Interface capturing ................................................................................................. 25
2.4.3. SPH ....................................................................................................................... 26
2.5. Wave tank modelling .................................................................................................. 27
2.5.1. Wave maker .......................................................................................................... 27
2.5.2. SWENSE ............................................................................................................... 31
2.6. Hydro-mechanical modelling..................................................................................... 36
2.6.1. Introduction ............................................................................................................ 36
2.6.2. 6 DOF .................................................................................................................... 36
2.6.3. 6 DOF for ship manoeuvring ................................................................................. 36
2.7. Cavitation Modelling ................................................................................................... 39
2.7.1. Introduction ............................................................................................................ 39
2.7.2. Boundary Element Methods .................................................................................. 39
2.7.3. Field Methods ........................................................................................................ 40
3. COMPUTATIONAL THEORY............................................................... 44
3.1. Potential Flow Theory ................................................................................................ 44
3.1.1. Panel mesh generation.......................................................................................... 44
3.1.2. Boundary Conditions definition .............................................................................. 44
3.1.3. Non-linear methods ............................................................................................... 45
3.2. Viscous Computational Fluid Dynamics .................................................................. 46
3.2.1. Solver methodology ............................................................................................... 46

3.2.2. Meshing ................................................................................................................. 47

3.2.3. Boundary conditions .............................................................................................. 49
3.2.4. Numerical Resolution ............................................................................................ 49
3.2.5. Post processing ..................................................................................................... 52
3.2.6. Sensitivity studies .................................................................................................. 52
3.2.7. Dealing with uncertainties ..................................................................................... 52
3.3. Optimisation techniques ............................................................................................ 53
3.3.1. Data analysis - Design exploration - Design of experiments ................................ 53
3.3.2. Data modelling - Response surfaces .................................................................... 54
3.3.3. Optimisation ........................................................................................................... 55
3.3.4. Decision making .................................................................................................... 56
3.3.5. Robust design ........................................................................................................ 57
3.3.6. Hybridation ............................................................................................................ 58
3.3.7. Time demanding modelling and analysis .............................................................. 58
4. APPLICATION 1: NUMERICAL TOWING TANK ................................ 62

4.1. Introduction ................................................................................................................. 62
4.2. Work-package Tasks .................................................................................................. 63
4.2.1. The Numerical Resistance Test ............................................................................ 63
4.2.2. Hull/Propeller Interaction – The numerical propulsion tests .................................. 63
4.2.3. Optimisation based on RANSE solvers ................................................................. 64
4.3. Application cases ....................................................................................................... 64
4.3.1. Case 1-a: Resistance prediction in double-body flow ........................................... 64
4.3.2. Case 1-b: Resistance prediction with free surface (free sinkage and trim) ........... 71
4.4. Prediction of Resistance, wake and wave pattern .................................................. 78
4.4.1. Introduction ............................................................................................................ 78
4.4.2. Target Flow Variables............................................................................................ 79
4.4.3. Double-model ship hull .......................................................................................... 80
4.4.4. Free surface .......................................................................................................... 86
4.5. Hull / Propeller Interaction ......................................................................................... 91
4.5.1. Introduction ............................................................................................................ 91
4.5.2. Target flow variables ............................................................................................. 91
4.5.3. Selection of mathematical model .......................................................................... 92
4.5.4. Accuracy of modelling ........................................................................................... 95
4.5.5. Error sources ......................................................................................................... 96
4.6. Optimisation of ship hulls .......................................................................................... 97
4.6.1. Hull deformation scheme and design variables .................................................... 97
4.6.2. Objective functions ................................................................................................ 98
4.6.3. Constraints ............................................................................................................ 99
4.6.4. Optimisation algorithm ......................................................................................... 100
4.6.5. Investigation of the design space ........................................................................ 100
4.6.6. Automatic optimisation ........................................................................................ 101
5. APPLICATION 2: NUMERICAL SEA KEEPING TANK ..................... 102

5.1. Theoretical aspects of sea keeping ........................................................................ 102
5.1.1. General theory of sea keeping ............................................................................ 102
5.1.2. CFD approach for viscous flow as a complementary tool for sea keeping ......... 104
5.2. How to use CFD jointly with standard sea-keeping codes................................... 109
5.2.1. Viscous damping in ship motions ........................................................................ 109
5.2.2. Ship motions in wave ........................................................................................... 111

5.2.3. Influence of Sloshing, Compartment flooded and Water on deck ....................... 111
5.2.4. Slamming loads prediction .................................................................................. 112
5.3. Guidelines for numerical sea-keeping applications ............................................. 114
5.3.1. Definition of target variables ................................................................................ 114
5.3.2. Selection of adequate mathematical model for the flow physics ........................ 115
5.3.3. Selection of computational domain, adequate grid topology and spacing .......... 116
5.3.4. Selection of turbulence model ............................................................................. 117
5.3.5. Selection of boundary conditions and initial conditions ....................................... 118
5.3.6. Selection of time step/transient flow/simulation duration .................................... 118
5.3.7. Selection of method for free surface prediction ................................................... 119
5.3.8. Selection of wave models .................................................................................... 119
5.3.9. Selection of Discretization schemes.................................................................... 120
5.3.10. Selection of iterative convergence criteria ....................................................... 122
5.3.11. Post-processing ............................................................................................... 122
5.3.12. Estimation of numerical accuracy of the solution ............................................ 124
5.4. Application Cases ..................................................................................................... 124
5.4.1. Case 2-a: HTC hull with forward speed in roll motion ......................................... 124
5.4.2. Case 2-b: Wave propagation w/o structure ......................................................... 127
5.4.3. Case 2-c: Calm buoy at zero-speed .................................................................... 133
5.4.4. Case 2-d: LNG Carrier at zero-speed ................................................................. 138
5.4.5. Case 2-e: HTC with forward speed in waves ...................................................... 142
6. APPLICATION 3: NUMERICAL MANOEUVRING TANK .................. 144

6.1. Theoretical aspects of Manoeuvring ...................................................................... 144
6.1.1. Introduction .......................................................................................................... 144
6.1.2. The approaches of manoeuvring model tests ..................................................... 144
6.1.3. Virtue manoeuvring towing tank as a complementary tool .................................. 146
6.1.4. Coordinate system and nomenclature................................................................. 146
6.1.5. Equations of motion ............................................................................................. 146
6.2. Applications Cases: Derivative Approach ............................................................. 148
6.2.1. Introduction .......................................................................................................... 148
6.2.2. Grid ...................................................................................................................... 148
6.2.3. Boundary conditions ............................................................................................ 149
6.2.4. Case 3-a: Steady oblique motion ........................................................................ 150
6.2.5. Case 3-b: Steady turning motion ......................................................................... 152
6.2.6. Obtaining the hydrodynamic coefficients............................................................. 154
6.2.7. Other available cases suitable for application and validation ............................. 159
6.2.8. Case 3-c: PMM Simulations - Unsteady forced motion ....................................... 160
6.3. Application Cases: Direct numerical manoeuvring simulation ........................... 170
6.3.1. Case 3-d: Unsteady manoeuvring ship on calm water ........................................ 170
6.3.2. Case 3-e: Unsteady manoeuvring ship in Waves ............................................... 175
6.4. Guidelines for numerical manoeuvring applications ............................................ 180
6.4.1. Definition of target variables ................................................................................ 180
6.4.2. Selection of adequate mathematical model for the flow physics ........................ 180
6.4.3. Selection of computational domain ..................................................................... 181
6.4.4. Selection of boundary conditions and initial conditions ....................................... 181
6.4.5. Selection of adequate grid topology/spacing and time step ................................ 181
6.4.6. Selection of discretisation schemes .................................................................... 182
6.4.7. Selection of iterative convergence criteria........................................................... 182
6.4.8. Post-processing ................................................................................................... 182
6.4.9. Estimation of numerical accuracy of the solution ................................................ 182
6.4.10. How to introduce CFD derivatives in manoeuvring simulator .......................... 183

7. APPLICATION 4: NUMERICAL CAVITATION TANK ....................... 184

7.1. Special issues relating to propeller flow modelling .............................................. 184
7.2. Guidelines for predicting cavitating flows around 2D and 3D stationary foils .. 185
7.2.1. Type of simulation ............................................................................................... 185
7.2.2. Domain size and meshing ................................................................................... 186
7.2.3. Boundary conditions ............................................................................................ 187
7.2.4. Turbulence modelling .......................................................................................... 188
7.2.5. Numerical issues ................................................................................................. 188
7.3. Guidelines for predicting propeller flows including cavitation ........................... 189
7.3.1. Definition and planning phase ............................................................................. 189
7.3.2. Type of simulation ............................................................................................... 190
7.3.3. Domain size ......................................................................................................... 192
7.3.4. Meshing ............................................................................................................... 194
7.3.5. Boundary conditions and initial conditions .......................................................... 196
7.3.6. Controlling the calculation ................................................................................... 198
7.3.7. Post-processing ................................................................................................... 199
7.4. Application Cases ..................................................................................................... 200
7.4.1. Case 4-a: 2D Foil ................................................................................................. 201
7.4.2. Case 4-b: 3D twisted foil...................................................................................... 209
7.4.3. Case 4-c: The INSEAN E779A propeller............................................................. 213
8. CONCLUSION .................................................................................... 224
9. REFERENCES ................................................................................... 225

1. Introduction
1.1. Background
The availability of robust commercial Computational Fluid Dynamics (CFD) software and the
rapid growth in processing power have lead to an increasing use of CFD for the solutions of
fluid engineering problems across all industrial sectors. The marine industry is no exception:
computational methods are now routinely used, for example, to examine vessel boundary
layer and wake, to predict propeller performance and to evaluate structural loads.
There has been a growing awareness that computational methods can prove difficult to apply
reliably i.e. with a known level of accuracy. This is in part due to CFD being a knowledge-
based activity and, despite the availability of the computational software; the knowledge base
embodied in the expert user is not available. This has lead to a number of initiatives that have
sought to structure existing knowledge in the form of best practice advice. Few notable
examples are the best practice guidelines developed by ERCOFTAC, the European Thematic
network QNET-CFD and of course the best practice guideline applied to the marine
applications, MARNET-CFD.
The guidelines presented here build on the work of these initiatives and particularly update
the MARNET-CFD Best Practise Guidelines with recent results and conclusions obtained
during the VIRTUE project – The Virtual Tank Utility in Europe.
The guidelines provide simple practical advice on the application of computational methods in
hydrodynamics within the marine industry. It covers both potential and viscous flow
calculations.
1.2. Scope
This document provides some background on the modelling theory used for hydrodynamics
simulation in Chapter 2 - Modelling theory for Marine applications. It details equations, validity
domain, assumptions, strengths and weaknesses of the different mathematical models used
for representing the different physics involved in marine application.
The numerical approaches used for solving the above models are detailed in Chapter 3 -
Computational Theory. This covers Potential Flow, Computational Fluid Dynamics and
Optimising tools.
Having given the reader a comprehensive overview of the modelling and computational
theory in hydrodynamics, the next chapters will be focused on how these models and
numerical methods compare on some practical applications and how we can come up with
some specific guidelines for each of these hydrodynamics applications, namely:
 Chapter 4 - Application 1: Numerical Towing tank
 Chapter 5 - Application 2: Numerical Sea keeping tank
 Chapter 6 - Application 3: Numerical Manoeuvring tank
 Chapter 7 - Application 4: Numerical Cavitation tank
Underlying Flow regime and application challenges are illustrated in each hydrodynamics
application chapter in order to better understand the usage, the benefits and the challenges of
applying Computational Fluid Dynamics in the maritime industry.

1.3. Notes for readers

A specialist in Computational Fluid Dynamics will focus its reading mainly on the application
cases and how to apply its CFD knowledge to the Maritime applications. The theory can bring
him the necessary background in Potential Flow (applications, benefits and limitations), in
models encountered more regularly in the marine hydrodynamics (free-surface, wave tank,
hydro-mechanical and cavitation modelling) and update him on the numerical methods
(including optimisation).
A naval architect or marine industry specialist will need more in depth reading and learning
from the modelling and computational theory before going through the application cases.
Both types of readers should enjoy the different marine application cases, which form the core
of this document, where the usage of Computational Fluid Dynamics is illustrated, validated
against experiments, compared against Potential Flow methods and sometimes challenged.

2. Modelling theory for Marine applications

In marine fluid dynamics we are chiefly concerned with problems in hydrodynamics. In the
majority of problems being solved, we are attempting to calculate global pressures and fluid
velocity components in a 3 dimensional space as a function of time surrounding the
submerged portion of the marine vehicle or platform of interest. In this way, it is possible to
further calculate the forces and moments acting on the vessel, whether steady or unsteady.
In this chapter, we will concentrate on describing the mathematical models (assumptions,
equations, strengths and weaknesses) used to represent the different physics occurring in
Marine Fluid dynamics.
Therefore, we start from the beginning and provide definitions of the general fluid equations of
motion, from which incompressible and isothermal equations can be derived. It is customary
to treat the working fluid, in this case water, as incompressible and isothermal.
From this set of equations together with inviscid and irrotational assumption, we develop the
basis of the Potential Flow Theory and the way the free surface is modelled.
By considering the viscous effects, we introduce the Navier Stokes turbulent equations used
in Computational Fluid Dynamics. In particular, we further develop turbulence modelling
(RANSE, DES, LES), near wall treatment, free surface modelling, wave tank modelling, hydro
mechanical modelling and cavitation modelling.
2.1. Fluid equations of motion
2.1.1. General Fluid Dynamic Equations

The general equations of fluid flow represent mathematical statements of the conservation
laws of physics, such that:
 Fluid mass is conserved
 The rate of change of momentum equals the sum of the forces on a fluid particle
 The rate of change of energy is equal to the sum of the rate of heat addition to and the
rate of work done on a particle.
The governing equations for an unsteady, three dimensional, compressible viscous flows are:
Continuity equation:

   ( U )  0 (1)
t
Momentum equations:
( u ) p   yx  zx
   ( uU )    xx    f x (2)
t x x y z
( v) p  xy  yy  zy
   ( vU )       f y (3)
t y x y z
( w) p  xz  yz  zz
   ( wU )       f z (4)
t w x y z

Energy equation:
 U2   U2   T  T  T
  (e  )      U (e  )   q  (k )  (k )  (k )
t  2   2  x x y y z z
 (up)  (vp)  ( wp )  (u xx )  (u yx )  (u zx )
     
x y z x y z
(5)
 (u xy )  (u yy )  (u zy )  (u xz )  (u yz )  (u zz )
       fU
x y z x y z
Where:  is the fluid density, U = (u, v, w) the fluid velocity, p the pressure, T the temperature,
e is the internal energy per unit mass, f = (fx, fy, fz) is a body force, k is the thermal
conductivity, q is the rate of volumetric heat addition per unit mass and nn are the viscous
stresses.
These equations represent 5 transport equations in 7 unknowns, u, v, w, p, T,  and e. They
are completed by adding two algebraic equations; one relating density to temperature and
pressure:
   (T , p) (6)
and the other, relating static enthalpy to temperature and pressure:
h  h(T , p) (7)
2.1.2. Incompressible and Isothermal assumption

For hydrodynamic applications, incompressible and isothermal flow is assumed. Furthermore,
it is assumed that the fluid is Newtonian and that the laminar viscosity is constant throughout
the flow.
The continuity equation becomes:
U
. 0 (8)
The momentum equations become:

Du p
     2 u    f x (9)
Dt x
Dv p
     2 v    f y (10)
Dt y
Dw p
     2 w    f z (11)
Dt z
Where D/Dt is the substantial derivative given by:
D    
 u v  w (12)
Dt t x y z

2.2. Inviscid and irrotational hypotheses

Potential flow models remain the most important tool for studying offshore structures and
remain the most reliable approach to wave resistance. They also provide the basis for the
majority of propeller design methods. They all employ boundary integral formulations of
various kinds and are therefore quite computationally efficient and, up until now, have offered
the simplest approach to the modelling of free surface and propeller flows.
2.2.1. Potential Flow Equations

By making further simplifications to the previously described equations in 2.1.2, that is to say
by considering the flow inviscid and irrotational, the equations can be reduced to a single
scalar quantity, the fluid potential  , in order to describe the flow.
The kernel of potential flow theory is the definition of a velocity potential such that:
  
u v w (13)
x y z
The continuity equation becomes a Laplace equation:

 2   2 x   2 y 2   2 z 2  0
2
(14)
which is sufficient to determine the complete velocity field.
The momentum equations reduce to the statement that fluid acceleration is directly related to
the fluid pressure gradient:
Du p
  (15)
Dt x
Dv p
  (16)
Dt y
Dw p
  (17)
Dt z
As the Laplace equation is linear it is possible to combine elementary solutions, such as

sources, sinks, doublets and vortices for application to complex solutions. It should also be
appreciated that potential flow is the governing behaviour of gravitationally driven wave
systems, and hence represents the fundamental physics, with appropriate boundary
conditions, for free surface wave problems.
For potential flows, it is possible to derive a simple expression for the fluid pressure by
integrating the Navier Stokes equations along a stream-line to give the well known Bernoulli
equation.
Potential flows, and their characterisation using the Laplace equation, have many important
and useful properties that can be used in the formulation of numerical solutions. The use of
the Divergence Theorem by Gauss (to convert volume to surface integrals), Green’s theorem
(to convert a surface integral to a line integral), and the principle of superposition of solutions,
all provide the means to formulate boundary element solution methods. Various forms of
boundary element or panel methods have, thus far, been the principal means by which these
flows have been modelled.

2.2.2. Steady ship flow application

As an illustration of how these methods are developed, the particular case of the steady ship
flow problem with the boundary integral formulation of its solution is described.
As noted earlier, the domain over which the flow solution is required is bounded by the wetted
hull surface, the free surface, the sea bed (if sufficiently close), and a so-called far-field
boundary.
If the free surface height can be represented by z = (x, y, t), the flow field is then evaluated
by solving the Laplace equation everywhere for z < (x, y, t):
 2  0 (18)
Where u   is used to derive the flow velocity components
The following boundary conditions are formulated throughout the domain:

Kinematic boundary conditions: Water does not penetrate the free surface or the body
surface.
Dynamic free-surface boundary condition: Atmospheric pressure acts at the water surface,
which is considered to contain all surface streamlines. This allows the use of the Bernoulli
equation in the formulation of a condition for the unsteady potential in combination with the
kinematic condition mentioned above.
Radiation or far field boundary conditions: Which depend on the type of analysis undertaken,
but can be summarised as allowing the propagation of waves in the far field which satisfy the
need for consistency in the transport of energy away from the disturbance. For linear wave
resistance or radiation / diffraction problems, these conditions are implicit in the choice of
Green’s function (see below). For non-linear time domain or field methods, the computational
domain is truncated at some distance from the vessel and appropriate numerical models that
satisfy the required properties are applied.
The steady kinematic condition on the water surface z =  can be written:
     z (19)
The steady dynamic condition at z =  is:

1 1
gz  ( ) 2  U 2 (20)
2 2
The non-linear free surface boundary condition for is formed by combining the kinematic and
dynamic boundary conditions:
1
  ( ) 2  g z  0 (21)
2
It can be assumed that the total potential is made up of a free-stream potential and a smaller
perturbation potential. The linearised Kelvin free-surface boundary condition at the
undisturbed surface is then:
g
 xx  z (22)
U2

The wave resistance can be calculated from the energy in the free wave spectrum, or by
integration of the steady hull surface pressures arising from the solution.
In a higher order approach suggested by Dawson (1977), the total potential is divided into a
double-body potential and a perturbation potential. As the perturbation potential is small, the
double-body potential corresponds to the limiting solution as the Froude number goes to zero.
More advanced methods have since been developed, such as Raven (1996) and Janson
(1997) although Dawson’s method remains the basis of many computer codes. A fully non-
linear approach is also possible whereby both the kinematic and dynamic boundary
conditions are satisfied by an iteration procedure that has many similarities with time domain
simulation.
In order to solve for the fluid potential, the Laplace equation is transformed into a surface
integral taken over the ship hull as described earlier. This surface integral then provides the
means to develop a discrete set of integral equations by splitting the surface up into a number
of panels. One integral equation is then written for each panel in which the hull surface
boundary condition is satisfied locally.
Each panel is assumed to represent a fluid source, which may be a local point value, or may
be distributed in some pre-defined manner (i.e. constant strength per unit area, bi-linear
distribution, etc.).
Each panel source has an effect on every other panel source, contributing to the induced flow
over each panel surface. The influence of each panel on every other panel is represented by
a weighting function, which in classical hydrodynamics is known as a Green’s function.
The Rankine source is used for steady flow in an infinite fluid and is one of the simplest
functions used. It simply makes the assumption that the velocity potential induced at a field
point some distance from the source is inversely proportional to the distance between them.
This type of function is suitable for both linear and non-linear applications.
More complex functions are used for linear free surface flows with wave radiation. The main
classes of such functions are:
 The zero speed pulsating source Green’s function,
 The steady forward speed Green’s function,
 The translating, pulsating source Green’s function,
For linear problems, these functions provide weightings for the source potential which also
uniquely satisfy the linear free surface and far-field boundary conditions, removing the need to
distribute further sources over these regions, and greatly reducing computing times.
For each panel therefore, the normal fluid velocity induced over the panel surface is
calculated from the summation of all other source (or dipole) contributions weighted by a
Green’s function, and its own self- influence. This velocity is then equated to the boundary
condition at the panel surface, which requires no net flow through the panel. In the case of a
general steady flow, with a non-linear free surface boundary condition to be satisfied, and in
the presence of lifting surfaces, this is expressed in the pair of equations:
     B GB .dS     F GF .dS    W GW .dS (23)

SB SF SW
(U   )n  0 (on the hull) (24)
Where subscript B denotes the surface of the body, F the free surface and W a trailing wake
sheet comprising a dipole distribution, as applicable. Here  is the element source strength, 
the dipole strength, and G the corresponding Green’s function.

Equation (24) may be differentiated and substituted into (23) to allow the numerical
discretisation of the surface integral. This can then be expressed in matrix form as:
D   W   S U
ij ik ij   n (25)
where for panel j, Sij is the source influence coefficient of a unit strength panel, Dij is the
dipole influence coefficient and Wik is the influence of the constant strength wake strip
extending to infinity.
The overall solution is achieved by inversion of this matrix problem using standard
techniques, and from the subsequent recovery of potentials, source and dipole strengths.
2.2.3. Free surface modelling

The primary difficulty with free surface calculations is that the position and shape of the free
surface is not known, and often involves non-linear effects such as wave breaking and
fragmentation. In any case, wave diffraction and radiation effects can be substantial for many
marine structures with large dimensions.
Earlier, some background description of free surface flow boundary conditions was given,
dealing with both the linear frequency domain and the linearisedd steady ship wave problem.
In both cases, solutions are achievable using a suitable Green’s function on the hull which
explicitly satisfies the linear free surface boundary condition, thereby removing the need to
model further the behaviour of the free surface.
However, the full definition of the free surface boundary condition is both non-linear as a
mathematical statement and in its geometrical location. Recalling equations 2 and 3 given
earlier, and considering their unsteady or dynamic form, we get:
The unsteady kinematic condition,

  z     (26)
t
and the unsteady dynamic condition:
 1 1
 gz  ( ) 2  U 2 (27)
t 2 2
These coupled equations are posed on the free surface itself, and therefore to be considered
within a moving frame of reference. Note that the potential considered here is the total fluid
potential.
For non-linear, steady or unsteady problems solved using a panel method, one method of
approach is to simulate the evolution of a steady state or transient behaviour of the flow by
discretising the above equations in time as well as space. For example the first order time
derivative in the above equations can be replaced with a simple forward finite difference
expression or by more advanced Runge-Kutta like time marching schemes. In these cases,
the kinematic condition is used to update the location of the free-surface panels used to
describe the boundary at each time step, usually using the values of velocity potential and
surface elevations at the current step. The formulation of the boundary condition for the
potential using the unsteady dynamic condition is more complex. The simplest approach is
clearly to update the potential, and use this predicted value in the boundary integral
formulation. However, other methods which seek to couple the solution of the free surface
potential to that on the surface of the vessel directly have also been developed but are
beyond the scope of this document to describe in detail.
It should also be noted that these non-linear potential flow problems are still open to the use
of the principle of superposition of solutions. For sea-keeping or offshore engineering

problems, it is therefore common to assume a single frequency harmonic incident potential,

and to express the problem in terms of the non-linear perturbation or solution potential
accordingly.
2.2.4. Integration of viscous effects

Potential flow methods offer a convenient and relatively simple way of determining general
flow patterns and forces around arbitrary bodies. The main drawback is the inherent neglect
of viscous effects. In many applications where a potential flow method is used, such as in the
high Reynolds number regime for ship hulls, the viscous effects will be confined to thin
attached boundary layers, wakes and regions of free shear. For problems such as the
solution of roll damping, viscous effects will become large. There are a number of ways of
including some estimation of the viscous effects.
2.2.4.1. Empirical and semi-empirical methods

A number of empirical techniques have been developed to include the effects of the boundary
layer with regard to skin friction and the alteration to the pressure distribution around the
body. A fully empirical approximation to the skin friction drag can be used (e.g. the ITTC 1957
correlation line). In this case the local skin friction for each panel can be calculated based on
a parametric length from the leading edge or stagnation point. The form drag can be
approximated by assuming that the important region for viscous shear is confined to a narrow
domain next to the body and the trailing wake. The displacement thickness of the boundary
layer is calculated and the corresponding panels then displaced by this amount.
One problem with such formulations is that it is hard to prescribe the correct form of viscous
correction equations for effects such as stern wave systems, as this is highly sensitive to hull
form shape.
A second problem is that traditional ship correlation lines already contain some effects
associated with hull form. These techniques are no longer in frequent use.
2.2.4.2. Solution of Navier Stokes equations

An alternative to empirical techniques is to solve the Reynolds Average Navier Stokes
Equations (RANSE) in the domain surrounding the vessel and with the potential flow solution
used to define the shape and location of the free surface.
This is currently only considered suitable for steady ship flow problems. The free surface
boundary is treated as a free-slip wall, and viscous effects at the free surface assumed to be
negligible. The main benefit of the method us that it allows the full wetted surface of the
vessel to be included in the calculations.
The use of empirical formula to estimate additional viscous effects should be used as an
approximate method only, and care should be exercised in the choice of skin friction
correlation line. Such methods can only be applied where the flow remains attached. For
accurate resolution of stern wave and transom effects, where viscous forces are significant,
empirical viscous approximations may not be sufficient.
2.3. Turbulence Modelling

Whilst the equations in 2.1.2 are sufficient for the description of incompressible, laminar flow,
and being a description of a continuum, in principle apply to all scales, they are also non-

linear and subject to instability. Physically, these instabilities grow to provide a mechanism to
describe turbulence. Practically, this renders the equations impossible to solve analytically,
and requires that numerical methods be formulated to solve them.
Flow velocities and pressure exhibit random variations and the fluid flow contains a wide
range of turbulent eddy sizes as shown in Figure 2-1.
 The size/velocity of large eddies is the order of the mean flow and large eddies derive
energy from the mean flow
 Energy is transferred from larger eddies to smaller eddies and in the smallest eddies,
turbulent energy is converted to internal energy by viscous dissipation.
Figure 2-1: Turbulent Energy Cascade (Richardson, 1922)
Most flows of practical engineering interest are turbulent, and the turbulent mixing of the flow
then usually dominates the behaviour of the fluid. The turbulent nature of the flow plays a
crucial part in the determination of many relevant engineering parameters, such as frictional
drag, flow separation, thickness of boundary layers, extent of secondary flows, and spreading
of jets and wakes.
The turbulent states which can be encountered across the whole range of industrially relevant
flows are rich, complex and varied. After a century of intensive theoretical and experimental
research, it is now accepted that no single turbulence model can span these states and that
there is no generally valid universal model of turbulence. A bewildering number and variety of
models have appeared over the years, as different developers have tried to introduce
improvements to the models that are available. The extremely difficult nature of this
endeavour caused Bradshaw (1994) to refer to turbulence as "the invention of the Devil on
th
the 7 day of creation, when the Good Lord wasn't looking".
It is possible, in theory, to directly resolve the whole spectrum of turbulent scales using an
approach known as direct numerical simulation (DNS). No modelling is required in DNS.

However, DNS is not feasible for practical engineering problems involving high Reynolds
number flows. The cost required for DNS to resolve the entire range of scales is proportional
3
to Ret , where Ret is the turbulent Reynolds number. Clearly, for high Reynolds numbers, the
cost becomes prohibitive.
2.3.1. RANSE
In the Reynolds Averaged Navier Stokes Equations (RANSE), the fluid equations of motion
are time averaged and all turbulence length scales are modelled. This is the most widely used
approach for calculating industrial flows.
Ensemble (time averaging) is used to extract the mean flow variables from the instantaneous
flow variables. The components of the flow velocity and the pressure are thus decomposed
into a mean value with superimposed fluctuations.
U  U ( x)  U ( x, t ) (28)
Where:
U (x) is the mean
U ( x, t ) is the unsteady disturbance quantities in the flow, such that U   0
By inserting the above equation in the hydrodynamic equations described in 2.1.2 and then by
time averaging the equations, one can obtain the well known Reynolds Averaged Navier
Stokes Equations (RANSE).
After time averaging, the x-component momentum equation becomes:
 (u 2 ) (uv) (uw)  dP   u    u 

        u  2      uv
 x y z  dx x  x  y  y 
  u 
    uw (29)
z  z 
The equations for the other components take a similar form. The Reynolds Stresses
(  uv,  uw , etc.) are treated as extra stresses that arise from the turbulent nature of the
flow.
The Reynolds Stresses Tensor arises as an additional unknown that needs to be modelled in
order to close the system of equations. There are many ways in which this can be achieved,
all relying to greater or lesser extents on further assumptions and simplifications. It is worth
mentioning few particular approaches for modelling the Reynolds Stresses Tensor since this
form an important aspect of the application guidelines given later.

2.3.1.1. Eddy Viscosity model

The simplest turbulence modelling approach is built on the concept of a turbulent viscosity,
T. This relates the turbulent stresses appearing in the RANS equations to the gradients of
time-averaged velocity (i.e. the rate of strain) in direct analogy to the classical interpretation of
viscous stresses in laminar flow by means of the fluid viscosity, . Thus for example, in a
shear layer where the dominant velocity gradient is u/y (u is time-averaged velocity in the
principal direction of flow and y is the cross-stream co-ordinate) the turbulent shear stress is
given as ·T·u/y.
In the Boussinesq hypothesis, Reynolds stresses are modelled using an eddy viscosity T:
u i u j 2 u 2
Rij    u iu j  T (  )  T k  ij  k ij (30)
x j xi 3 xk 3
From dimensional considerations, T/ is proportional to V·L, where V is a velocity scale and
L is a length scale of turbulent motions (often called the mixing length). Both the velocity scale
V and the length scale L are determined by the state of turbulence, and, over the years,
various prescriptions for V and L have been proposed.
2.3.1.1.1 Algebraic (or zero-equation) models
The simplest prescription of V and L is with the so called algebraic (or zero-equation) class of
models. These assume that V and L can be related by algebraic equations to the local
properties of the flow, see, for example, Cebeci and Smith (1974) and Baldwin and Lomax
(1978). For example, in a wake or free shear layer V is often taken as proportional to the
velocity difference across the flow and L is taken as constant and proportional to the width of
the layer. In a boundary layer close to the wall V is given as L·u/y (or L· where  is the
magnitude of the vorticity) and L is related to the wall-normal distance from the wall (y-
direction). The outer part of the boundary layer is treated in a similar manner to a wake. The
turbulent Prandtl number is given a constant value close to unity except very close to a wall
where viscous effects become important.
Algebraic models of turbulence have the virtue of simplicity and are widely used with
considerable success for simple shear flows such as attached boundary layers, jets and
wakes.
It should be noted that for ship flows, this modelling approach is generally accepted to be
unsatisfactory as the state of turbulence is not locally determined but related to the upstream
history of the flow. A more sophisticated modelling is required.
2.3.1.1.2 One-equation models
The one-equation models attempt to improve on the zero-equation models by using an eddy
viscosity that no longer depends purely on the local flow conditions but takes into account
where the flow has come from, i.e. upon the flow history. The majority of approaches seek to
determine V and L separately and then construct T/ as the product of V and L. Almost
1/2
without exception, V is identified with k , where k is the kinetic energy per unit mass of fluid
arising from the turbulent fluctuations in velocity around the time-averaged velocity. A
transport equation for k can be derived from the Navier Stokes equations and this is the single
transport equation in the one-equation model. This is closed (i.e. reduced to a form involving
only calculated variables) by introducing simple modelling assumptions thereby furnishing a
robust prescription for V which accounts for non-local effects. It is then possible to

algebraically prescribe L with reasonable confidence (e.g. in regions close to a wall) whilst
solving the k-equation for the velocity scale V.
Spalart and Allmaras (1992) have devised an alternative formulation of a one-equation model
which determines the turbulent viscosity directly from a single transport equation for T.
Menter (1994b) introduced a transformation of the two equations turbulent viscosity model
into one equation turbulence model (KE1E) by expressing the time derivative of the eddy
viscosity by the time derivatives of k and ε.
These models have been proved very successful for practical turbulent flows in aerospace
applications, particularly in the USA. For ship flows, Spalart and Allmaras (1992) is probably
the most popular one-equation model these days. It has also been tested in Application 1:
Numerical Towing tank by at least two participants.
2.3.1.1.3 Two-equation k-epsilon models
For general applications, it is usual to solve two separate transport equations to determine V
and L, giving rise to the name two-equation model. In combination with the transport equation
for k, an additional transport equation is solved for a quantity which determines the length
scale L. This class of models (two-equation models) is the most commonly used in industrial
application since it is the simplest level of closure which does not require geometry or flow
regime dependent input.
A very popular version of two equation models has been the k- model, where  is the rate at
which turbulent energy is dissipated by the action of viscosity on the smallest eddies (Launder
and Spalding, 1974). A modelled transport equation for  is solved and then L is determined
as C k / where C is a constant. The key point to remember is that the k- model is
3/2
generally applicable only to high Reynolds number flows with a turbulence structure that is
homogenous, and in which production and dissipation of turbulence is in balance.
The standard k- model with wall functions, as set out by Launder and Spalding (1974) has
been extensively used in industrial applications. It is therefore of value to catalogue the major
weaknesses associated with this model in practical application (The manuals of commercial
and in-house codes may proffer alternative and equally effective advice, and many
commercial codes will include alternatives to the standard k- model):
 The turbulent kinetic energy is over-predicted in regions of flow impingement and re-
attachment leading to poor prediction of the development of flow around leading
edges and bluff bodies. Kato and Launder (1993) have proposed a modification in the
production term of the k transport equation to tackle this problem.
 Regions of re-circulation in a swirling flow are under-estimated.
 Highly swirling flows are generally poorly predicted due to the complex strain fields.
 Mixing is poorly predicted in flows with strong buoyancy effects or high streamlines
curvature.
 Flow separation from surfaces under the action of adverse pressure gradients is
poorly predicted.
 Flow recovery following re-attachment is poorly predicted.
 The spreading rates of wakes and round jets are predicted incorrectly.
 Turbulence driven secondary flows in straight ducts of non-circular cross section are
not predicted at all. Linear eddy viscosity models cannot capture this feature.
 Laminar and transitional regions of flow cannot be modelled with the standard k-
model.

It should also be noted that for problems in steady ship flows, this modelling approach is
generally accepted to be unsatisfactory, other than for the most preliminary of assessments of
the flow field.
2.3.1.1.4 Two-equation k-omega models
A widely used type of two equation model is the k- model, where  corresponds to a
frequency for the large eddies (Wilcox, 1998). A modelled transport equation for  is solved
1/2
and L is then determined as k /.
The main strengths of the k- model are:
 The model equations do not contain terms which are undefined at the wall (unlike k-
model) and they can be integrated to the wall without using wall functions.
 The model is accurate and robust for a wide range of boundary layer flows with
pressure gradient.
 Shear Stress Transport, SST model (Menter, 1994) uses a blending function to
gradually transition from the standard k- model near the wall to a high Reynolds
number version of the k- model in the outer portion of the boundary layer
 SST model contains also a modified turbulent viscosity formulation to account for the
transport effects of the principal turbulent Shear Stress
 A more advanced option for rotational correction (Hellsten, 1998) is often also
available
The k- model performs very well close to walls in boundary layer flows, particularly under
strong adverse pressure gradients. However it is very sensitive to the free stream value of 
and unless great care is taken in setting this value, spurious results are obtained in both
boundary layer flows and free shear flows. The k- model is less sensitive to free stream
values but generally inadequate in adverse pressure gradients and so Menter (1993, 1994a,
1994b, 1996) has proposed a model which retains the properties of k- close to the wall and
gradually blends into the k- model away from the wall: Shear Stress Transport. This model
has been shown to eliminate the free stream sensitivity problem without sacrificing the k-
near wall performance.
The performance of two-equation turbulence models deteriorates when the turbulence
structure is no longer close to local equilibrium. This occurs when the ratio of the production
of turbulence energy to the rate at which it is dissipated at the small scales (i.e. ) departs
significantly from its ‘equilibrium value’, or equivalently when dimensionless strain rates (i.e.
absolute value of the rate of strain times k/) become large. Various attempts have been
made to modify two equation turbulence models to account for strong non-equilibrium effects.
For example, the so-called SST (Shear Stress Transport) variation of Menter's model, Menter
(1993, 1996), leads to marked improvements in performance for non-equilibrium boundary
layer regions such as may be found close to separation.
One of the weak points of the k- models is the  wall boundary condition (Eça and Hoekstra,
2004). The exact solution of  in the viscous sub-layer goes to infinity at a wall.
2.3.1.2. Non linear Eddy Viscosity model: EASM

Approach for dealing with complex strain is provided by the non-linear eddy viscosity class of
models; see for example, Apsley et al. (1997). These models retain the idea that the turbulent
stresses can be algebraically related to the rate of strain (i.e. time averaged velocity
gradients), but higher order quadratic and cubic terms are included. Such models are gaining

in popularity since they involve the same number of equations as two equation models and
thus are computationally efficient.
In the Explicit Algebraic Stress Model, EASM, the equation for Reynolds Stress anisotropy
Tensor is derived from the Reynolds Stress Transport equation (Gatski and Speziale, 1993).
Obtaining an explicit expression for the anisotropy tensor involves 2 main assumptions:
 A weak equilibrium condition on the turbulent stress anisotropy first proposed by Rodi
(1976)
 Anisotropy of the turbulence transport and viscous diffusion is proportional to the
anisotropy of Reynolds stresses
An explicit analytical solution of the algebraic equation for the Reynolds Stress anisotropy
tensor is thus derived and a 2 equations model is then used to obtain the turbulent velocity
and length scale.
It should be noted that during the EFFORT-project (Verkuyl et al. 2003) which involved
computations of ship-hull flows at model scale and full scale, the analysis of the computed
wakes (in particular looking at the strength of the bilge vortex) gave better results with EASM
compared to eddy viscosity k- models. This model looks promising for ship hydrodynamics
as it proved to be accurate and rather economical being based on a 2 equations model.
2.3.1.3. Reynolds Stress Transport Model

Abandoning the isotropic eddy-viscosity hypothesis, the RSM closes the Reynolds-averaged
Navier-Stokes equations by solving transport equations for the Reynolds stresses, together
with an equation for the dissipation rate. This means that five additional transport equations
are required in 2D flows and that seven additional transport equations must be solved in 3D.
These Reynolds Stress Transport methods are becoming accepted as feasible in ship
hydrodynamics application, and have been shown to give superior results to eddy viscosity
modelling albeit at the cost of increased computing time. It should be emphasised that these
too contain modelled terms, derived from a combination of theoretical argument and
empiricism, and should be used with care.
2.3.1.4. Conclusion on RANSE turbulence modelling

Many other models have been developed which are not referred to here, and for further
details the interested reader is referred to the very extensive literature on this subject.
The user should be aware that there is no universally valid general model of turbulence that is
accurate for all classes of flows.
 Validation of the turbulence model is necessary
 If possible, the user should examine the effect and sensitivity of results to the
turbulence model by changing the turbulence model being used.
 The relevance of turbulence modelling only becomes significant in CFD simulations
when other sources of errors, in particular numerical errors, have been removed or
properly controlled.
Finally, it should be recalled that all of the above discussion relating to turbulence modelling
applies chiefly to flows with a steady mean. These modelling approaches may also be used
where the mean flow varies over a time scale which is sufficiently large (slowly varying), or
the eddies contained in the flow are sufficiently large, slow and weak, that the primary
assumptions underlying the above equations are valid. It remains a matter of debate as to
whether such assumptions are appropriate to hydrodynamic flows.

2.3.2. LES
In Large Eddy Simulation, LES, large eddies are resolved directly, while small eddies are
modelled. Large eddy simulation (LES) thus falls between DNS and RANSE in terms of the
fraction of the resolved scales. The rationale behind LES can be summarised as follows:
 Momentum, mass, energy, and other passive scalars are transported mostly by large
eddies.
 Large eddies are more problem-dependent. They are dictated by the geometries and
boundary conditions of the flow involved.
 Small eddies are less dependent on the geometry, tend to be more isotropic, and are
consequently more universal.
 The chance of finding a universal turbulence model is much higher for small eddies.
Resolving only the large eddies allows one to use much coarser mesh and larger times-step
sizes in LES than in DNS. However, LES still requires substantially finer meshes and smaller
time-step sizes than those typically used for RANSE calculations. In addition, LES has to be
run for a sufficiently long flow-time to obtain stable statistics of the flow being modelled. As a
result, the computational cost involved with LES is normally orders of magnitudes higher than
that for steady RANSE calculations in terms of memory (RAM) and CPU time. High-
performance computing (e.g., parallel computing) is a necessity for LES, especially for
industrial applications.
The following sections give details of the governing equations for LES, the subgrid-scale
turbulence models, and the boundary conditions.
2.3.2.1. Filtered Navier Stokes equations

The LES equations are obtained by filtering the time dependant Navier Stokes equations in
either Fourier space (wave number) or configuration space (physical). The filtering process
filters out the eddies whose scales are smaller than the grid spacing used in the computation.
A filtered variable is defined as:
 ( x)    ( x' )G( x, x' )dx' (31)
D
Where D is the fluid domain and G is the filter function that determines the scale of the
resolved eddies.
The continuity equation becomes:
ui
0 (32)
xi
The momentum equations become:

u i u i u j p   ij  ij
   ( ) (33)
t x j xi x j x j x j
Where  ij is the stress tensor due to molecular viscosity and  ij is the subgrid scale stress
defined as
 ij   ui u j  ui u j (34)

2.3.2.2. Sub-grid scale turbulence models
The sub-grid scale stresses are modelled using a Boussinesq hypothesis similar to RANSE,
involving a sub-grid scale turbulent viscosity,  t .
This sub-grid scale turbulent viscosity can be modelled using different formulations
(Smagorinsky Lilly, Dynamic Smagorinsky Lilly, WALE, Dynamic kinetic energy…).
2.3.2.3. Perturbations on the inlet

It is generally important to specify some realistic turbulent and velocity inflow boundary
conditions in order to get accurate prediction of the downstream flow.
Different methods exist in order to generate some representative time dependant inlet
conditions where perturbations are added on top of a specified mean velocity profile. It is
possible to solve for a periodic LES channel flow upstream of the boundary conditions in
order to represent velocities and turbulence at the inlet or some methods also exist in order to
generate anisotropic, inhomogeneous turbulence such as the Vortex method (Mathey et al,
2003), the Spectral synthesizer (Kraichnan, 1970 and Smirnov et al., 2001)...
However, in hydrodynamics it should be noted that certain applications may not in practice
have a turbulent inflow and it may be that attention needs to be given to transition modelling
where boundary layers are of interest.
2.3.3. Hybrid RANSE/LES
For High Reynolds wall bounded flows, LES becomes prohibitively expensive to resolve the
near-wall region. Using RANSE in the near wall regions would significantly mitigate the mesh
resolution requirements.
In the DES approach, the unsteady RANSE models are employed in the near-wall regions,
while the filtered versions of the same models are used in the regions away from the near-
wall. The LES region is normally associated with the core turbulent region where large
turbulence scales play a dominant role. In this region, the DES models recover the respective
sub-grid models. In the near-wall region, the respective unsteady RANSE models are
recovered.
DES models have been specifically designed to address high Reynolds number wall bounded
flows, where the cost of a near-wall resolving Large Eddy Simulation would be prohibitive.
The difference with the LES model is that it relies only in the required resolution in the
boundary layers. There are different DES flavours depending on the RANSE model used in
the near wall regions.
The application of DES, however, may still require significant CPU resources and therefore,
as a general guideline, it is recommended that the conventional turbulence models employing
the Reynolds-averaged approach be used for practical calculations.
2.3.4. Near wall treatment

In wall attached boundary layers, the normal gradients in the flow variables become extremely
large as wall distance reduces to zero. A large number of mesh points packed close to the
wall is required to resolve these gradients. Furthermore, as the wall is approached, turbulent

fluctuations are suppressed and viscous effects become important in the region known as the
viscous sub-layer. This modified turbulence structure means that many standard turbulence
models (see summary given above) are not valid all the way through to the wall; thus special
wall modelling procedures are required.
2.3.4.1. Wall functions

This is the procedure most commonly used in industrial practice. The difficult near-wall region
is not explicitly resolved with the numerical model but is bridged using so called wall functions
(Rodi, 1981; Wilcox, 1998). In order to construct these functions the region close to the wall is
characterised in terms of variables rendered dimensionless with respect to conditions at the
wall.
where w is the wall shear stress.
1/2
The wall friction velocity u is defined as (w/)
Let y be normal distance from the wall and let U be time-averaged velocity parallel to the wall.
+ +
Then the dimensionless velocity, U and dimensionless wall distance, y are defined as U/u
and y··u / respectively.
+
If the flow close to the wall is determined by conditions at the wall then U can be expected to
+ +
be a universal function of y up to some limiting value of y . This is indeed observed in
+ +
practice, with a linear relationship between U and y in the viscous sub-layer, and a
logarithmic relationship, known as the law of the wall, in the layers adjacent to this (so-called
+
log-layer). The y -limit of validity depends on external factors such as pressure gradient and
the penetration of far field influences. In some circumstances the range of validity may also be
affected by local influences such as buoyancy forces if there is strong heat transfer at the
1/2
wall. The turbulence velocity (k ) and length scales, when treated in the same way also
exhibit a universal behaviour.
These universal functions can be used to relate flow variables at the first computational mesh
point, displaced some distance y from the wall, directly to the wall shear stress without
+
resolving the structure in between. The only constraint on the value of y is that y at the mesh
point remains within the limit of validity of the wall functions. A similar universal, non-
dimensional function can be constructed which relates the temperature difference between
the wall and the mesh point to heat flux at the wall (Rodi, 1981). This can be used to bridge
the near-wall region when solving the energy equation.
2.3.4.2. Near wall resolution

As already mentioned, universal near-wall behaviour over a practical range of y+ may not be
realisable everywhere in a flow. Under such circumstances the wall-function concept breaks
down and its use will lead to significant error, particularly if wall friction and heat transfer rates
are important. The alternative is to fully resolve the flow structure through to the wall. Some
turbulence models can be validly used for this purpose, others cannot. For example, the k-
two-equation model can be deployed through to the wall as can the one-equation k-L model
(e.g. Wolfstein, 1969). The standard k- and RSM models cannot. Various so-called low-
Reynolds number versions of the k- and RSM models have been proposed incorporating
modifications which remove this limitation (Patel et al., 1985; Wilcox, 1998). Alternatively the
standard k- and RSM models can be used in the interior of the flow and coupled to the k-L
model which is used to resolve just the wall region. This is known as a two-layer model.
Whatever modelling approach is adopted, a large number of mesh points must be packed into
a very narrow region adjacent to the wall in order to capture the variation in the flow variables.
A y+ value of 1 at the wall should be targeted for near wall resolution.

2.4. Free surface Modelling

There are essentially two approaches to free surface modelling for viscous flows using
RANSE solvers: interface tracking and interface capturing.
The main disadvantage of interface capturing over interface tracking is the need to predict
where grid refinement is required as the location of breaking waves, etc. will not generally be
known in advance. Dynamic grid refinement could be used with Interface capturing
techniques in order to refine the cells in the vicinity of the interface and coarsen the cells
elsewhere.
Another method used for free surface modelling is Smooth Particle Hydrodynamics, SPH.
This method differs from the traditional grid based CFD calculation as it is a mesh free
lagrangian method.
2.4.1. Interface tracking

Interface tracking involves generation of a grid covering just the liquid domain. One of the
domain boundaries is then, by default, the free surface where the boundary conditions are
applied. The grid is adapted to the position of the free surface at each time step. Grid
adaptation may be made computationally more efficient by methods such as moving points
along predefined lines or by updating the free surface position only after several time steps,
having solved the free surface using the pressure boundary condition at intermediate steps.
The method can currently only be used in the absence of steep or breaking waves to avoid
contortion of the grid.
2.4.2. Interface capturing

The alternative approach, interface capturing, involves solving the RANSE on a
predetermined grid which covers the whole domain. Three main methods are covered below:
2.4.2.1. Marker-and-cell
Mass-less tracer particles are introduced into the fluid near the free surface and tracked
throughout the calculation. This scheme can cope with non-linearities such as breaking waves
and has produced some good results. However, it is computationally expensive.
2.4.2.2. Volume Of Fluid method, VOF

The two fluid phases are considered to make up one single fluid. The position of each phase
is described by assigning a volume fraction of either 0 or 1. The free surface is then identified
with the region of raid change in this volume fraction.
The tracking of the interface is accomplished by solving a continuity equation of the volume
fraction of one phase, either through implicit or explicit scheme. A specific discretisation
scheme should also be used for the volume fraction in conjunction with the implicit / explicit
resolution of the volume fraction equation.
Some volume fraction discretisation schemes apply a special interpolation treatment to the
cells where the interface lies by considering the volume fraction as a discontinuous variable,
such as the geometric reconstruction scheme (generalisation from the work of Youngs (1982).

Other volume fraction discretisation schemes treat the volume fraction as a continuous
variable together with high resolution / compression in order to maintain a sharp resolution of
the interface:
 High Resolution Interface Capture, HRIC (Muzaferija et al., 1998)
 Compressive Interface Capturing Scheme for Arbitrary Meshes, CICSAM (Ubbink,
1997).
A single transport equation is generally solved throughout the domain, similar to the
momentum equations detailed in 2.1.2. The dependency on the volume fraction appears in
the momentum equations throughout the volume fraction averaged fluid properties.
Recently, many advances have been made in modelling ship motions using VOF methods,
and these appear to be gaining in popularity for practical applications.
2.4.2.3. Level set technique

In this method, a scalar “level set” function is defined in each cell. Initially, it is set equal to
the distance from the free surface, positive in one direction and negative in the other. At
every later instance the function is computed from the condition that its total (material)
derivative with respect to time is zero. This means that the value of the function is constant
with time on all fluid particles on the free surface. These points will always be on the free
surface, since the relative normal velocity is zero. Thus the surface can be found at each
time by finding the surface of zero value of the level set function. The surface is obtained in
both air and water, but a smoothing layer needs to be introduced at the interface where the
density and viscosity exhibit large jumps.
2.4.3. SPH
The smoothed particle hydrodynamics (SPH) method is a mesh free method that works by
dividing the fluid into a set of discrete elements, referred to as particles. These particles have
a spatial distance (known as the "smoothing length", typically represented in equations by h),
over which their properties are "smoothed" by a kernel function. This means that any physical
quantity of any particle can be obtained by summing the relevant properties of all the particles
which lie within the range of the kernel. For example, using Monagahan's popular cubic spline
kernel the temperature field at position r depends on the temperatures of all the particles
within a radial distance 2h of r (Wikipedia on Smoothed Particle Hydrodynamics, 2009).
The contributions of each particle to a property are weighted according to their distance from
the particle of interest, and their density. Mathematically, this is governed by the kernel
function (symbol W). Kernel functions commonly used include the Gaussian function and the
cubic spline. The latter function is exactly zero for particles further away than two smoothing
lengths (unlike the Gaussian, where there is a small contribution at any finite distance away).
This has the advantage of saving computational effort by not including the relatively minor
contributions from distant particles.
Smoothed particle hydrodynamics is being increasingly used to model fluid motion as well.
This is due to several benefits over traditional grid-based techniques. First, SPH guarantees
conservation of mass without extra computation since the particles themselves represent
mass. Second, SPH computes pressure from weighted contributions of neighbouring particles
rather than by solving linear systems of equations. Finally, unlike grid-based techniques which
must track fluid boundaries, SPH creates a free surface for two-phase interacting fluids
directly since the particles represent the denser fluid (usually water) and empty space
represents the lighter fluid (usually air). For these reasons it is possible to simulate fluid
motion using SPH in real time. However, both grid-based and SPH techniques still require the

generation of renderable free surface geometry using a polygonisation technique such as

metaballs and marching cubes, point splatting, or "carpet" visualisation.
One drawback over grid-based techniques is the need for large numbers of particles to
produce simulations of equivalent resolution. In the typical implementation of both uniform
grids and SPH particle techniques, many voxels or particles will be used to fill water volumes
which are never rendered. However, accuracy can be significantly higher with sophisticated
grid-based techniques, especially those coupled with particle methods (such as particle level
sets).
2.5. Wave tank modelling

The wave tank modelling tends to mimic wave tank testing facilities by generating waves
upstream and minimising wave reflection downstream. Two different approaches are
presented, namely:
 a wave maker approach which uses a conventional Navier Stokes solver and
introduces a wave maker upstream and a wave damper downstream.
 a coupling approach between the resolution of the undisturbed incident wave field
(using fully nonlinear potential flow theory) and the resolution of non linear diffracted
flow (using a modified Navier-Stokes solver).
2.5.1. Wave maker

The propagation of a wave can be implemented in a CFD code by solving the case transient
and generating the right type of wave at the domain inlet (wave maker). Some particular
attention should be paid at minimising the numerical diffusion with higher order numerical
scheme (grid refinement and spatial / time discretisation) for propagating the wave at the right
amplitude/frequency and at absorbing the waves close to the outlet (wave beach) for damping
the reflection effect from the outlet.
The propagations of the water waves should be so as to minimise the loss of wave energy
and conserve wave speed. At the same time, it is necessary that the waves experienced by a
floating body within the domain of interest should be of required wave amplitude, frequency
and kinematics.
2.5.1.1. Wave maker inlet

The boundary condition at the inlet is applied in the form of a harmonic velocity and volume
fraction position according to different wave types: regular, bi-chromatic and irregular. This
inlet formulation can also be used in the entire domain as an initial field solution for velocities
and volume fraction.

2.5.1.1.1 Regular wave
The first order Airy wave theory defines the following velocity boundary conditions as a
function of time and position.
H gT cosh[2 ( z  d ) /  ]
u cos(kx  t )
2  cosh(2d /  )
(35)
H gT sinh[ 2 ( z  d ) /  ]
w sin( kx  t )
2  cosh(2d /  )
Where: u, w are the horizontal and vertical velocity components respectively

x, z are the horizontal and vertical coordinates
g is the acceleration due to gravity
H is the wave height
T is the wave period
λ is the wave length
k is the wave number (2π/λ)
d is the water depth.
Airy wave theory provides a solution to the linear problem in which the pressure can be found
from the sum of the dynamic and static pressures, and the wave elevation calculated from the
simple expression:
H
 cos(kx  t ) (36)
2
However, the theory assumes that the wave elevations are small and hence expressions for
velocity do not apply in the trough to crest region. In addition, the expression for the wave
elevation does not include the non-linear effect resulting from the application of the pressure
term in the Bernoulli equation proportional to the square of the velocity.
Both of these factors give rise to some incompatibilities between the wave theory and the
solutions generated close to the wave inlet boundary. The elevation η is therefore corrected
for the dynamic pressure,
1
  1  (u 2  w2)
2g (37)

Where 1 is the linear first order wave elevation as given above, and u,w are the horizontal
first order velocity components evaluated at z = 0, the mean still water level.
The volume fraction of water is set to 0 above

 and 1 below
.

2.5.1.1.2 Bi-Chromatic Waves
For bi-chromatic waves the boundary condition at the inlet is applied in the form of a velocity,
prescribed from the sum of two first order (Airy) waves as describes earlier, i.e.
Hn gTn cosh[2 ( z  d ) / n]

u 
n 1, 2 2 n cosh(2d / n)
cos(knx  nt  n)
(38)
Hn gTn sinh[ 2 ( z  d ) / n]
w 
n 1, 2 2 n cosh(2d / n)
sin(knx  nt  n)
Where: u, w are the horizontal and vertical velocity components respectively

x, z are the horizontal and vertical coordinates
g is the acceleration due to gravity
Hn is the wave height of each component n
Tn is the wave period of each component n
λn is the wave length of each component n
kn is the wave number (2π/λn) of each component n
n is the phase of each component
d is the water depth.
Airy wave theory provides a solution to the linear problem in which the pressure can be found
from the sum of the dynamic and static pressures, and the wave elevation calculated from the
simple expression:
Hn
 
n 1, 2 2
cos(knx  nt  n) (39)
This expression for the wave height may again be corrected according to equation (37) given
earlier to second order. At the inlet boundary, as with the mono-chromatic wave, this
expression is used to calculate the VOF fraction at the boundary, and hence defines also the
height of the fluid inlet boundary over which the velocities defined by equation (38) are
calculated.
2.5.1.1.3 Irregular Waves
For Irregular Waves, the boundary treatment is identical to that of mono- and bi-chromatic
waves in terms of its implementation as a User Defined Function. Thus far, the form of wave
spectrum used is that of the JONSWAP Spectrum as given by the following formulation:

f 0 
 2
4
 ( f  f 0) 
g (2 ) f exp  ( ) 

2 4 5 5 exp    (40)
S( f )   
 2 f
2


4 0
 f 
Where:
0.22
 gX 
  0.076   (41)
2 
 U 10 

0.33
 gX  g
f 0  3.5  2  2
(42)
 U 10  U 10
And
σ = 0.07 if f  f0 or σ = 0.09 if f  f0
X: is the fetch,
U10: is the wind speed at 10m above the sea surface
The implementation of this expression as a realisation in terms of wave particle velocities and
surface elevations requires the further assumption that the wave field is comprised of the sum
of a series of linear Airy waves. The wave elevation at any point in space and time is
synthesised from the wave spectrum using a numerical method in which the frequency space
(ω) is discretised into a number (i) of segments Δωi, creating wave packets defined in energy
content by the product of the spectral ordinate at ωi and frequency segment width Δω. The
first order wave elevation at any point in space and time is given by:
n
1 x, t    ai sin  i  (43)
i 1
χi = 2π(kix-ωit+εi) = mi (x-cit), so mi = 2πki and wave speed, ci = ωi /ki.

ki = wave number = 1/λi
ωi = wave frequency = 1/T = sqrt(gki/T) = ci /λi
εi = random phase (between 0 and 2π) assigned to each wave packet.
½
ai = [2 Si1(ωi) ∆ωi] which is the amplitude of the wave at frequency ωi with the same
energy as the defined wave packet.
Si1 = 1-Sided power spectral density associated with η1
∆ωi = frequency interval.
There is also a second order correction to this for irregular waves which is ignored in the
following.
The velocity components at the inlet boundary are given in an analogous way by:
n
ki ai g coshsi z  d 
u   sin  i  (44)
i 1 i coshsi d 
n
ki ai g sinh si z  d 
w cos i  (45)
i 1 i coshsi d 
The implementation of these expressions consists of an initialisation of the entire flow field at
in start of the simulation, and then evaluation of the fluid velocities at the inlet boundary
according to equations (44) and (45) up to and including all cells wherein the volume fraction
of fluid calculated at the boundary using (43) is appropriate.
2.5.1.2. Damping beach

A number of options exist for the definition of the outflow boundary conditions. The choice, as
in all CFD calculations, is to specify either the outflow velocities or the outflow pressure (and
allow the velocities to be calculated in order to retain mass continuity).

The definition of the outflow velocity would require an exact knowledge of the diffracted and
radiated wave field at the boundary. Although theoretically possible, it was found during the
EXPRO-CFD project to be difficult to implement at a practical level, since there is some
incompatibility between classical methods for the calculation of a diffracted wave field
(potential flow) and that which would be generated by an unsteady RANSE calculation. The
specification of both wave and pressure field by this route is known to lead to instabilities.
A practical option is to dampen the wave near the outlet boundary. This could be achieved by
using a numerical beach to diffuse the waves away before they reach the outflow boundary:
 Grid stretching in a portion of the domain upstream of the outlet
 Momentum loss in the vertical direction is applied in a portion of the domain
upstream of the outlet.
The momentum loss is defined as follow:

S Mz   K Loss U Uz
2
 x 
with K Loss  K B  1 X B , X B  L   1 X B  L , X B  L  
 L  (46)
The coefficients β and KB are adjusted depending on the wavelength and/or the length of the
numerical beach.
L
βL
Start of the numerical

beach: XB Outlet
KB
Strength of the
numerical beach
Figure 2-2: Numerical Damping Beach
2.5.2. SWENSE
The Spectral Wave Explicit Navier Stokes Equations (SWENSE) approach has been
developed to treat wave-structures interaction under viscous flow theory efficiently: the
undisturbed incident wave field is modelled using fully nonlinear potential flow theory, while
the non linear diffracted flow is simulated using a modified Navier-Stokes solver.

2.5.2.1. Problems caused by wave simulation under viscous flow theory

Resistance and propulsion analysis are today more and more often addressed using viscous
flow solvers based on the solution of Reynolds Averaged Navier-Stokes Equations (RANSE),
because turbulence, viscosity or flow separation effects play an important role in the physics
of those phenomena (particularly in the vicinity of the ship). The most usual method to
introduce influence of waves on the ship in RANSE solvers is a wave maker approach based
on wave generation on the outer boundary of the computational domain by imposing wave
kinematics or boundary motions (Section2.5.1-Wave maker). Unfortunately problems are
linked to this technique: very fine refinement of the free surface is required in order to ensure
a good propagation of the waves with no noticeable damping (typically at least 50 nodes per
wavelength for regular waves) which leads to very large CPU time. Moreover it can be hard
during simulation of wave-body interaction to avoid reflection of the total wavefield (composed
by incident and diffracted waves) at the outside boundaries because it is complicated to
define an efficient open field condition. So successive wave reflections on the body or the
paddle affect the incoming wave train and reduce the useable duration of the numerical
simulation.
Finally if direct generation of 2D regular wave trains can be considered it seems very
problematic to achieve computations with complex waves like irregular wave trains or 3D sea
states.
2.5.2.2. Principle of the SWENSE approach

In fact it appears that the potential flow theory is well suited (more than viscous flow theory) to
compute nonlinear wave generation and propagation (with no breaking) with accuracy and
weak CPU time consuming. Thus it seems interesting to use methods combining potential
flow theory for wave propagation and viscous flow theory for studying phenomena where
viscosity and turbulence will have a strong influence on the flow: wave breaking, interaction of
traveling waves with a ship or an offshore structure.
In the literature many examples of such combinations or coupling can be found : to study
wave breaking, Grilli et al. (2004) use a BEM (Boundary Element Method) for the wave
propagation; this computation is then used to initialise a wave breaking computation with a
VOF (Volume OF Fluid) method solving RANS Equations.
To solve the problem of a ship advancing in water at rest and the computation of the wake far
from the ship, Guillerm and Alessandrini (2003) combine a RANSE procedure to compute the
flow in the vicinity of the body and a potential flow computation far from the ship obtained by a
Fourier-Kochin method. The coupling between these two methods is based on a surface of
connection around the ship where velocities are needed (to be used as boundary conditions)
for the RANSE procedure and the potential flow computation. Unfortunately this kind of
coupling was limited to linearised potential flow models.
Instead of splitting time or space, Dommermuth et al (1997) use a decomposition of the
unknowns (for exemple velocities) in a sum of two terms: a first term which is an irrotational
velocity (deriving from a scalar potential) and a second one which is rotational.
In the method presented here (Luquet et al., 2004) all unknowns of the problem (velocities,
pressure and free surface elevation) are decomposed in a sum of two terms: a term
corresponding to the incident wave train which is propagated in the computational domain
(calculated by a spectral method under nonlinear potential flow theory) and another term
corresponding to the diffracted part of the flow. By assuming that the total flow verifies
RANSE, modified RANSE verified by the diffracted flow are solved by a 3D viscous flow
solver and will be named in the following SWENSE (Spectral Wave Explicit Navier-Stokes
Equations).

Figure 2-3: Three steps of the SWENSE approach

At each time step the SWENSE procedure can be summarised as follows:
 Incident terms are described explicitly using a nonlinear potential flow model
 Then these terms are introduced in a modified viscous flow solver to compute the
diffracted flow due to influence of the ship on the incident wave train
 The total field is then obtained by summing incident and diffracted parts.
Using this kind of procedure the propagation of gravity waves is no more calculated in the
RANSE solver:
 The computational time is more or less the same as a computation on still water since
the size of the mesh used is equivalent to those of a still water computation (no more
wave generation and propagation calculation).
 Usual meshes refined near the ship and stretched near the outer boundary can be
used and lead to an efficient damping of the diffracted field generated
 The useable incident fields are only limited by the potential model. By using HOS
(High Order Spectral) methods, it is possible to generate regular and irregular 3D
waves like prescribed focused wave packets or real sea-states. Only breaking waves
can not be simulated.
For comparison, at the CFD symposium held in Tokyo (2005), the use of SWENSE reduces
the CPU time requirement of a factor one hundred on a forward speed diffraction test case
(even if we assume the fact that meshes do not represent accurately the farfield but only the
nearfield and the forces coefficients).

2.5.2.3. Governing Equations

The previous set of RANSE is modified in order to formulate a problem for the nonlinear
diffracted flow. Primitive unknowns (Cartesian components of velocity U , i  1,2,3 , pressure
i
P and free-surface elevation h ) are decomposed as follows:
U i  U Ii  U Di

 P  PI  PD
h  h  h (47)
 I D
Variables with the subscripts I and D represent incident and diffracted variables respectively.
This decomposition is then introduced in the set of initial RANSE in Cartesian coordinates
assuming that the incident wave flow fulfils Euler equations. So we obtain a new set of
equations (issued from mass conservation and RANSE) called SWENS Equations:
U Di
aik 0 (48)
 k
U D  j j  t j  U D
t 

 ai U I  U Dj  v g 
i 1
Re eff
f j  a ki 
 i
ak
  j
 
2 
1  UD P
 g kk  ak D
Re eff  k
2
 k
 2U D  t j 

1
Re eff
g jk
 j  k
 a ki
 i
a
 j

U Ik  U Dk  (49)
k l
1  jk  U2 
U I  k  j   U I
 g fj
I
 a j U D  a ij ti 
Re eff   j  k  j      k

The same decomposition is used for free surface boundary conditions. In SWENSE incident
variables (velocities, dynamic pressure, free-surface elevations and their gradients) are
known explicitly. Their values are directly computed at each time step knowing kinematics
and interface position of the incident flow. Then, only the diffracted variables are unknowns of
the problem and have to be solved by the modified viscous solver.

2.5.2.4. Incident Flow Model

In order to apply the SWENSE method it is necessary to be able to know at each time step all
of the characteristics of the incident fields (velocities, pressure, free surface elevations).
To compute a non-linear regular incident wave train, an algorithm based on the stream
function theory of Rienecker & Fenton (1981) has been implemented. This algorithm has
been chosen as it can generate the solution of steady progressive periodic waves of an
irrotational flow over a horizontal bed for a wide range of depths, amplitudes and wavelengths
(in the limit of breaking waves) with large accuracy.
Even if the field is represented in the water only, the solution is independent of the Navier-
Stokes grid and can be computed everywhere. This is essential because in the SWENSE
method, values of the incident field will be possibly needed above the undisturbed incident
free surface. The total free surface elevation can indeed be higher than the incident elevation.
That is why this point is decisive for the choice of the incident field model. Figure 2-4 shows
the axial velocity field to demonstrate the continuity of the wave field under, through and also
upon the free surface.
Figure 2-4: Contours of the axial velocity in the whole space. The bold line represents the free
surface.
An irregular wave train could be prescribed as well by using a spectral formulation. This kind
of procedure has been already developed by Ferrant & Le Touzé (2002) combining a
spectral formulation and a Boundary Element Method (BEM) to simulate an irregular wave
train interacting with a 3D body under potential flow theory.

2.6. Hydro-mechanical modelling

The flow solver has to be coupled to a solver of the flight dynamics equations for the
computation of the trajectory of an immersed body free in all or a number of its 6 degrees of
freedom (6 DOF).
2.6.1. Introduction
The barycentric reference frame is attached to the centre of gravity of the body and remains
parallel to the absolute reference frame of the fluid at rest in the far-field.
The resulting forces (pressure, viscous stresses) exerted at a given instant on the body are
used to compute the translation and angular accelerations in the barycentric reference frame.
The formulation of the solid body dynamics adds yet another high non-linearity to the set of
equations, and a coupling algorithm has to be implemented to iterate over these non-
linearities and those of the flow equations
2.6.2. 6 DOF
The matrix expression of the inertial and external forces balance, when written at the centre of
inertia of the solid body, with vector components in the barycentric reference frame, is
expressed in the following way:

 For the translational degrees of freedom, with F the external force, [M b] the mass

matrix, and X t the vector of translational d.o.f. of the solid body (absolute
coordinates of the centre of gravity):
 
X t [M b ]1 F (50)

 For the rotational degrees of freedom, with M the external moment vector, [Kb ] the
inertia matrix in the Galilean reference frame and [Rb ] the third-order entrainment

tensor, non-linear function of the rotational angles X  :
   
X [Kb ] M  XT[Rb ]X (51)
The model is consistent in that all vectors (fluid velocity, forces and moments, accelerations,
translation DOF) are expressed in projection on the same absolute reference frame.
2.6.3. 6 DOF for ship manoeuvring

In order to compute unsteady manoeuvres of ship, the solver must be able to compute the
motions of the ship under forces computed at each time step. This is done by solving the
standard Euler’s law in the body fixed coordinate frame centered on G. The six components
of ship velocity and position are then obtained, and allow to move the ship by moving the grid,
whereas the velocity are used for boundary conditions on the hull.
Navier-Stokes equations are solved in the fixed general axis center on the origin of the
reference frame (R0). This coordinate frame is Galilean, so acceleration terms of the fluid in
the Navier-Stokes equations do not have to be taken into accounts, and then reduce the
complexity of the problem to solve.

The Euler’s laws are solved in the body fixed coordinate frame centered on G (RG).
Figure 2-5: Fixed reference and body fixed axis
Figure 2-6: Decomposition of the rotation of the ship
The rotation matrix used to transform coordinates from the fixed coordinate frame to the body
fixed coordinate frame is defined by:
 c c  s c   s  
PR3 R c s s   s c  c c   s s s  c s  (52)
 
c s c   s s   c s   s s c  c c 
The ship motion equations are written using the Euler’s laws:
  dVG / R 
m   R  VG / R   F/ R
  dt 
 
, (53)
 d  R 
I R   R   I R R   M G / R
 dt
where:
m is the mass of the ship
IR is the inertial matrix of the ship

VG / R is the velocity vector of the center of gravity

R is the angular velocity vector


F/ R is the total forces acting the ship (hydrodynamic, aerodynamic, gravitational and
external forces)

M G / R is the total momentum acting on the ship
Assuming that the integration scheme is of the following form (general multi-step methods):
p
vn 1  vn     Ci  f t, vn1i  (54)
i 0
where:
Ci are coefficients of implicit or explicit multi-step methods
 is the time step

vn 1 is the unknown velocity components at time step n+1
vn is the known velocity components at time step n
Euler’s laws can then be written in the following form:
MV  F (55)
with:
m 
 m 
 
 m 
M   (56)
 Ixx Ixy Ixz 
 Iyx Iyy Iyz 
 
 Izx Izy Izz 
Vectors components are defined by:
Vi  vi n 1  vi n (57)
 p 
Fi   
  
C j fi j   j  V  j
.i   (58)
 j 0 
Where i from 1 to 3 refers to x, y and z forces coordinates, and i from 4 to 6 refers to
momentum coordinates.
Velocity components V are calculated by solving the linear system. Positions of the ship are
then directly integrated from ship velocities.
A predictor corrector scheme, based on explicit and implicit second order multi-step methods
(Adams Baschford and Adams Moulton methods) is used to integrate the ship motions.
Predictor step:

 3 1 
 zn 1  zn  h   f n  f n 1 
  2 2 
 (59)
z  1 1 
 zn  h   zn 1  zn 
 n 1 2 2 
Corrector step:
 1 1 
 zn 1  zn  h   f n 1  f n 
 2 2 
 (60)
z 1 1 
 zn  h   zn 1  zn 
 n 1 2 2 
2.7. Cavitation Modelling

2.7.1. Introduction
This Chapter aims to provide a brief introduction into the models available to numerically
simulate cavitating flows. Unlike turbulence modelling, where the models are inserted in the
Reynolds stress terms of the Navier-Stokes equations, cavitation modelling is generally not
easily added to a single phase flow code, as it affects not only the right hand term of the
momentum equation, but also interferes with the mass and energy conservation equations.
A first classification for computational methods dealing with cavitating flows is to distinguish
Boundary Element Methods from Field methods.
2.7.2. Boundary Element Methods

Early work on the mathematical modelling of cavitating flows can be considered to have
th
started already at the end of the 19 century with conformal mapping techniques. The
development of lifting surface theory in the sixties and the seventies of the 20th century
offered possibilities for a better mathematical and numerical modelling through linearized
cavity theory. The first applications of Boundary Element Methods for cavitating flows over
foils and propellers were published in the eighties of the last century. A review of
developments in this type of computational methods can for example be found in Vaz (2005).
These Boundary Element Methods or BEMs are very efficient for non-viscous potential flows,
as they only solve the singularities on the surface, thereby rendering the flow variables over
the complete computational domain. The Laplace equation that is solved in these methods is
obtained after assuming a non-viscous (  0 ), non rotational (  u  0 ) flow. The
momentum equations then reduce to the statement that fluid acceleration is directly related to
the pressure fluid gradient. The continuity equation takes the form of a Laplace equation for
the velocity potential. The energy equation reduces to the well known Bernoulli equation,
relating the pressure to the velocity field and the unsteady potential time derivative.
The velocity field is subsequently obtained from solving the Laplace equation where the cavity
dynamics are obtained from the application of suitable boundary conditions (BC’s). The
boundary conditions pertinent to the cavity are the prescription of a zero normal velocity
component on the cavity surface and the prescription of vapour pressure on the cavity
surface. The differences between the various Boundary Element Codes are essentially in the
implementation of the BC’s, which can be done through elements on the actual cavity surface,
or on elements that are positioned on the foil surface, but that use transpiration velocities to
meet the BC’s on the cavity surface. Other important sources for differences between the
codes are the implementation of the cavity detachment point near the leading edge of the foil,

and the formulation and implementation of the closure condition at the trailing edge of the
sheet.
2.7.3. Field Methods

An important difference between potential flow methods that solve the Laplace Surface
Integral problem and the Euler and RANS methods is that the latter classes of flow equations
cannot be reduced to a boundary integral problem. Instead, their solution needs a
discretization of the whole fluid domain that is relevant to the problem. These latter methods
are therefore also referred to as field methods. Obviously, these so-called field methods
require more computational effort than BEM’s.
A major distinction in field methods is made in codes dealing with the flow as a homogeneous
mixture of phases and codes solving for the flow for each separate component in the flow,
which may be more than two (liquid and vapor). The first category will be referred to as
“homogeneous mixture” methods, the second group as “multi component models”.
Homogeneous mixture models are by far dominating in multi phase codes for propeller-rudder
analysis. Multi component models are for instance used in flows that are dispersed in sprays
consisting of little droplets, such as occur in the fuel injection system of combustion engines.
2.7.3.1. Homogeneous mixture models

An essential assumption in the mixture models is that the flow can be described as the flow
of a continuum, thus allowing for a solution of the Navier Stokes equations for only one
(mixed) fluid with variable density and viscosity. The contents of a grid cell are represented by
the ratio of the phases present in that cell (the vapour or liquid ratio). This ratio strongly varies
in the vicinity of liquid-vapour interfaces. This is precisely where some of the difficulties with
this model occur, because it is principally impossible to describe a sharp interface through a
uniform mixture ratio per cell.

The mixture quantities like specific mass m and kinematic viscosity
m are assumed to be
constituted by a proportionality relation with the respective fractions:
m  l l   v v (61)
where the sum of fractions needs to be unity by definition:

l   v  1 , and so eq. (61) can
be expressed in the unknown

l .
Compared to the case of a single-phase fluid, where the specific mass of the fluid is usually

known, an additional unknown l is introduced now, requiring an additional equation. This
equation has the form of a transport equation for either the liquid or the vapour fraction, and
has the following form:
 
 (u i )  Q
t xi (62)
Despite the inherent problem of predicting sharp interfaces, the mixture approach seems to
offer a reasonable model for simulating the dynamics of developed sheet cavitation. This
approach is by far the most adopted model in cavitating flow CFD.

Within the class of Homogeneous mixture models, the following subdivision can be made
following Koop (2008):
 barotropic flow models
 transport equation models (TEM)
 bubble two phase transport models (BFT)
 mass transfer rate models
 thermodynamic-equilibrium models
2.7.3.1.1 Barotropic flow models
The simplest multiphase flow model is a formulation that couples the mixture density to the
pressure through a barotropic state law    ( p) ranging from the liquid density through a
transitional zone to the vapour density (see e.g. Delannoy and Kueny, 1990). Although this
simple model seems to give reasonable predictions for e.g. a cavitating propeller, it lacks two
vorticity production terms that occur in the vorticity transport equation. One of these is
obviously the shear stress term. The other lacking term is the so-called baroclinic production
term:
1
    p
 (63)
This production term is zero in a barotropic model, as the gradients in density and pressure
are always parallel. Although vorticity production plays an important role in cavitating flows,
the importance of the shear stress term and the baroclinic production term have not yet been
quantified. Our current understanding is that much of the vorticity production seems inertia
driven.
2.7.3.1.2 Transport Equation models (TEM)
The Transport Equation Models treat the flow of the vapour or liquid by means of a transport
equation for the volume or mass fraction of liquid or vapour (and sometimes even a third
fraction for e.g. an inert gas). In these models it is assumed that the state variables such as
pressure, velocity and temperature, are continuous (homogeneous mixture assumption).
Transition from one phase into the other is accounted for by production/destruction terms
(also referred to as source terms) in the transport equation. These methods have the
advantage that they can take into account the time dependency of the mass transfer
phenomena through empirical laws for the source term, as opposed to barotropic models.
However, the determination of the constants in the source terms appears to be somewhat
arbitrary. Broadly two classes of source models can be distinguished: source terms based on
bubble dynamics and fully empirical source terms.
In the first class, cavitation is modelled as the growth and collapse process of vapour bubbles,
excluding interaction between the bubbles. The bubbles originate from nuclei and their growth
and collapse depends on the surrounding conditions. The dynamics of the bubbles is
supposed to be governed by the Rayleigh or the Rayleigh-Plesset equation. Important
contributions to this way of modelling are a.o. provided by Kubota et al. (1992) and Sauer et
al. (2000).
Important contributions to the development of the fully empirical models are a.o. given by
Merkle (1998) and Kunz et al. (2000). Senocak and Shyy (2002) evaluated different
formulations of this type of models. They concluded that the results of the methods for the

pressure distribution generally agree, and that the biggest differences in pressure distribution
occur in the closure region of the sheet cavity.
2.7.3.1.3 Thermodynamic equilibrium models
Another approach to get rid of the empirical source terms in the Transport Equation models is
based on a thermodynamic equilibrium cavitation model, employing the full set of
conservation laws for the flow of a homogeneous mixture. Additional equations of state are
needed to close the system of equations. The two-phase regime is enforced to be in
thermodynamic and mechanical equilibrium at all locations in the flow. In regions where two
phases can exist, the homogeneous equilibrium model returns a prediction of the bulk density
of the two-fluid mixture, from which a volume fraction of vapour is obtained given the
temperature dependent saturation-state relations. This method is capable of capturing phase
transitions, condensation shocks and other multi-phase flow features, but non-equilibrium
effects are by definition excluded. Furthermore, due to the hyperbolic nature of the equations,
wave propagation phenomena can be studied for all phases. Important contributions to this
model are a.o. made by Saurel et al. (1999) and Schmidt et al. (2006).
The only empiricism involved in this model is the constants in the equations of state of the
pure vapour and liquid phases. These coefficients are however well established in the physics
domain. To close the description of the two-phase mixture and for the compressible water,
physical mixture properties can be obtained from for example tabulated steam tables for
specific mass and enthalpy. Alternatively the Sanchez-Lacombe equation of state for real-fluid
flows, accompanied with the equilibrium assumption in the two-phase region to guarantee a
real-valued speed of sound, has been applied.
2.7.3.2. Multi component models

Multi component models or multiple species methods have first been proposed for detonation
waves in granular explosives and modified for the resolution of multi-phase mixtures and
interface problems between purely compressible materials. These models describe the
cavitating flow by adopting a full set of equations, i.e. continuity, momentum and energy
conservation equations for the vapour phase, the liquid phase and sometimes even an inert
gas phase, together with their own thermodynamic relations. They allow for both mechanical
and thermal non-equilibrium to be taken into account. The coupling between the different
phases is accounted for by appropriate transfer relations, derived from two-phase flow
modelling considerations. Most models lead to non-conservation forms due to the interface
interaction terms. These methods are assumed to possess more generality, but at the cost of
a higher computational effort since a full set of conservation equations has to be solved for
each phase.
2.7.3.3. Compressibility
A far reaching choice in modelling cavitating flows, is the choice for an incompressible fluid, or
for a compressible fluid. The vast majority in the domain of marine propulsor analysis consists
currently of codes for incompressible fluids, and these codes seem to be able to capture the
macroscopic phenomena in cavitating flows. When the attention is focused on the
development and collapse of smaller structures than the main cavity sheet and tip vortex,
such as secondary cavitating vortices originating from the main sheet or vortex, local cavity
events such as collapse become important. For the smaller spatial and temporal scales
involved, compressibility becomes important as the speed of sound of the vapour-fluid mixture
rapidly decreases for only small percentages of vapour (see Figure 2-7).

Figure 2-7: Speed of sound in a mixture of water and vapour (Koop, 2008)
Experience from the VIRTUE project indicates that the convergence behaviour of
incompressible codes may suffer from the pressure waves generated upon collapse of a
cavity, that then occur instantaneously throughout the complete computational domain.
2.7.3.4. Viscosity and turbulence modelling

Another important choice in the modelling of cavitating flows addresses the inclusion of
viscosity, and further, the choice of turbulence model. Viscosity is understood to play an
important role in the detachment of the boundary layer flow as occurring in the Leading Edge
region of a sheet cavity. It also controls the vortex strength of a Leading Edge vortex or tip
vortex. There are however accumulating indications that viscosity has a small if not negligible
influence on the cavity dynamics in the closure region of a cavitating sheet.
When using RANSE solvers for the simulation of sheet cavitation on a foil, the turbulence
model may have to be adjusted. Numerical exercises have shown that such adjustments can
have considerable effect on the shape and dynamics of the cavity (see e.g. Reboud et al.,
1998). However, since the effects of cavitation on turbulence and vice versa are not well
understood, there is no clear guideline yet. It has however been demonstrated in the project
that the grid density in the closure region does have an important effect on the viscous
dissipation as well (see e.g. Section 7.4.1:Case 4-a: 2D Foil).

3. Computational Theory
The differences between viscous turbulent flow solvers and the previously described potential
flow methods are numerous. They stem from the considerably more complex forms of non-
linear partial differential equation being addressed, and the need to carry out the numerical
discretisation in 3D, rather than being able to reduce the problem to a set of surface integrals.
3.1. Potential Flow Theory
3.1.1. Panel mesh generation

The body surface is usually modelled using combinations of quadrilateral and triangular
panels. It is important to remember that the panels are not physical but actually represent
distributions of sources, vortices or dipoles. Distributions can take the following forms:
 A single point source distribution is used for simplicity where the distance to the panel
source is large, as a refined distribution will not produce a more accurate result.
 A cluster of point sources improves the accuracy over a single point source and is still
simpler and faster than a true surface distribution.
 A plane panel of constant source strength was suggested by Hess and Smith (1964)
and is used by many present day codes.
 A non-planar panel of constant source strength, made up of triangular elements, was
developed by Jensen (1988)
 A curved panel with bilinearly varying source strength was developed by Wei (1987)
and is used in some higher order methods.
 Methods based on spline representations of both the surface and the potential are
also now in use (Newman)
For most ship flows, flat panels give sufficient accuracy and are simpler to construct.
The source strength must be continuous across the panel joints, except at a trailing edge.
Here the Kutta condition must be satisfied, such that the flow can't go around the trailing
edge, but must leave the body there.
Many basic panel codes use flat, planar panels with the vertices located on the body. A
common method of curved surface panel generation uses a parametric cubic spline or
NURBS procedure to approximate the true body curve. Automatic panel generation programs
are widely available for this. The facility to define a number of bodies or separate parts of the
same body independently allows complex bodies and flows to be investigated.
For steady free surface applications the number of panels required for accurate calculation of
the wave profile is inversely proportional to the square of the Froude number. The lower the
speed range to be investigated, the more panels there will be in the longitudinal free surface
mesh.
3.1.2. Boundary Conditions definition

The inherent assumptions of potential flow methods are that the flow is inviscid and
irrotational.

The conditions imposed on the disturbance potential  are that:

 the velocity potential satisfies Laplace’s equation everywhere outside of the body and
the wake;
 the disturbance potential due to the body vanishes at infinity;
 the normal component of velocity is zero on the body surface;
 the Kutta condition of a finite velocity at the trailing edge is satisfied;
 the wake sheet is a stream surface with equal pressure either side.
For free surface flows the following boundary conditions are also invoked:
 Kinematic boundary condition: Water does not penetrate the free surface or the body.
 Equilibrium condition: The weight and external forces acting on the body are in
equilibrium.
 Dynamic boundary condition: Atmospheric pressure acts at the water surface.
 Radiation boundary condition: This depends upon the forward speed of the vessel,
depth of water, and the problem type. For the infinite water depth, steady forward
speed problem, waves exist only in the sector behind the ship and do not propagate
ahead. As the water depth decreases, and in very shallow water, issues relating to
depth effects may become important. For sea-keeping problems at forward speed,
care needs to be taken with the combination of ship speed and wave frequencies
used (the important parameter is (U/g).
 Domain boundary conditions: Waves generated by the vessel should pass out of the
computational domain without reflection.
3.1.3. Non-linear methods

Many of the phenomena associated with a ship or free structure in a seaway will be non-linear
(e.g. roll damping, motions of water on deck, added resistance in waves).
Most of these effects cannot be accounted for by even second or third order linear
perturbation methods. The traditional method of dealing with these non-linear effects is by
linearising the motion and solving the motion equations in the time domain. Time integration is
performed by standard methods such as Euler, Runge-Kutta or predictor-corrector methods.

3.2. Viscous Computational Fluid Dynamics
3.2.1. Solver methodology

It should be noted that, of all the methods described below, the Finite Volume method is by
far the most common approach to be found in current commercial CFD codes. Much of the
guidance given in this document is based on the assumption that the reader is following this
approach.
3.2.1.1. Finite difference method

The finite difference method is the oldest of the methods, considered to have been developed
by Euler in 1768, and is used to obtain numerical solutions to differential equations by hand
calculation. At each node point of the grid used to describe the fluid domain, Taylor series
expansions are used to generate finite difference approximations to the derivatives of the
RANS equations. The derivatives appearing in the governing equations are then replaced by
these finite difference expressions, yielding an algebraic equation for the flow solution at each
grid point.
It is the simplest method to apply, but requires a high degree of regularity of the mesh. In
general, the mesh must be structured. Grid points should form an ordered array in three
dimensions, allowing the finite difference approximations to be formed from local, easily
addressed locations. Grid spacing need not be uniform, but there are limits (guidelines given
later) on the amount of grid stretching or distortion that is possible, and at the same time
maintain accuracy. Topologically, these finite difference structured grids must fit the
constraints of general co-ordinate systems with, for example, Cartesian grids must fit within 6
sided computational domains. However the use of an intermediate co-ordinate mapping
allows this otherwise quite major geometrical constraint to be relaxed, such that complex
shapes (including ship hulls) can be modelled.
3.2.1.2. Finite element method

The finite element method was developed initially as a procedure for constructing matrix
solutions to stress and displacement calculations in structural analysis. The method uses
simple piecewise polynomial functions on local elements to describe the variations of the
unknown flow variables. When these approximate functions are substituted into the governing
equation it will not hold exactly, and the concept of a residual is introduced to measure the
errors. These residuals are then minimised by multiplying by a set of weighting functions and
then integrating. This results in a set of algebraic equations for the unknown terms of the
approximating functions and hence the flow solution can be found.
Finite element methods are not used extensively in CFD, although there are a number of
commercial and research based codes available. For certain classes of flow, FE methods
bring a high degree of formalised accuracy to the numerical modelling process. However, it
has generally been found that FE methods require greater computational resources and CPU
effort than equivalent Finite Volume methods, and therefore their popularity, at least in
Europe, is limited.
3.2.1.3. Spectral method

Spectral methods use the same general approach as the finite difference and finite element
methods by again replacing the unknowns of the governing equation with truncated series.

The difference is that, where the previous two methods use local approximations, the spectral
method approximation is valid throughout the entire domain. The approximation is either by
means of truncated Fourier series or by series of Chebyshev polynomials. The discrepancy
between the exact solution and the approximation is dealt with using a weighted residuals
concept similar to finite element.
3.2.1.4. Finite volume method

The finite volume method was first introduced by McDonald (1971) and MacCormack and
Paullay (1972) for the solution of two dimensional time dependent Euler equations, and
extended to three dimensional flows by Rizzi and Inouye (1973). The method discretises the
integral form of the conservation laws directly in physical space. The resulting statements
express the exact conservation of relevant properties for each finite cell volume. Finite-
difference-type approximations are then substituted for the terms of the integrated equations,
forming algebraic equations that are solved by an iterative method.
As the method works with the cell volumes and not the grid intersection points, unstructured
meshes can be used where a large number of options are open for the definition of the shape
and location of the control volumes around which the conservation laws are expressed. A
‘finite element’ type mesh can be used where the mesh is formed by combinations of
triangular or quadrilateral cells (or tetrahedra and pyramids in three dimensions), where the
mesh cannot be identified with co-ordinates lines. This type of unstructured mesh, although
requiring careful bookkeeping, can offer greater flexibility for complicated geometries.
Flow variables can be stored either at Cell Centre or Cell Vertex locations. Conveniently, the
cells coincide with the control volumes if using the Cell Centred scheme. For the Cell Vertex
scheme, additional volumes are required to be constructed; however, the scheme has the
advantage that boundary conditions are more easily applied since the variables are known on
all boundaries.
3.2.2. Meshing
The fluid computational domain is split into a three dimensional grid of data points. This grid
may be “structured” or “unstructured” depending upon the details of the numerical scheme
and solvers employed.
3.2.2.1. Mesh generation

The computational grid represents the geometry of the region of interest. It consists of grid
cells that provide an adequate resolution of the geometrical features. In hydrodynamics, body-
fitted grids are used almost universally. However, several kinds of mesh topology are
available:
 Structured grid: The points of a block are addressed by a triple of indices (ijk). The
connectivity is straight-forward because cells adjacent to a given face are identified
by the indices. Cell edges form continuous mesh lines which start and end on
opposite block faces. Cells have the shape of hexahedral, but a small number of
prisms, pyramids and tetrahedra with degenerated faces and edges are sometimes
accepted.
 Block structured grid: For the sake of flexibility the mesh is assembled from a
number of structured blocks attached to each other. Attachments may be regular, i.e.
cell faces of adjacent blocks match, or arbitrary (general attachment without matching
cell faces).

 Unstructured grid: Meshes are allowed to be assembled cell by cell freely without
considering continuity of mesh lines. Hence, the connectivity information for each cell
face needs to be stored in a table. The most typical cell shape is the tetrahedron, but
any other form is possible including hexahedra, prisms, pyramids, polygons etc.
 Hybrid grid: This grid combines structured with unstructured meshes.
The discretisation error decreases with increasing number of cells, i.e. with decreasing cell
size. However, due to limitations imposed by the increased computer storage and run-time
some compromise is nearly always inevitable.
In addition to grid density, the quality of a mesh depends on various criteria such as the
shape of the cells (aspect ratio, skewness, included angle of adjacent faces), distances of cell
faces from boundaries or spatial distribution of cell sizes. The introduction of special
topological features such as O-grids or C-grids and care taken to locate block-interfaces in a
sensible manner can help to improve the overall quality of a block-structured mesh.
Unstructured meshing techniques may take advantage of prism layers with structured
submeshes close to domain boundaries.
3.2.2.2. Mesh deformation

Mesh deformation needs to be used when the shape of the domain is changing with time; for
example, the motion of the domain boundary due to:
 Prescribed motion: Forced Heave, Forced Pitch, Propeller blades
 Resulting motion: 6 Degrees Of Freedom solvers (6 DOF), where the motion of the
solid body is a resultant of the force balance on the body
There are different motion methods available for updating the volume mesh in the deforming
zone such as:
 Smoothing method, this uses the spring analogy for deforming cells and maintains
the same mesh topology. It can be used in a number of situations as long as the
mesh does not get squeeze or stretch too much.
 Layering method, this adds and removes layer of cells adjacent to a moving
boundary.
 Sliding method, where one mesh zone slides relative to another zone along a grid
interface. It can be efficiently used for pitch motion or for propellers.
 Remeshing method, this allows remeshing a zone. This method is particularly useful
for bigger local displacement.
3.2.2.3. Mesh superposition

The superposition mesh technique is referred as Chimera or Overset grid approach:
Structured mesh blocks are placed freely in the domain to fit the geometrical boundaries and
to satisfy resolution requirements.
Blocks may overlap, and instead of attachments at block boundaries information between
different blocks is transferred in the overlapping region.
3.2.2.4. Mesh adaption

The grid must be fine enough to capture all important flow features. This may be achieved by
local grid refinement. Unstructured meshes are especially well suited for this purpose. If block

structured grids are used local refinement results in block attachments with dissimilar number
of grid lines. Some CFD codes provide algorithms to adapt the grid resolution locally
according to numerical criteria from the flow solution, such as gradient information or error
estimators.
An example of local grid refinement is adapting / coarsening the cells in as a function of the
gradient of volume fraction is order to locally refine the mesh around the free surface. This
can also been done dynamically at every n time steps in a time dependant simulation.
3.2.3. Boundary conditions

Two types of boundary conditions and combinations of them are most commonly
encountered. The Dirichlet condition, specifies the distribution of a physical quantity over the
boundary at a given time step and the Neumann condition, defines the distribution of its first
derivative.
There is common consent that good practice for outflow boundaries is to set the convective
derivative normal to the boundary face equal to zero and to combine this with a streamwise
extrapolation of transported quantities. At pressure boundaries the same treatment is usually
applied. Open boundaries bring about the following difficulties:
 Non-physical reflection of outgoing information back into the domain, including free
surface waves.
 Difficulties in providing information about the properties of the fluid which may
inadvertently enter the domain from the outside.
 Difficulties may also arise if open boundary conditions are placed in regions of high
swirl, large curvatures or pressure gradients.
Some CFD codes prevent fluid from entering into the domain through open boundaries. In
order to avoid undesirable side effects open boundaries should be placed very carefully.
In many real applications, there is a frequent difficulty to define some of the boundary
conditions at the inlet and outlet of a calculation domain in the detail that is needed for an
accurate simulation. A typical example is the specification of the turbulence properties
(turbulence intensity and length scale) at the inlet flow boundary, as these are practically
arbitrary in marine CFD. However, for special cases such as propeller or water jet flows the
user needs to be aware of these problems and needs to develop a good feel for the certainty
or uncertainty of the boundary conditions that are imposed.
3.2.4. Numerical Resolution
3.2.4.1. Spatial discretisation

Different numerical methods evaluate the fluxes at the same grid locations as the transported
quantities or somewhere in between (collocated or staggered grids). In both cases, an
algebraic approximation of the spatial functions is required to calculate the gradients at these
locations. This approximation is called the differencing scheme in finite volume or difference
methods or the basis function in finite element methods. The accuracy of the scheme
depends on the form of the algebraic relationship and on the number of grid points used in it
(stencil). The spatial discretisation or truncation error equals the difference between the
scheme and the exact formulation based on a Taylor expansion series. A formally second
order scheme is consistent with the exact formulation up to the term with a power of two, a
third order scheme also takes into account the next higher term. The formal order of accuracy
is not preserved on irregular meshes, where it reduces by one. Reducing the cell size by
introducing a finer grid has the biggest impact on the accuracy of the solution if higher order

rd
schemes are applied. Halving the elements in all directions using a 3 order scheme will
st
reduce the numerical error by a factor of 8, while this factor is only 2 with a 1 order scheme.
If the solution of the physical problem considered is smooth and exhibits only small gradients
even a first order scheme can do a good job, but it is not at all suitable for general
engineering applications involving complex flows with large gradients and thin boundary
layers. The large truncation error introduced by the first order upwind scheme, particularly
popular in finite volume methods, is known as numerical viscosity or diffusivity as it gives rise
to artificial diffusion fluxes, which may be much stronger than the real molecular or turbulent
contributions.
On the other hand higher order schemes suffer from a different more obvious problem,
namely the appearance of a characteristic wavy pattern with a wavelength of two cell sizes in
the neighbourhood of steep gradients. These so-called wiggles are caused by dispersion
errors, i.e. waves with different wave lengths are not transported with the same speed.
Dispersion errors are most prominent in central differencing schemes for finite volume
methods and quadratic basis function schemes for finite element methods. Higher order
upwind schemes are less prone to it. If necessary, this problem may be remedied using
special (non-linear) TVD or shock-capturing schemes. Due to their capability to resolve steep
gradients or interfaces while avoiding dispersion effects they are frequently applied in
supersonic flows with shock waves or for the transport of scalar quantities with weak
molecular diffusion.
3.2.4.2. Temporal discretisation

Purely steady flow fields with the time-derivative equal zero are only a special case of the
time-dependent equations. In general, fluid flows are transient, whereby the sources for this
time-dependent behaviour are:
 External transient or non-transient forces.
 Transient boundary conditions, moving walls (e.g. the fluttering of an airfoil).
 Vortex stretching, a three-dimensional phenomenon due to the non-linear term of the
governing equations, which also gives rise to the fluctuating nature of turbulence.
The computation of steady turbulent flow is the most common kind of simulation for the
general use of CFD. In these cases the Reynolds-averaged flow is steady while the average
turbulent quantities account for the time-dependence of the turbulent fluctuations. However,
RANSE also allows time-dependent Reynolds-averaged flow fields to be computed, based on
the assumption that the temporal average of the turbulent quantities is not affected by the
global unsteadiness. This is physically correct if the spatial scale of the turbulent eddies is
much smaller than the geometrical scale of the analysed geometry. A time-dependent
simulation is always needed if the scale of eddies or vortices becomes larger and is in
comparable size to the dimensions of the geometry (e.g. the computation of vortex shedding).
If an accurate spatial discretisation is applied, flows which are physically time-dependent will
fail to converge using a steady-state method. Very often convergence problems with a steady
simulation can be interpreted as a hint that the flow is unsteady and a time-stepping scheme
would be appropriate. On the other hand, symmetry boundary conditions may impose a
steady flow, although it would be transient in reality. If the complete geometry including both
sides of the symmetry plane were used the velocity field would oscillate perpetually.
Averaging the solution over a long time interval would lead to a symmetrical field, which,
however, differs from the steady state solution with the symmetry plane.
The temporal discretisation scheme provides an approximation of the time derivative. Most
CFD codes offer first order and second order schemes, which are unconditionally stable and
most effective in terms of computer memory and stability requirements. Low-storage higher-
order Runge-Kutta methods are also available. The order of the scheme and the choice of the

time step influence the size of the amplitude and the phase error, the two components of the
temporal discretisation error. To improve time-accuracy self-adaptive time-stepping
procedures (such as predictor-corrector methods) can be used.
The choice of the time step depends on the time scales of the flow being analysed. If time
steps are too large the simulation might fail to capture important flow and mimic unphysical
steady behaviour. It is therefore advisable to start with relatively small CFL numbers even
though this is not required from the point of view of numerical stability.
3.2.4.3. Initialisation
Before starting the iterative process, an “initial guess” of the solution flow field must be given.
Some attentions must be taken for the steady state initialisation in order to reduce the time to
convergence. The initialisation in steady state should not affect the final converged solutions
as the steady state solutions should not depend from initial conditions but it will influence the
CPU time needed to converge:
 The entire domain and cells can be initialised with a constant value or with some
functions (for example, initialising the hydrostatic pressure head in the water part)
 Part of the domain can be initialised with specific values
 If a parametric study is done (small change in ship velocity, small change in hull
shape…), it is often very CPU efficient to restart the calculation from the results of
previous solution
Some extra care must be taken for the unsteady calculation in order to start from
representative initial conditions, as they will influence the time dependant solutions.
3.2.4.4. Convergence
Iterative algorithms are used for steady state solution methods and for procedures to obtain
an accurate intermediate solution at a given time step in transient methods. Progressively
better estimates of the solution are generated as the iteration count proceeds.
There are no universally accepted criteria for judging the final convergence of a simulation,
and mathematicians have found no formal proof that a converged solution for the Navier-
Stokes equations exists. In some situations the iterative procedure does not converge, but
either diverges or remains at a fixed and unacceptable level of error, or oscillates between
alternative solutions. Careful selection and optimisation of control parameters (such as
damping and relaxation factors or time-steps) may be needed in these cases to ensure that a
converged solution can be found.
The level of convergence is most commonly evaluated based on residuals, on values of
globally integrated parameters, such as lift coefficient, or on time/iteration signals of a
physical quantity at a monitor point, which is an arbitrarily chosen location in the flow domain.
 Residuals are 3D fields associated with a conservation law, such as conservation of
mass or momentum. They indicate how far the present approximate solution is away
from perfect conservation (balance of fluxes). Usually, the residuals are normalised
by dividing by a reference value.
 Convergence is also monitored on the basis of some representative number
characterising the residual level in the 3D flow field.
The large number of variants makes it difficult to give precise statements how to judge
convergence and at which residual level a solution may be considered converged. In
principle, a solution is converged if the level of round-off error is reached. Special care is

needed in defining equivalent levels of convergence if different codes are used for
comparison purpose.
3.2.5. Post processing

Don’t be seduced into believing that the solution is correct just because it has converged and
produced high-quality colour plots (or even seductive video presentations) of the CFD
simulations. Make sure that an elementary interpretation of the flow-field explains the fluid
behaviour and that the trends of the flow analysis can be reconciled with a simple view of the
flow.
Most commercial codes come with some kind of post-processing package. This allows many
of the flow phenomena to be visualised or plotted in graphical form. The two main steps of
post-processing are to determine:
 whether the result is sensible
 whether the result is accurate
Checking the believability of the solution may involve several steps such as checks on
conserved variables, visual confirmation that velocities and pressures are smoothly
distributed and comparison with other similar problem results. The convergence history will
give some indication of whether the problem has reached a steady state solution.
3.2.6. Sensitivity studies

Most CFD problems are dependent on mesh quality and resolution, and this may be even
more so for free surface problems, such as where the free surface meets the body. It may be
easy to find in the literature a suggested panel or mesh density for a particular problem but it
is important to examine the sensitivity of variables such as these on the solution. This may
also take an iterative form, where the initial solution has highlighted an area of insufficient
mesh resolution or skewed panels or grid that need improvement.
3.2.7. Dealing with uncertainties

Uncertainties arise through lack of knowledge. This can be a lack of knowledge of the details
of the problem to be modelled, or of the methods and approximations used to solve the
problem. The latter can only be solved by increased user awareness to the theories and
methods used.
Uncertainties can also occur because of simplification of the problem due to modelling
constraints.

3.3. Optimisation techniques

In addition to modelling and analysis tools integration capabilities, optimal design
technologies have become more and more widely used. Distinction must be made between a
design solution and optimal design:
 Design solution, or individual, is to be considered as a n-tuple [zk]k=1,N, where zk are
the N attributes which describe the solution. For some design methodologies, these
attributes can be divided in two groups: inputs xi and outputs yj so that a solution is
defined by [(xi)i=1,Ninp, (yj)j=1,Nout].
 Optimal design must be considered as the whole set of techniques or methods that
enable to generate knowledge and extract information on the dependencies between
attributes, and thus better understand the design characteristics and make proper
decisions. This is not only restricted to optimisation in the mathematical sense
(function minimisation).
Therefore, optimal design consists in the operation of one or several techniques, to
manipulate and analyse design solutions, and finally make decision. The whole process must
be controlled and monitored by the designer himself, which remains at the centre of the
design search strategy and decision making process.
3.3.1. Data analysis - Design exploration - Design of experiments

The first mission of an optimal design environment is to manipulate and analyse design data
sets, i.e. sets of candidate designs, each described by the above defined attributes.
In practice, two solutions are possible to have such data available either by recovering data
sets from existing information (database or previous designs, data available after a previous
optimal design approach, etc…) or by generating data sets through:
 the definition of designs to be calculated (inputs x i), for example by means of Design
Of Experiments techniques (DOE) which will enable to get a set of candidate
solutions distributed in the most appropriate manner,
 the real calculation of the outputs corresponding to these inputs, by means of the
modelling and analysis tools that are integrated or connected to the optimisation
environment.
Several families of DOE can be used, depending on the problem characteristic such as
number of inputs, cost of calculations, search objectives…
Random or pseudo-random DOE can be used to generate data with uniform distribution in the
stochastic sense. In cases where it is very difficult to find one feasible solution, some specific
algorithms exist: one such algorithm, for example, is CSP (Constraint Satisfaction Problem),
which is based on and heuristic approach (RRT, record to record from Poles, 2004).
When only the statistical response of a system has to be analysed, a Monte-Carlo design of
experiment can be used, which enables to generate data with given distributions around a
mean value, for each design variable. A variant of this is the Latin hypercube for which the
variation domain of each variable is divided in equal intervals within which data are distributed
along a given law.
A natural design of experiment is the full factorial plan. However this becomes prohibitive in
terms of number of data as soon as the number of variables and number of levels increases.
It can then be replaced by other deterministic plans (reduced factorial, cubic face centered,
Box-behnken, Latin square, Plackett Burman, etc…) which enable to provide the maximum
information on main trends and coupling between variables, within a given number of planned
evaluations.

Other design of experiments plan can also be used such as those coming from quality control
world. “Taguchi plans” enable to find data sets which provide the system less sensitive to
uncontrolled noise. D-Optimal plans can be used when classical factorial plans cannot be
used because of the number of points or a constrained space.
Cross-validation design of experiment is based on the use of a Kriging approach (Cressie,
1993) to homogenise the distribution of data within the space, by minimising internal error
criteria calculated by the Kriging method.
Once such data are available, the optimisation environment offers means to analyse the
distribution and interactions between the attributes, the inputs and the outputs, through
 Charts that represent the designs and the design space in many ways,
 Statistical analysis tools that provide quantified information on distributions,
correlations, sensitivities, trends, etc…
From this analysis, the user is able to extract most suitable designs within the ones available,
understand design trends, and make first decisions on design evolutions.
3.3.2. Data modelling - Response surfaces

In many cases, it is relevant to be able to assess quickly the design performance, for new
design alternatives: providing quick answers, leading design searches with reduced modelling
and calculation time. This can be done thanks to Response Surface Methods, which enable to
build such approximations from a given set of existing solutions, check their quality, and use
them within the design search (data analysis, optimisation, etc…), or even in other
applications which require such simplified models.
The principle consists in building approximate multiple variables functions on the basis of an
existing designs database. Several methods exist for this purpose (see examples hereafter),
adapted to several types of behaviours.
Most popular approaches are least square approximation methods which enable to adjust
predefined functions (linear, polynomial, etc…) to a cloud of points. This approach is very
easy to undertake and can lead to good results for behaviour to which predefined functions
are convenient.
Another family of methods are interpolation methods, to which belong k-nearest approaches
in which a solution is deduced from the knowledge of the closest neighbours, with appropriate
weighting factors. A variant of this is the Kriging (Cressie, 1993) approach which exploits the
covariance between points to assess a better estimate of weights and produce intrinsic quality
assessment criteria.
More complex multidimensional functions, including non-linear behaviours, may require the
use of more elaborate approximation methods. Bayesian approaches, like Gaussian
processes, can be efficient and robust in this case (MacKay, 2003). More generally, Neural
Networks are also more and more used, and show a good behaviour for many applications
(MacKay, 2003). However, these approaches may need a good deal of experience and
manpower to build good response surfaces that can be used in the design search.
The construction of response surfaces, whatever the method, is facilitated through optimal
design environments, which provide all the information to assess the quality of the
approximations built (absolute or relative errors, visualisation of approximations, etc…), and
also give indications and even wizard to help this construction. However, this process
normally requires a great care of the user to ensure really usable approximations.

3.3.3. Optimisation
Once first solution(s) is(are) available, and possibly some approximations of the design
behaviour have been built, the problem consists in exploiting this information to progressively
adjust the design variables, so that the design objectives are met. This can be ensured by the
designer himself, who can also be assisted in this process by optimisation algorithms which
will propose alternative solutions towards improvement.
A large number of optimisation algorithms are available, which now enable to cover a wide
range of applications. Many classifications can be used to sort out the various types of
algorithms: deterministic (heuristic or mathematically based) or stochastic, continuous or
discrete, mono-objective or multi-objective, using gradients or not, using internal response
surfaces or not, one or several starting point, constraints intrinsically satisfied or through
penalty methods, etc… Many combinations of these characteristics have been used, which
led to many variants of algorithms. A summary of main reliable algorithms is provided below.
Historically first optimisation algorithms were based on deterministic methods, either based on
heuristic or similar approaches, like for example the SIMPLEX algorithm (Nelder and Mead
downhill Method) which uses a geometric progression in the N+1 dimensional space (N
number of design variables – Fletcher, 1989), or based on the exploitation of the gradients
(derivatives of functions in each design variable), like the BFGS algorithm which is a quasi-
newton approach, using an approximated Hessian matrix (Fletcher, 1989). In this category
lies also the NLPQLP method, which is a very efficient (now a standard) Sequential Quadratic
Programming method developed by Pr.Schittkowski, and includes intrinsic satisfaction of
constraints through a Lagrangian approach, as the other methods are using penalty functions
for this purpose (Schittkowski, 2006). Some of these algorithms, like Levenberg-Marquardt,
are particularly well suited to solve least square problems as found in curve fitting (Fletcher,
1989).
Another family of algorithms, more recently appeared for design optimisation, is using
stochastic approaches. The most well known belong to evolutions strategies, like genetic
algorithms (Poloni, 2003), which mimic the life evolution process of an initial population,
through selection, cross-over, mutation operations. These approaches have led to a huge
number of variants in terms of genes coding and operations models, which are each adapted
to certain problems characteristics: some of them are more capable to keep solutions
diversity and some others can have a more local but efficient convergence behaviour, etc…
See for example Distributed Evolution Strategies algorithm (Bäck, 2004) which is tuned to
quickly find a local solution, just like deterministic algorithms, but without the restrictions
associated with these methods (bearing noisy functions).
Another category of stochastic algorithms is simulated annealing, which uses similarity with
material cooling process, through the random generation of a solution from a given point. The
solution selection with a given probability depends on a decreasing temperature law. Several
variants of this algorithm also exist.
All the above methods are able to solve mono-objective problems of different characteristics.
The traditional way to solve multi-objective problems is to combine all objectives in a weighted
sum of the objectives, and use one or the other algorithms. A major advance in recent
developments is the capability to solve the true multi-objective problem, which consists in
finding all the non-dominated solutions of the multi-objective problem (pareto frontier). This
avoids making combination of quantities of different natures, and allocating weights to the
objectives well before knowing how these objectives behave with each other. Stochastic
algorithms are well suited for searching the pareto front, and several implementations are
available for this purpose with different characteristics: MOGA, Multi-Objective Genetic
Algorithm (Poloni, 2003), MOSA, Multi-Objective Simulated Annealing method (Rigoni, 2003),
NSGA, Non-dominated Sorting Genetic Algorithm (Deb, 2000), MMES, Multimembered
Multiobjective Evolution Strategy (Bäck, 2004). A particular family of algorithms is based on

game theory (ex: MOGT, multi-objective game theory algorithm from Clarich, 2004), which
enables to quickly drive solutions toward the pareto frontier.
A particular implementation of MOGA, which aims at improving the convergence of the
stochastic approach in the main zones of interests, is ARMOGA, Adaptive Range Multi-
Objective Genetic Algorithm (Sasaki, 2005). It consists in analysing the evolution of the
algorithm to progressively adjust the search domain, so that uninteresting zones are excluded
form the search, which then concentrates on relevant areas.
Another approach for solving multi-objective problems has recently appeared and is based on
deterministic algorithms. This is using the NBI (Normal Boundary Intersection – Das, 1996)
methods for the multi-objective purpose, which consists in leading several weighted
deterministic optimisations from several points, taking advantage of the knowledge of the
evaluations history to orientate the search towards the pareto frontier, and improving
convergence of each mono-objective optimisation by exploiting the mutual information of the
intermediate solutions. This can be used together with any deterministic method (NLPQL,
BFGS, SIMPLEX…).
Some other specific algorithms can be used to solve particular problems. For example, in
case of highly constrained problems, where even one feasible solution is difficult to find,
Constraint satisfaction algorithms (CSP – Poles, 2004) can be used, which are based on the
principle of simulated annealing, with an heuristic approach for the selection of worse (i.e.
less feasible) solutions within the randomly generated solutions.
One problem can also consist in trying to add points in some areas where they are missing to
consistently cover the global behaviour of the system. The method objective is of "design of
experiments" type, but the method (MACK, modeFrontier) belongs to the optimisation
methods, as it requires the evaluation of the outputs. Additional points are distributed so that
the relative or absolute error estimated with a Kriging approach must be minimal.
Of course, no algorithm is able to deal with all optimisation problems, and in front of a given
problem, the challenge is to find the most adapted algorithm, or the most efficient combination
of algorithms, which can lead to the relevant solutions.
3.3.4. Decision making

The design process, especially when optimal design approaches as above are undertaken,
may lead to several design variants possibly relevant for a given set of specifications, which
best meet a number of objective criteria. The next phase is to outline a unique final design,
which is the "best" trade-off solution, between these objective criteria, but also possible taking
into account additional judgement criteria, that may not be in the optimisation process itself,
and that may even not be numerically quantifiable. This requires a rational choice or selection
process that embeds the designer’s preferences together with the numerical information. This
is covered by the "multiple criteria decision making" techniques.
These techniques can be operated through a number of steps:
 analysis of the problem and description through a given number of attributes, which
describe the product or system to be designed,
 allocation of importance to these different attributes,
 generation or import of a given set of relevant solutions (described through their
attributes) that best meet the specifications (coming from an optimisation, designs of
the pareto frontier, for example),
 definition of preferences between these designs, through pair-wise comparisons,

 run of a numerical algorithm which solves the above described problem, sorts out the
set of available solutions along the given preferences, and possibly provides non
linear weights variations laws for all the attributes, as a function of their relative value.
This last functionality makes it possible, when relevant, to define a global non-linearly
weighted objective (which can replace separate objectives of the initial problem), and perform
mono-objective optimisation which enables to refine the final solution, taking into account all
the designer's preferences.
The resolution algorithms can be of different types, depending on the complexity of the
problem. Linear resolution can be achieved when attributes are not numerous. The problem
can also be solved through GA based optimisation algorithms, to provide approximate
solutions in case of numerous attributes, with some variants. Specific algorithms can also be
used, which can give importance to either optimistic (favour maximum value of the best
attribute, maximax) or pessimistic (favour best value of the worst attribute, maximin)
behaviour, or any intermediate solution such (see Hurwics MADM from Lecocq, 2000). Some
algorithms also allow selecting the solution which provides the lowest loss with respect to
maximum performance of each attributes, through a norm to be minimised (see Savage
MADM from Lecocq, 2000).
In all cases these approaches should be used in an iterative way in order to achieve the best
solution, under the designer's control.
3.3.5. Robust design

Pure optimisation on performance criteria can lead to solutions which quality is very sensible
to any perturbation on inputs (design variables) or on other variables of the problem. Robust
design consists in ensuring stability of solutions versus such perturbations as well as
performance of these solutions.
Three types of robust design problems can be considered, defined by the area where the
perturbation lies:
 perturbation in the design variables (ex: uncertainties related with the manufacturing
process)
 perturbation in the operational conditions (ex: variations of speed or attitude of an
aircraft or a ship around its nominal operational conditions)
 uncertainty in the design specifications (ex: trying to find early solutions that will be
robust to variations in the design specifications that may occur during the design
process)
The principle to solve such problems is to introduce uncertainty through perturbations in the
design description, defined by given statistical distributions, then to assess the design for all
the defined perturbations, and extract from the results the characteristics of the outputs
distributions. Then the relationship between variability of outputs and variability of inputs can
be analysed, and solutions which reduce this variability selected.
This can be done through classical Monte-Carlo analyses, which provide such variability of
outputs for given sets of designs. This variability can then be used in standard robust design
approaches (ex: 6σ). This can also be done through a true multi-objective problem, where the
performance and the variability tend to be improved at the same time, and which lead to a
performance and variability pareto front, from which trade-off can be decided by the designer
(see Mordo, modeFrontier).
This approach may lead to large computing time, as perturbed designs have to be calculated
in addition to basic designs. This can be overcome by specific developments in the resolution

methods (taking advantage of the smallness of perturbations) or by using response surfaces

approaches to speed up the process (see 3.3.2)
3.3.6. Hybridation
One given technology or algorithm will never be able to solve any type of problem. Depending
on the nature of the problem (discrete or continuous, smooth or noisy, few or many local
extrema, quick or high time consuming assessment, search of trends or refined optimal
solutions, etc…), different methods, and more likely different combinations of methods
(hybridation) should be applied.
Hybridation can concern any combination of any technique: combination of optimisation
algorithms with response surface techniques to speed up the optimisation process,
combination of widespread search algorithms (like GA) with more focused algorithms (e.g.
gradient methods) starting from most interesting solutions from the first one, progressively
modify the parameters of one given algorithm as the design search progresses, etc…
Ideally, hybridation could be an automated process: switching from one method to another on
the basis of rational calculated criteria, adjusting response surfaces as soon as new solutions
are calculated by the optimisation algorithm, etc… In many cases, such an approach is not
practicable to real problems, as the criteria which drive the interaction between several
methods are difficult to estimate, or to tune, for the current problem.
Another approach is to rely on the human capability to globally understand and interpret the
composite behaviour of the algorithms and of the design, so that the designer itself can
decide for hybridation and adapt the search strategy during the search process. This is much
more efficient in many cases, at the condition that the optimal design environment used
allows to smoothly and easily shift from one method to another, or combine several methods,
without loss of information. This capability is offered in most advanced optimal design
environments.
3.3.7. Time demanding modelling and analysis

One obstacle to the diffusion of optimal design technologies is related with the global time
needed to assess many design alternatives, as needed in an optimisation process, when
assessment is performed with high demanding calculation tools, like in CFD. So a specific
field of research and development lies in the study of means to reduce the global response
time of an optimal design approach.
Three main tracks must be followed there:
 work on parametric modelling,
 work on optimisation strategies,
 work on assessment computing time.
3.3.7.1. Parametric modelling

The number of evaluations within an optimal design approach is strongly related with the
number and efficiency of the parameters used to model the freedom in the design. The
objective is here to get the best compromise between
 the number of parameter, which must be as low as possible to reduce the number of
possible combinations,

 the magnitude of the design variations, which must be as large as possible to keep
the diversity of solutions and authorise innovative solutions to be found,
 the accuracy of the modelling, which must also be as large as possible to get solution
as close as possible to the final design (to be produced),
 the reliability of the model (ex CAD models), which must be able to generate viable
variants with as many parameters combinations as possible.
These objectives are widely conflicting objectives, and parametric modelling remains a key
point of an efficient optimisation process. Several steps and methods can be used to ensure
this. The following ones are described below, mainly concerning shape deformation
parameterisation.
3.3.7.1.1 CAD base modelling
The most natural ways to build parametric shape variants is to rely on a CAD tool. This is
satisfactory as CAD tools are generally full part of the classical design procedure, and as it
can be expected that solutions coming from the optimisation process will then be fully
compliant with the other aspects of design or with next stages towards production.
However, even if recent CAD tools claim to be "parametric", it often happens that the CAD
parameters are not necessarily the best parameters for shape optimisation. It is often very
difficult to define the freedom at the place useful for the behaviour which has to be optimised,
with a reduced number of parameters.
Moreover, CAD tools are not designed themselves to support large or various combinations of
parameters values, as it may happen within an optimisation process. So in many cases, a
great number (up to more than 90%) of tentative designs, requested by an optimisation
approach, cannot be generated by the CAD tool because of this limitation.
So, direct CAD based parametric modelling can be useful in some cases, but often lead to
some restrictions which, even if not preventing from any optimisation process when robust
algorithms are used, may lead to a larger number of evaluations than strictly necessary.
In some particular cases, dedicated CAD tools, specifically designed for parametric shape
modelling, can be efficient for this purpose anyway. See for example, the ship hull design
system Friendship-System (Harries, 2003).
3.3.7.1.2 Shape deformation modelling
Alternative solutions to CAD parametric modelling exist, which allow overcoming the above
restrictions. They mainly consist in modelling variations, rather than modelling the shape
itself, in a parametric way. Many implementations of such approaches exists, some working
directly on CAD generated shapes (ex: Iges files) see Bataos (Jacquin, 2003) and Marin
(Hoekstra, 2003), some working directly on grids to be used by the analysis tool (See Sculptor
reference).
These approaches normally allow defining exactly the freedom wished by the designer in
relation with his optimisation problem, with the lowest number of useful parameters.
Grid deformation tools have in addition the advantage to produce grids with same topologies
for all variants, which enables a smoother behaviour of the grid related error during
optimisation approaches, provided that the grid quality criteria remain satisfied during the
variations. In addition, this makes it easy to re-use fields calculated for one variant as initial
condition of another variant, which dramatically reduces calculation time (see 3.3.7.3).

The shortcoming of these approaches is that the optimisation is performed outside the current
design environment, and the recovery of the solution within a CAD tool might be difficult.
However, some solutions are being developed in this domain.
3.3.7.1.3 Multi-level parametric modelling
One solution to reduce the total number of evaluations is to perform multi-level parametric
modelling. In this case, optimisation search can be performed with a restricted number of
main parameters during the first stages, where main design orientations are looked at, then a
more refined parametric modelling can be used in final stages where a better accuracy is
expected and when the search domain is reduced.
This requires however to rely on a modelling which is consistent between the several levels of
details. This is not obvious to ensure, and some work is currently performed on these topics
(Desideri, 2003).
3.3.7.1.4 Upstream work on parametric modelling
In all cases, the parametric modelling must be carefully considered before undertaking any
optimal design approach. This can be eased by the use of an optimal design environment,
reduced to the modelling part (without analysis), and which will allow to deeply investigate the
parametric model: check the reliability of shape generation, assess the parameters influence
and correlations, select appropriate parameters, tune their ranges of variations, etc…
Once this is done, the optimal design search can be started with much more confidence.
3.3.7.2. Efficient optimisation strategies

The second topic to be worked on in case of expensive analysis is the reduction of necessary
evaluations through the use of an appropriate optimisation search strategy. Moreover, in such
circumstances, the calculation process is often complex and may be subject to failure, for a
number of reasons (impossibility to generate the shape, quality of the mesh, convergence
problems, etc..).
So the challenge is to use as few assessments as possible, and to be robust against failure.
The first item can be solved by gradient methods which have very good convergence
capabilities. However these algorithms require the complete availability of all intermediate
points and all their derivatives, which might be difficult to ensure.
On the other hand, stochastic algorithms are much less sensitive to such failures, but often
require a large number of assessments. However, some of these algorithms can be tuned to
have improved convergence towards local minima, which can be interesting in some cases.
So most often, the convenient approach consists in combining several methods (see 3.3.6),
and define a compound "real time" search strategy, i.e. exploiting the interactive capabilities
of optimal design environments to monitor the design search, analyse the available data at
various stages, redefine search space or variables, redefine search algorithm, redefine
starting points or populations, etc…and thus develop the strategy up to convenient results.
Typical approaches used for example in CFD can consist in the following steps:
 calculate a first set of design variants (3.3.1),
 build response surfaces on this basis (3.3.2),

 launch a wide spread stochastic algorithms for some generations, partially using
response surfaces,
 analyse the results, outline domains of interest,…
 rebuild response surfaces,
 search optimum of response surface,
 check with direct calculation of the solutions.
Any variant, to be defined on the basis of the current results, can be considered.
3.3.7.3. Computing time reduction

The last topic to be considered is the reduction of computing time for individual calculations.
The first thing is to rely on efficient calculation infrastructure and simulation software, and the
capability to exploit multi-processing.
In addition to this, the simulation capabilities to reduce computing time have to be exploited.
In particular, CFD codes offer the capability to interpolate fields from one calculation to
another one. In optimisation, most often the changes occur in limited physical areas, and
starting from interpolated fields can dramatically decrease the calculation time. This is even
more efficient if the modelling process ensures the topology consistence between variants, in
which case, no more interpolation is needed and grid related errors are smoothed (see
3.3.7.1.2). A field of investigation exists here to improve such an approach, to exploit these
aspects.
Another current field of research lies in the calculation of gradients, either through adjoint
equations methods, or similar, or even through code derivation. This is however rather
specific to each code and cannot be considered as a generic solution.
The development of simplified solutions on the principles of reduced bases is a promising
area, as it should lead to decisive reduction in calculation times (Chatterjee, 2000 and Bui-
Thanh, 2002).
Another approach is the use of multi-level models. This consists in using rough or simplified
models (either rough grids, or low level physical model) in a first stage of the search, then
refine these models when needed or when the search is more focused.
In conclusion, lot of research is going on in these domains, and efficient methods or tools
should be emerging soon.

4. Application 1: Numerical Towing tank
4.1. Introduction
This part of the Best Practice Guidelines addresses the field called “the Numerical Towing
Tank” in the VIRTUE project. It considers the use of CFD for predicting ship resistance and
flow around the hull; ship propulsion; and optimisation regarding these aspects.
The overall aim of the Numerical Towing Tank, WP1 was to improve the accuracy, reliability
and applicability of existing state-of-the-art CFD codes, through improving problem set-up and
methodology. This development intended to lead to a viable complement to the physical
towing tank testing in the prediction of hydrodynamic performance of ships, in particular the
speed-power relationship, at the design stage. The application of the methods in a dedicated
optimisation process was a second main objective.
To this end and to enable further exploitation of CFD techniques in ship design, the following
objectives were targeted:
 redictive capability through improved flow modelling;
advanced modelling techniques for turbulence, free surface including transom stern
flow, hull roughness and propulsors have been tested for selected test cases. Their
influence on the accuracy of resistance and propulsion performance prediction is
carefully studied.

improved discretisation schemes featuring enhanced robustness, adaptive grids and
numerical efficiency. Extensive uncertainty analysis has been undertaken for
parameters that influence the quantitative accuracy of performance prediction; in
particular, the prediction of wake, resistance and propulsion and its consequences for
scale effects.

test cases; the validation study includes prediction of all resistance components (hull
only) and propeller open water characteristics (propeller only), propulsion test
simulation (hull + propeller) and nominal wake computation.
 Use of the improved and validated CFD methods to investigate scale effects on wake,
wave pattern, viscous and wave resistance, and propeller performance behind the
ship; comparison with existing full scale extrapolation techniques and exploration of
the use of CFD-based extrapolation techniques.

performance prediction) CFD codes with geometry variation/modification modules
and optimisation and sensitivity analysis routines, thus facilitating hull form changes
towards improved designs and multi-disciplinary optimisation.
The guidelines given below have largely been derived from the research results and
experiences collected in this work.
For all the problems addressed in WP1, which include the numerical calculation of viscous
flows around ships, the Reynolds-Averaged Navier-Stokes equations (RANSE) are still the
6 9
only viable mathematical model; due to the large Reynolds numbers involved (10 to 10 ).
Therefore, for the remaining of this section we focus only on RANSE solvers. Complementary
models (as for example optimisers) required by some of the problems presented above are
not addressed in this document.

In the next section, we first discuss the background and main objectives of the work done in
VIRTUE on resistance prediction, hull/propeller interaction prediction and optimisation. Next,
two application cases are discussed that played a central role in workshops in VIRTUE, and
which illustrate current capabilities and remaining issues. Then, specific guidelines are given
on problem setup, modelling and numerics.
4.2. Work-package Tasks
4.2.1. The Numerical Resistance Test

Resistance prediction is of prime importance for all parts of hydrodynamic analysis of ships.
The accurate determination of the overall resistance – wave, viscous and interaction – will
influence all subsequent analysis steps, e.g. the prediction of power requirements. Analysis
and comparison of results obtained from state-of-the-art flow codes, e.g. the Gothenburg
2000 and Tokyo 2005 workshops, show that there is still a large scatter of predicted results. It
is therefore not surprising that the participants in the European thematic network project
MARNET ranked accuracy as the most important topic for future developments in this area.
Precise CFD-based resistance and power predictions would be a substantial step forward for
the maritime industry. Thus, to establish the Virtual Towing Tank it is of utmost importance to
address accuracy and uncertainty of resistance predictions.
The main goal of this task is to improve certain aspects of existing RANSE codes which
seem responsible for an insufficient accuracy of the resistance predictions. Past numerical
experiments indicated that mainly grid resolution (and orientation) and turbulence modelling
are important factors for accuracy and their influence is therefore of particular interest. For
free-surface viscous flows, there are additional aspects such as the capturing schemes, grid
deformation techniques, computational approach for including trim and sinkage, and
modelling of transom stern flows. In the work in Virtue, also other numerical aspects have
been found to contribute to the observed scatter between resistance predictions, and studies
have been done to better control these Validation and uncertainty analysis is difficult and time
consuming but necessary in order to draw correct conclusions.
The substantial differences between the viscous flow around the hull at model and full scale
introduce a major difficulty for performance predictions. Using the improved CFD codes, the
influence of this scale effect on wave pattern, viscous and wave resistance and propeller
performance has therefore been investigated to improve the full-scale predictions. Finally,
some topics relating to modelling needs, in particular regarding free surface or code specific
improvements are carried out in order to achieve the final goal of improved accuracy.
4.2.2. Hull/Propeller Interaction – The numerical propulsion tests

This task was aimed at simulating the performance of the integral ship + propeller system.
The numerical tools enhanced in the previous provide means to predict the propulsive power
and propeller rotation rate by integrating the propulsor (propeller) into the model of the hull
flow, and compute the effects of the propulsor and the interaction between the propeller and
the hull.
Usually, an actuator disk is introduced to simulate propeller effects in RANSE methods. By
applying body forces to the numerical cells in the propeller disk the flow may be accelerated
in the same way as by a propeller with an infinite number of blades, producing the same
thrust and torque. This method has been frequently used both for investigating propeller/hull
interaction and for podded propulsors. More recently considerable developments in the
prediction of propeller blade flows using RANSE methods have been made. Over a range of
advance ratios many methods predict the thrust coefficient within a few per cent of the
measured one and the torque is not much worse; but the pressure distributions are generally

not in good agreement with experimental data, particularly at the inner radii. Some of these
methods have been applied to the operating propeller behind a ship. Promising results were
obtained in predicting propeller/hull interactions effects, but for more accurate predictions of
blade flows and pressure fluctuations the more advanced methods addressed in WP4 will be
required. Both methods to model the 'numerical propulsion test' have been applied and
compared. Therefore, the objectives of this task were to determine what level and precision of
modelling of the propeller is necessary for an accurate prediction of power, RPM, thrust
deduction, effective wake fraction and relative rotative efficiency; and additionally, what scale
effects are found from the computational methods.
4.2.3. Optimisation based on RANSE solvers

Optimisation based on RANSE predictions are especially time consuming, automatic
procedures are hampered due to the complex problem set-up and the grid generation
involved. The nature of the flow to be predicted, especially flow separation phenomena, does
not lend itself easily to a fully automated process. These “extra problems” – when being
compared with optimisations based on other techniques which will be mainly addressed in
WP5 – call for a dedicated study that includes:
 Application of RANSE-based optimisation, by deterministic and genetic algorithms.
 RANSE based multi-objective optimisation is investigated for simultaneously
minimising power and optimising wake quality.
Ship designers are very often faced with the problem of satisfying conflicting demands
between good wake qualities and limit the resistance increase. This design problem can be
solved based on accurate RANSE computations together with multi-objective optimisation
techniques. The formulation of the optimisation problem has been investigated, constraints
set, and extensive optimisations have been run using accurate RANSE solutions.
Verifications of the methods have been done by checking the required grid density,
convergence level and grid generation approach to deduce the correct trends with hull form
variation. As a validation the aftbody hull form of an initial design was requested to be
optimised by the optimisation systems developed, with respect to wake flow quality and
viscous resistance. The most promising design alternative is to be tested for validation.
The guidelines given in Sections 4.4, 4.5 and 4.6 are grouped according to these main tasks.
4.3. Application cases
4.3.1. Case 1-a: Resistance prediction in double-body flow
4.3.1.1. Test case

The application case presented here formed the subject of two VIRTUE workshops in January
and October 2007. It addresses double-body viscous flow computations for the KVLCC2
tanker, which is a standard test case of the CFD Workshops in ship viscous flows.

The main particulars of the KVLCC2 are described in the table below:
Model Ship
Lpp (Length between perpendiculars) 5.5172 m 320.0 m
B(Breadth) 1.0 m 58.0 m
T(Draft even keel condition) 0.3586 m 20.8 m
CB(Block coefficient) 0.81 0.81
2 2
S(Wetted surface area) 8.0838 m 27,194 m
Table 4-1: KVLCC2 geometry definition
6 9
The Reynolds number is Re=4.6x10 at model scale and Re=2.03 x 10 at full scale.
Main objectives with computations were to analyze and compare results of different methods
and study the origin of differences, thus helping us to drive guidance on the most promising
ways to achieve further improvement. Presented in this Application case is the influence of
turbulence models, grid topologies on the prediction of the viscous resistance (friction and
pressure components) and wake at model and full scale.
4.3.1.2. Overview of methods used

ECN, FLOWTECH and MARIN have used different numerical, modelling or meshing
techniques for the workshop that are summarised in the table below:
Methods ECN FLOWTECH/SSPA MARIN/IST
Pressure-
pressure correction direct coupling direct coupling
velocity coupling
overlapping multiblock
Grid type unstructured
structured structured
Wall functions no/yes no no/yes
Turbulence
SST, EASM, RSM SST, EASM SST, Menter
models
Table 4-2: KVLCC2 methods used
4.3.1.3. Grids
Meshing used for the first workshop was generated by individual participants. Table 4-3:
shows meshing from each participant that was used for comparisons of results at model
scale.
Code Grid Type Grid density at model scale
ECN O-O, 225x161x81
MARIN/IST H-O, 673x121x41
FLOWTECH/SSPA H-O, 141x121x65
Table 4-3: KVLCC2 grid description
IST created sequence of H-O grids for the second workshop ranging from 401x66x21 to
801x131x41nodes.

4.3.1.4. Resistance coefficients

One of the aims with the first workshop was to study the influence of the turbulence model.
 ECN used one Reynolds stress turbulence model (RSM), one explicit algebraic stress
model (EASM) and the two equation k-ω SST model.
 MARIN/IST used six eddy viscosity models including the k-ω SST model.
 FLOWTECH used the same EASM as ECN and k-ω SST model.
In total, 12 sets of calculations were thus presented at model scale. All three groups also
carried out full scale computations, but not all turbulence models were used in this case. In
total, eight such computations were performed.
The table summarizes the results of the model scale computations.
Turbulence
Id Code Grid Cpv CF CT
modelling
0 Exp. - - - - 4.06e-3
1 ECN SA O-O 225x161x81 5.88e-4 3.43e-3 4.02e-3
2 ECN SST O-O 225x161x81 6.75e-4 3.38e-3 4.05e-3
3 IECN EASM O-O 225x161x81 7.86e-4 3.33e-3 4.11e-3
4 ECN RSM-SSG O-O 225x161x81 8.07e-4 3.45e-3 4.26e-3
5 MARIN/IST MNT H-O 673x121x41 6.04e-4 3.45e-3 4.05e-3
6 MARIN/IST SA H-O 673x121x41 5.50e-4 3.35e-3 3.90e-3
7 MARIN/IST TNT H-O 673x121x41 6.37e-4 3.51e-3 4.15e-3
8 MARIN/IST SST H-O 673x121x41 6.58e-4 3.42e-3 4.08e-3
9 MARIN/IST SKL H-O 673x121x41 6.12e-4 3.24e-3 3.86e-3
10 MARIN/IST KSKL H-O 673x121x41 6.26e-4 3.28e-3 3.91e-3
11 FLOWTECH EASM H-O 141x121x65 8.48e-4 3.23e-3 4.08e-3
12 FLOWTECH SST H-O 141x121x65 8.27e-4 3.22e-3 4.05e-3
Table 4-4: KVLCC2 results at model scale
The mean value of 12 computed total resistance CT in model is 4.043e-3, which is strikingly
close to the measured CT 4.056e-3. All CT predicted were within an accuracy of 5% and 7 out
of 12 predictions within accuracy of 1% as compared with measurements. Although the total
resistance was quite similar in all results the distribution between the two resistance
components (viscous pressure, CPV and frictional resistance, CF) varied more. The CF varies
between 3.22e-3 and 3.51e-3, and the mean value is 3.365e-3; which is quite comparable to
the ITTC line, CF = 3.44e-3.
It is noted that CF depends a lot in MARIN/IST computations on turbulence models. This is a
consequence of the significantly different level of eddy-viscosity predicted by the several
turbulence models in the bow region. All the calculations were performed assuming fully-
turbulent flow, i.e. the turbulence model is supposed to handle transition. In the MARIN/IST
solutions, the SKL and KSKL models proposed by Menter exhibit the largest region of
“laminar flow” whereas the k-ω models present an unreasonably large level of eddy-viscosity
at the bow. In the latter case, the limiter applied to the production of k was not effective in the
H-O topology adopted. FLOWTECH predicted also lower C F, but this is due to neglect of bulb
in CF integration in the zonal approach that was used in these computations. The C PV shows a

considerably large variation (0.550e-3 to 0.848e-3), and SA model gives lower CP in both
ECN and MARIN/IST computation. Like in the Tokyo Workshop 2005 an increase in one
component was compensated by a decrease in the other. Previous experience suggests that
a coarse grid exaggerates CPV while CF is reduced.
Similar observation for resistance comparison can be made one the full scale case. The mean
CF from seven predictions is 1.48e-3 and it is well comparable with the ITTC line, C F=1.44e-3.
The change in form factor predicted at model scale (1+k=1.2) and at full scale (1+k=1.25) is
reasonable. It was an issue at 2000 Gothenburg Workshop due to almost two times bigger
form factor in full scale. However, these results must be interpreted carefully, because the
accuracy of the numerical prediction of the pressure resistance is strongly dependent on the
details of the calculations.
Id Code T.m. Grid Cpv CF CT

1 ECN SST-WF O-O 113x81x41 2.36e-4 1.48e-3 1.71e-3
2 ECN EASM-WF O-O 113x81x41 2.85e-4 1.51e-3 1.79e-3
3 ECN SST Low Re O-O 169x141x61 2.22e-4 1.52e-3 1.74e-3
4 MARIN/IST MNT H-O 673x141x41 1.97e-4 1.57e-3 1.77e-3
5 MARIN/IST SA H-O 673x141x41 1.78e-4 1.51e-3 1.69e-3
6 MARIN/IST TNT H-O 673x141x41 1.98e-4 1.55e-3 1.75e-3
7 FLOWTECH EASM H-O 141x121x65 5.61e-4 1.38e-3 1.94e-3
8 FLOWTECH SST H-O 141x121x65 5.42e-4 1.37e-3 1.97e-3
Table 4-5: KVLCC2 results at full scale
The variations due to numerical uncertainty, domain size…etc can yield variations of the
same order as the differences in the table. However, it can be noted a better control in
resistance prediction by turbulence models, grid topology and grid size and therefore smaller
scattered results are obtained between computations at full scale.
Another important finding is that the change in CF in model/full scale is consistency with ITTC
line. This represents a clear improvement in accuracy as compared with the Tokyo Workshop
in 2005.
4.3.1.5. Wake field

There are two sets of measurement for the KVLCC2, one from a towing tank and one from a
wind tunnel. Some discrepancies are observed between these two sets of wake
measurements as noted in the figure below. Considerable difference is found for 0.4 iso-
contour line while comparing towing tank with wind tunnel measurements.

Figure 4-1: KVLCC2 wake fields measurements
Below are results from ECN using three different turbulence models SST, EASM and RSM.
EASM model performs better in general. SST model indicated the hook shape, but not strong
enough. Results from MARIN/IST with six different one equation models showed a too
smooth wake. However, wake predictions were improved with empirical correction in these
models.
Figure 4-2: KVLCC2 wake fields computations – Turbulence modelling influence

The figure below shows the results at the plane X=0.9825*Lpp for the same code using the
SST turbulence model but with a H-O (dashed) and a O-O (solid) grid. The grids have
approximately the same number of cells but the different grid topology yields substantial local
variations of cell volume and shapes. This can to some extent explain differences in the outer
part of the wake. Other important differences appear in the inner part of the wake due to the
treatment of the ship contour.
Figure 4-3: KVLCC2 wake fields computations – Grid Type influence
The figure below shows the results at the plane X=0.9825*Lpp for two different codes using
the SST turbulence model and the same H-O grid. Different results come from differences in
numerical techniques and model implementation. With this in mind, the results are
surprisingly similar.
Figure 4-4: KVLCC2 wake fields computations at model scale – k-ω SST results in 2 codes

The figure below shows the results at the plane X=0.9825*Lpp for two different codes using
the SST turbulence model and the same H-O grid for full scale wakes. The results showed a
reduction of the Reynolds stresses with at least one order of magnitude. It is noticeable how
the “hooks” have been reduced.
Figure 4-5: KVLCC2 wake fields computations at full scale – k-ω SST results in 2 codes
4.3.1.6. Conclusions
 Resistance
Resistance predictions in model and full scale are now achievable. The total resistance is well
predicted within an accuracy of 5% in model scale. The full scale resistance prediction can be
better. The reduction in CF between the model and full scale results is reasonable, compared
with the ITTC -57 formula.
 Wake, model scale

The wake flow can be predicted very accurately with more advanced turbulence models
(EASM and RSM) or k-ω SST model with curvature correction. The predicted flow quantities
(for instance, wake, velocity, pressure and limiting streamlines) can be used for propeller
design, wake improving device and evaluation of hull form improvement.
 Wake, full scale

Reliable full scale wake computation is possible. The boundary layer thickness change and
disappearance of “hooks” with increased Reynolds number already observed in the
experiments suggest that full scale computations could be handled by turbulence models.
Indeed there is a smaller difference between advanced models and eddy viscosity models at
full scale compared with model scale. This is consistent with the reduction of at least one
order of magnitude in the Reynolds stresses from model to full scale.

4.3.2. Case 1-b: Resistance prediction with free surface (free sinkage
and trim)
4.3.2.1. Test case

The application case discussed here formed the subject of a VIRTUE workshop in October
2007. It addresses viscous flow computations with free surface and free trim and sinkage for
the ‘Hamburg Test Case’, a container ship for which extensive experimental data have been
measured in the past by HSVA.
The main objectives of these computations were:
 To judge the efficiency and accuracy of the different methods, for the model-scale
case;
 To compare the predicted wave patterns, viscous flow fields and resistance
components, for the full-scale case, and to assess the physical scale effects;
 To assess the ability of the methods to compute the wave resistance and its scale
effect, and the scaling of the viscous resistance; and to compare with model-ship
extrapolation methods used in towing-tank work.
Therefore, the application has some complications due to the need to compute both for model
scale (1:24) and for full scale, to include the dynamic trim and sinkage, and to reach a
numerical accuracy sufficient to determine the scale effects with some certainty.
No appendages and rudder have been taken into account, and computations addressed an
unpropelled (towed) condition. The static drafts were Tf = 0.383 m model / 9.2 m ship; Ta =
0.429 m model / 10.3 m ship. In the tests and in the free-surface computations, the model
was free to assume a dynamic sinkage and trim. The Froude number was Fn = 0.238, the
Reynolds number Re = 11.77E06 for model scale, Re = 1.2E09 for full scale.
4.3.2.2. Overview of methods used

A variety of methods has been used in this workshop. The following table attempts to
summarise some of the main features.
Aspect HSVA ECN Principia MARIN
pressure- pressure pressure artificial direct coupling
velocity coupling correction correction compressibility
grid type quasi-structured unstructured multiblock multiblock
structured structured
grid generator ICEM Hexpress ? Marin tools
wall functions yes yes yes no
free-surface capturing, VOF capturing, VOF- tracking free-surface
modelling like fitting
free-surface time integration time integration time integration steady iterative
solution method
trim/sinkage body motion eq. body motion eq. body motion eq. iterative
+ artificial
damping
Table 4-6: Simulation methods used by different participants

4.3.2.3. Grids
The participants all used grids consisting of 2 to 2.8 million cells. However, there are
important differences. On the one hand, capturing codes not only require a gridding in the air
domain, but also a refined band in the area where the free surface is expected, to reduce the
errors in the capturing algorithms. This means at least a doubling of the number of cells. On
the other hand, the surface-fitting method of Marin used no wall functions. Use of wall
functions would reduce the number of cells by some 20%.
Also, there were important differences in grid topologies and density distributions.
The grids ranged from single-block structured to fully unstructured. All participants generated
the grid using their favourite tools and according to their experience. Also the domain sizes
differed, but in general were larger than would be used for computations without free surface.
4.3.2.4. Trim and sinkage
For computing trim and sinkage different approaches have been used. Most
participants used time integration for the RANS solution including a coupling with body motion
equations. Below we show an example of the time history of the trim.
0.8 FS mesh 2
MS mesh 2
0.6
0.4
0.2
trim (°)
measurement
0
-0.2
-0.4
-0.6
-0.8
0 1 2 3 4 5
t V/Lpp
Figure 4-6: Time history of the trim
The graph shows that the coupling with the body motion equations can lead to a pitching
oscillation that decays only slowly and substantially prolongs the computation. In some cases
this indirectly led to trouble with sloshing in the domain, as reflected waves came back to the
ship within the longer integration time. It was found that by deviating from the physical body
motion equations these trim and sinkage oscillations could be largely controlled.
Nevertheless, including free trim and sinkage tends to ask for longer integration times than for
a fixed trim and sinkage. The physical simulation times used varied between T.V/Lpp = 5 and
33.
Marin used a steady iterative solution for the free-surface problem, and also an iterative
adjustment of the trim and sinkage, with corrections based on zero-speed (hydrostatic)
geometric quantities only. This simple method converged monotonously and quickly.
All computed model-scale trim and sinkage are essentially within the experimental range. It
shows that computing trim and sinkage for a case like this is an easy task. Also a free-surface

potential-flow computation gave a result in the experimental range. No significant scale

effects on trim and sinkage were computed.
4.3.2.5. Wave pattern
The computed wave patterns were mostly rather similar. Two sets of results were very close
together (ECN’s and Marin’s) as shown in Figure 4-7 even while they were obtained by
completely different methods. This suggests that numerics can be well controlled today.
Figure 4-7: Wave comparisons on model scale

0.004 model HTC at Fn = 0.238

longitudinal wave cut at y/Lpp=0.184
0.002
(z-zwl)/Lpp
-0.002
0.006
-0.004
x/Lpp
-1.5 -1 -0.5 0 0.5 1 1.5
x/Lpp
AP FP
FP
AP
0.004 model HTC at Fn = 0.238
longitudinal wave cut at y/Lpp=0.418
(z-zwl)/Lpp
0.002
-0.002
-1.5 -1 -0.5 0 0.5 1

x/Lpp
Figure 4-8: Wave cuts for Hamburg test case application on two longitudinal wave cuts.
Top figure y/Lpp=0.184 and bottom figure y/Lpp=0.418
1
Black = experiment, coloured lines are computed results from different participants .
1
Orange curve results have been obtained after the workshop, after a substantial change in the code

Clearly, in the wave cut close to the hull almost all results are close together, with minimal
differences in phase but some in amplitude. The shorter diverging bow waves near x/L = 0.3-
0.5 show more variation. In the distant wave cut the variations are substantially larger, as
numerical errors in wave propagation properties accumulate.
Due to the same cause, the grid dependence of these wave cuts is larger at larger distances.
Figure 4-9 shows the grid dependence for one participant, due to halving the grid spacing in
longitudinal and vertical direction. The grid dependence differed
Green: coarse quite a bit between
grid, model scale
Red: fine grid, model scale.
participants and was often much larger than this for the outer wave cut.
Wavecut y=0.418L PP
Green: coarse grid, model scale

Red: fine grid, model scale.
Wavecut y=0.184LPP
0.002
0.002
(z-zwl)/Lpp
(z-zwl)/Lpp
0
0
-0.002
-0.002
-0.004 -0.004
-1 0 1 -1 0 1
x/Lpp
AP FP x/Lpp
AP FP
Figure 4-9: Grid dependence of wave cuts at y/Lpp=0.184 (left) and 0.418 (right), for MARIN.
red = fine grid, green = coarser grid. Refinement ratio = 4.0.
4.3.2.6. Wave pattern scale effects
Figure 4-10: Model (right) and full-scale (left) wave pattern predicted by ECN.

Figure 4-11: Computed wave cuts at centreline aft of the stern, for model and full scale.
Although there is no experimental information, we expect the wave pattern scale effects to be
confined to the stern area for a rather slender ship like this, and this was actually computed
by two participants; with some additional differences in the bow wave system for one other.
The scale effect computed, also in some other cases, is that the stern wave system is
significantly higher at full scale than at model scale. To really estimate the wave pattern scale
effects, a good control of the numerics is needed because these are often relative small
effects.
4.3.2.7. Wake field

It was found that for this ship, the influence of the free surface on the wake field is limited.
This is not a general conclusion however, as it must depend on hull form, speed, afterbody
shape etc.
In this application, there appeared to be a limited effect of the turbulence model on the wake
field, as illustrated below. Figure 4-12 shows one comparison of a computed wake field and
the experimental field.

Resistance
Remarkably, while the flow field predictions seem accurate and in good agreement
with the data, the resistance predictions differed from the experimental result by up to 10%.
For full scale the difference in computed resistance were smaller.
Grid dependence of the resistance prediction has not been thoroughly checked in this
application, but fine/coarse differences were of the order of 2 % for most participants.
Resistance scale effects have been computed and compared with the usual model-
to-ship extrapolation assumptions. It turned out that this is a quite demanding use of the
computations, asking for a very high numerical accuracy.
Figure 4-12: Turbulence sensitivity in the wake field
4.3.2.8. Conclusions
Many conclusions could be drawn from this large effort. We repeat some.
 The computed wave patterns of several participants were in good agreement
mutually and with the experimental data. Some still had local deviations due to visible
grid effects or other numerical issues. Grid dependence was reasonably limited in the
nearby wave cut, where it acts mostly on the bow wave system; but for some it is still
quite pronounced in the more distant wave cuts, at which the wave amplitude can
differ by a factor of 3 between coarse and fine grids.
 Predicted scale effects on the wave pattern are confined to the stern wave system,
which is found to be about 40% higher at full scale than at model scale. One
participant also got a substantial increase of the bow wave system, which is believed
to be due to numerics rather than physics.
 Double body and free surface flow computations performed on grids of similar density
showed no real influence of the free surface flow on the wake field for this case,
except somewhat more upflow due to the presence of a stern wave.
 The wake field computed by two participants are quite similar and in good agreement
with experiments. Another participant got reasonable trends, but their use of the K-ε
model may prevent a better agreement. Contrary to what was observed on other hull
shapes, it appears that in this case the influence of the turbulence modelling
(between K--SST and EASM) is very limited, even at model scale.
 The transient computation of trim and sinkage resulted in an oscillating approach to
steady state, but one participant deviates from the transient physics by successfully
adding a damping term that yields an almost monotonous approach to steady state.
One participant used an iterative method without time-dependence which avoids the
problem altogether.
 Two of the model-scale resistance predictions significantly underestimated the
experimental value; one was close to the experimental value. The choice of the
turbulence model explained part of the discrepancy. The full-scale resistance
predictions were much closer together.

 While capturing methods are generally more flexible than free-surface fitting methods,
in this application they showed disadvantages: a larger number of grid cells to resolve
the interface and the air flow, rather slow oscillatory convergence to steady state (in
trim, sinkage and resistance), larger grid dependence, and much more computation
time than the surface fitting method.
4.4. Prediction of Resistance, wake and wave pattern

4.4.1. Introduction
The present section makes a distinction between double-body flow computations, and free-
surface flow computations. Subsection 4.4.3 discusses recommendations for computing a
double-body flow. Subsection 4.4.4 gives recommendations for free-surface flow
computations, as far as these are different from those for double-body flow. There are
additional subsections dealing with the incorporation of dynamic sinkage and trim; and the
particular recommendations for computing a wake field.
Double-body flow computations disregard the effect of the free water surface, and instead
impose a symmetry condition at the still water surface. This can be considered as an
‘Underlying Flow Regime’ that includes the essence of the viscous flow around the hull
without the complications of a free surface.
While in view of the development of solution methods for viscous flow with free surface some
might think double-body flow computations to be obsolete, they are still very relevant for
several reasons:
 Wave effects on the afterbody flow can be very speed-dependent (due to the phase
of the waves relative to the aft shoulder and stern), and by disregarding them, double-
body flow computations provide a clearer impression of the viscous flow and its
relation with the hull form. Even so, for a final assessment the free-surface effect
often should be included.
 In particular at low Froude numbers, the wave effects on the viscous flow play a role
primarily near the water surface, and the double-body flow is quite accurate for most
of the flow field. In VIRTUE, comparative computations have shown that for the
Hamburg Test Case at Fn=0.238, there was very little free-surface effect on the wake
field.
 Many current computational methods for free-surface viscous flow require a refined
grid in the region around the free surface for an acceptable accuracy; and often a grid
also in the air region above the wave surface. Thereby, grid requirements for free-
surface flow are harder to meet than for double-body flow. Besides, free-surface flows
usually impose additional restrictions on time steps, simulation time and start up
procedure. Thus incorporating the free surface makes the computation more time-
consuming and may not help the accuracy of the prediction of some flow features.
 In VIRTUE, a procedure has been developed to separate resistance components and
determine their scaling using both free-surface and double-body flow computations.
The latter type of calculations is also required to check the common extrapolation
methods that split the ship resistance in viscous, C V, and wake, CW, resistance
components. The wave component is assumed to be dependent only on the Froude
number whereas the viscous resistance is assumed to be only a function of the
Reynolds number. Therefore, CV, is estimated for zero Froude number, i.e. double-
model flow.
On the other hand, some care may need to be taken in double-body flow computations for
ships with a bulbous bow close to (or piercing) the still-water surface. A double-body flow
around such a bow would often differ strongly from a free-surface flow, e.g. showing a large

flow separation at the bulb that affects the resistance and the viscous flow downstream of it
substantially. In such cases, a local adjustment of the bulb shape (or hull attitude) to prevent
large deviation of the flow around the bulb can be the best option.
4.4.2. Target Flow Variables
The resistance force of a ship is composed of two contributions: friction resistance and
pressure resistance. Both quantities can be determined from integrations over the ship
surface: shear-stress at the wall for the first component and pressure for the latter. Total
resistance may also be obtained from a momentum balance.
Pressure is one of the dependent variables of RANSE, whereas the shear-stress at the

wall w is related to the velocity field. Its determination is intimately linked to the boundary
conditions: if the no slip condition is directly imposed at the wall, w is obtained from the
normal velocity derivative at the wall; if wall functions are applied as boundary c
is computed from the analytical profiles assumed for the near-wall region using the velocity
magnitude (and possibly the turbulence kinetic energy) at the nearest location to the wall,
Conventionally, double-body flow computations solve for the hydrodynamic pressure (relative
pressure) contribution only, i.e. hydrostatic pressure (-gz, with z=0 at the still-water surface)
is used as the reference pressure and so the acceleration of gravity drops from the
momentum equations. In free-surface flow computations, different formulations are used
dependent on the method, with or without incorporation of the hydrostatic pressure.
For full-formed, relatively slow vessels (e.g. tankers), the viscous pressure resistance
component is extremely sensitive to all sorts of parameters, and a careful check of that
sensitivity needs to be made if quantitative accuracy of the total resistance prediction is
required. In VIRTUE, for the KVLCC2 test case significant effects have been observed of grid
density, size of the computational domain, the type of the boundary conditions, and the
discretisation of those boundary conditions. The recommendations below address these
effects.
The wave elevation, for steady flow, is directly related with the hydrodynamic pressure at the
surface as found from a free-surface flow computation. In order to be able to assess the wave
making, the wave spectrum, and possibilities to improve the hull form, it is important that the
wave pattern is accurately computed up to some distance away from the hull, e.g. to 0.3 – 0.5
Lpp off the centreline (LPP is the distance between perpendiculars).
Sinkage and trim are here understood as the change of attitude of the ship at speed due to
the hydrodynamic pressure distribution on the hull and possibly the effect of the propulsion or
towing force. Sinkage is often defined as the downward vertical movement of the hull
measured at midship, while trim is the angle of rotation. The sign of the trim depends on the
coordinate system and can differ between methods. Other definitions of sinkage are
sometimes used.
The wake field is commonly defined as the velocity field (axial, tangential and radial
components) at the location of the propeller: Nominal wake is the velocity field without
propeller present and total wake is the velocity field including the propeller action.
The total wake differs from the nominal wake by:
 the propeller induction;
 the effect of the propeller action on the flow around the hull,
However, the split between these two is not very clear-cut. E.g. HSVA, in their ‘method A’
(see section 4.5), use a different split than others.

Effective wake is the total wake, minus the velocity induced by the propeller; so it includes the
second contribution but not the first. Thereby it depends on the split between both effects and
on the propeller model. The objective of the effective wake definition, however, is to
approximate the propeller operation behind the ship by that in an assumed infinite domain
with an inflow equal to the effective wake field.
4.4.3. Double-model ship hull
4.4.3.1. Selection of the computational domain

Double-model flow calculations have three “natural” boundaries: the ship surface, the
symmetry plane of the ship and the still water plane (also a symmetry plane). The definition of
a computational domain requires the specification of “artificial” boundaries at a certain
distance from the ship, where approximate boundary conditions must be imposed. The shape
of the domain is usually a consequence of the grid topology adopted (O-O, C-O, H-O,…), but
it always includes an inlet, an outlet and an external boundary, which require different types of
boundary conditions (discussed below).
The size of the computational domain is related to the approximate boundary conditions
adopted at the “artificial” boundaries. However, with the most usual choices (undisturbed flow
or potential flow solution) it may have a significant influence in the prediction of the viscous
resistance coefficients. Figure 4-13 and Figure 4-14 present an illustration of this influence for
the calculation of the flow around the KVLCC2 tanker at model scale Reynolds number. The
calculations were performed with the RANSE solver PARNASSOS [?] in a H-O single-block
grid with the outlet (xout) and external (rext) boundaries at different distances from the ship
surface (LPP is the distance between perpendiculars). The plots present the friction and
pressure resistance coefficients for two types of boundary conditions: tangential velocity and
pressure imposed from the potential flow solution (Figure 4-13) and uniform (undisturbed)
flow at the external boundary (Figure 4-14). The estimated numerical uncertainty of the data
plotted in these figures is below 1%.
3.45 0.75 xout=1.25LPP

xout=1.3LPP
xout=1.5LPP
0.7 xout=1.88LPP
xout=2.5LPP
xout=4LPP
3.4 0.65
0.6
CF10 3
CP10 3
xout=1.25LPP
xout=1.3LPP
3.35 xout=1.5LPP 0.55
xout=1.88LPP
xout=2.5LPP
xout=4LPP 0.5
3.3 0.45
0.4
3.25 0.35
0 0.5 1 1.5 2 0 0.5 1 1.5 2
rext/LPP rext/LPP
Figure 4-13: Friction and pressure resistance coefficients of the KVLCC2 tanker at model
6
scale Reynolds number, 5×10 . Potential flow boundary conditions at the external boundary.

The results presented in Figure 4-13 and Figure 4-14 suggest the following
recommendations:
 When undisturbed flow conditions are adopted as boundary conditions, the external
boundary condition location should be at least two ship lengths away from the ship
surface to have Cp unaffected. For the full hull form considered, a boundary at 1 Lpp
gives a 3% error in Cp.
 Imposing potential flow conditions at the external boundary enables the reduction of
the size of the computational domain. However, to obtain a negligible effect in the
pressure resistance coefficient (<0.5%) the external boundary should be at least one
ship length away from the ship surface.
For more slender hull forms than the KVLCC2, the viscous pressure resistance is a
smaller contribution and the importance of a large domain size for a total resistance
prediction is supposed to be less.
3.45 0.75 xout=1.25LPP

xout=1.3LPP
xout=1.5LPP
0.7 xout=1.88LPP
xout=2.5LPP
xout=4LPP
3.4 0.65
0.6
CF10 3
CP10 3
xout=1.25LPP
xout=1.3LPP
3.35 xout=1.5LPP 0.55
xout=1.88LPP
xout=2.5LPP
xout=4LPP 0.5
3.3 0.45
0.4
3.25 0.35
0 0.5 1 1.5 2 0 0.5 1 1.5 2
rext/LPP rext/LPP
Figure 4-14: Friction and pressure resistance coefficients of the KVLCC2 tanker at model
6
scale Reynolds number, 5×10 . Uniform (undisturbed) flow imposed at the external boundary.
4.4.3.2. Boundary conditions

Small details in the numerical implementation of the symmetry conditions may have a drastic
effect on the accuracy of the pressure resistance coefficient. The use of first-order schemes
may deteriorate severely the accuracy of the numerical prediction of the pressure resistance
coefficient.

Frame 001  23 May 2006 
0.9
0.8
0.7
Cvp *1000
0.6
0.5
0.4
0.3
0.2
0.1
00 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
h_i/h_1
Figure 4-15: Grid dependence of predicted viscous pressure resistance, for model scale (full
lines) and full scale (dashed lines) using different implementations of the symmetry conditions
st nd rd
at the still water surface (Delta markers: 1 order; Gradient markers: 2 order; circles: 3
order; squares: Indirect formulation)
This effect may be dependent on the formulation adopted, but it is important to check it
carefully. An example taken from [ICHD] is presented in Figure 4-15, showing the
convergence of the pressure resistance coefficient for different implementations of the
symmetry conditions at the still water surface that can produce different resistance levels and
scale effects in a given grid.
Figure 4-16: Computed limiting streamlines with and without wall functions for the KVLCC2
tanker at model scale Reynolds number. Black - no slip; Red – Wall functions

Figure 4-17: Computed limiting streamlines with wall functions (top) and without (bottom) for
the Virtue tanker at model scale Reynolds number.
 Wall functions and direct no-slip conditions produce similar results in the absence of
flow separation and strong pressure gradients. However, at regions of flow separation
the differences between the two approaches may be significant. Logically, in these
situations the use of wall functions is questionable. Two examples are presented in
Figure 4-16 and Figure 4-17 for the KVLCC2 and Virtue tankers. In both cases, there
is an effect on the size of the separation bubble below the shaft and on the location of
the longitudinal vortex.
 Grid refinement studies with wall functions should preserve the same distance to the
wall for the grid nodes where the boundary conditions are applied (wall functions are
part of the mathematical model).
 ''Automatic'' wall functions along a route outlined by Menter [“automatic wall
functions”] are numerically more robust than wall functions based only on the log law.
4.4.3.3. Turbulence model
 The friction resistance coefficient is more dependent on the choice of the turbulence
model than the pressure resistance coefficient. An example taken from [JMST] is
presented in Figure 4-18 for the flow around the KVLCC2 at model scale Reynolds
number.

HO(), Mt HO(), k-kL HO(), TNT HO(), Mt HO(), k-kL HO(), TNT
3.6 HO(), Mt HO(), k-kL HO(), TNT 0.7 HO(), Mt HO(), k-kL HO(), TNT
HO(), Mt HO(), k-kL HO(), TNT HO(), Mt HO(), k-kL HO(), TNT
3.5 0.6
CF103
CP103
3.4 0.5
3.3 0.4
3.2 0.3
1 1.25 1.5 1.75 2 1 1.25 1.5 1.75 2
hi/h1 hi/h1
Figure 4-18: Convergence with the grid refinement of the friction and pressure resistance
coefficients of the KVLCC2 tanker computed with three eddy-viscosity turbulence models at
model scale Reynolds number.
 Due to the  wall boundary condition, k- models require smaller near-wall grid line
spacing for the direct application of the no-slip condition than other eddy-viscosity
turbulence models to obtain the same accuracy in the friction resistance coefficient.
This is seen in Fig.6 in the larger slope of the lines for the TNT-model, which is a k-
model.
 For a given grid, the numerical accuracy of the flow field may depend on the selected
turbulence model. Therefore, the numerical uncertainty for accuracy estimated for a
given turbulence model is not necessarily valid for a different turbulence model, see
Figure 4-18.
4.4.3.4. Numerical aspects

 In body-conforming grids generated with Hexpress, the total number of points
depends on the hull complexity. Typical numbers in the near-wall region are 10 layers
in the viscous layer for wall functions boundary conditions and 30 for the direct
application of the no-slip condition.
 If H-O grids are used, these are typically not matching the bow and stern contours. It
is recommended to contract the grid towards the bow and stern to limit the local
inaccuracy. In systematic grid refinement studies, the description of the contours can
show up as some scatter in the resistance coefficients, see Figure 4-18.
 Cpx103: -4 -3.2 -2.4 -1.6 -0.8 0 0.8 1.6 2.4 3.2 4
Figure 4-19: Comparison of pressure distributions on the ship surface for the KVLCC2 tanker
computed with two eddy-viscosity turbulence models at model scale Reynolds number.

 The friction resistance coefficient features a significantly smaller grid dependency

than the pressure resistance coefficient. Performing graphical comparisons of surface
pressure isolines (isobars) is not sufficient to address the accuracy of the numerical
determination of the pressure resistance coefficient (it can even be misleading). An
example is presented in Figure 4-19, with data taken from the study reported in
[JMST]. The left picture shows a direct comparison of isolines from two turbulence
models, whereas the right picture includes the difference between both predictions
divided by the pressure resistance coefficient as the reference value.
 Convergence criteria have to be selected carefully. The iterative error may be 2
orders of magnitude (3 if convergence is based on root mean square values) larger
than the maximum differences between consecutive iterations or maximum residuals,
[ONR2006].
 In the context of CFD Uncertainty Analysis, a negligible influence of the iterative error
on the estimation of the discretisation error requires iterative errors 2 orders of
magnitude smaller than the discretisation error. Therefore, fine grids require more
demanding iterative convergence errors than coarse grids, [ONR2006].
 Performing single grid studies to evaluate scale effects may be completely
misleading. Tendencies observed in coarse grids may differ significantly from those
obtained when the grids are refined.
4.4.3.5. Post-processing
 Numerical details of the integration of pressure and shear-stress at the wall may have
a significant effect in the accuracy of the predictions, especially for grid topologies
that do not conform to the ship contour. In such cases, the exact definition of the ship
contour must be included in the integration procedure; otherwise there is a significant
loss of accuracy.

4.4.4. Free surface

In this section, recommendations are given as far as they differ from those for double-body
flow only.
In agreement with model testing, wave resistance is defined as the difference between the
computed total resistance with and without free surface. This requires two computations (for
each Fn and Rn): one for double-body flow, one for free-surface flow. It is important to use
very similar grid, turbulence model, input parameters etc in both computations, because
otherwise the wave resistance deduced may be excessively polluted by numerical errors.
4.4.4.1. Selection of the computational domain
 Generally, for free surface computations, it can be useful to choose a rectangular

computation domain. The extension of the computational domain in the streamwise
direction depends on the formulation and boundary conditions adopted.
 Methods that solve the problem in an unsteady formulation usually require numerical
wave damping downstream of the ship; therefore they use an extended domain with
the outlet boundary e.g. 3 to 5 Lpp (depending on the wave damping method
adopted) behind AP. The lateral vertical border must be sufficiently far from the
centre plane to guarantee that the entire Kelvin wedge ends up at the aft boundary.
Therefore, the domain becomes much longer and wider than for double-body flow.
 Methods in steady formulation can also have wave reflections at the outlet boundary,
but these do not travel back into the domain and need no damping. For these
methods, imposing a boundary condition at the lateral boundary found from a free-
surface potential flow computation removes reflections sufficiently well for most of the
wave system. Again the lateral boundary is preferably chosen such that the entire
Kelvin wedge ends up at the aft boundary. As a result, the domain dimensions are
comparable with those for double-body flow.
4.4.4.2. Boundary conditions

 If free-surface boundary conditions are imposed (in free-surface tracking and fitting
methods) it is essential that no first-order discretisation errors are present anywhere,
as these immediately lead to an unacceptable wave damping. Similar arguments are
applicable to interface-capturing methods.
4.4.4.3. Turbulence model
 In the ship applications considered, no significant effect of the turbulence model on

the wave pattern has been observed.
 Depending on the code, numerical robustness of free-surface flow calculations may
be affected by the choice of turbulence model. In COMET, eddy-viscosity models
tend to be more robust than Reynolds-stress models for this type of applications.
However, the Explicit Algebraic Stress Model (EASM) included in ISIS-CFD is as
robust as eddy-viscosity models for double-body and free-surface flows

4.4.4.4. Numerical aspects
 For free-surface viscous flow computations, a grid density must be chosen that is
sufficient for resolving both the hull form and viscous-flow details, and the wave
pattern (with accurate wave propagation). These are two independent aspects. The
required grid density for good wave propagation needs to be retained up to a distance
from the hull where a good wave pattern prediction is desired.
Figure 4-20: Illustration of the influence of the grid density on the Kelvin wedge angle.
 It is important to be precisely aware of the properties of the method used regarding

wave propagation. E.g. capturing methods on coarse grids or with inadequate
capturing discretisation scheme give interface smearing, which results in wave
damping and uncertainty in pressure integration over the hull. For one free-surface
fitting method, it has been observed that a large cell size in direction normal to wave
surface caused numerical dispersion, observed as a change of the Kelvin wedge
angle. In addition, too strongly increasing cell size in wall-normal direction towards
the outer boundary can lead to too large mesh size and can thereby cause numerical
damping.
 The numerical damping and dispersion can be determined analytically in some cases,
or numerically, in a simple, linearised (quasi) 2D test case, such as waves due to a
pressure distribution on the water surface. [Ref. Raven / van der Ploeg / Starke,
2004]. Such efforts are justified and recommendable, since the respective
conclusions are generally representative for more complex maritime applications.
 For one particular surface fitting method in which the numerical dispersion and
damping for such a 2D test case were determined analytically to be of third order,
about 30 nodes per wavelength are required for an accurate computation. For other
methods which have a larger numerical dispersion and damping, 50 cells per
wavelength are mentioned as a guideline. Hence the required grid density increases
with a decrease of the Froude number, and can be very different from the grid density
required for double-body flow.
 The wave lengths that determine the grid density in principle are widely variable,
dependent on wave propagation direction (divergence angle θ), being proportional to
2
cos θ in the far field. Evidently it is not possible to retain a density of 50 cells per
wavelength for all visible wave components, and an upper limit of θ is to be chosen.

 For viscous free surface flow computations using the interface-capturing method in
which an additional transport equation for the volume fraction between water and air
is solved, high resolution grid is required in the forebody region in order to resolve the
bow wave system and the possible wave breaking. Fine grid in aftbody region is
important for a correct prediction of wake distribution. In addition to the high
resolution of the hull grid with the required number of cells per wave length
(depending on the accuracy of the scheme), fine grid (about 15 dense layers each)
above and beneath the still water line is required. Significant predictive improvements
are obtained for specific high-resolution approaches. For free-surface flow
computations with ISIS_CFD it is recommended to use at least 15 points from crest to
trough and 40 points per wave length.
 A steady free surface flow can be simulated with an unsteady model. Then the
transient free surface flow needs to be correctly simulated. The velocity of the ship is
introduced progressively from the start of the calculation (a zero velocity of the ship)
to its maximal velocity, in order to avoid a too large deformation of the free surface at
the beginning of the transient flow.
 The time step needs to be adapted with respect to CFL criteria (including velocity and
size of cells), considering the numerical scheme used in the model (explicit,
implicit…).
 Concerning the total duration of the simulation, the ship must cover a distance at
least equal to 5 to 15 ship lengths (according to the results obtained in the VIRTUE
Workshops) whatever the forward speed.
 In the transient approach for RANS free surface computations oscillations are
observed in the computed pressure resistance when the computations are started
with the final speed at the inlet. As long as the amplitude of the persistent oscillations
is less than 5% of the mean value of the last timely 1000 iterations and the residual is
below the prescribed value, the mean value can be taken as the ‘representative’
pressure resistance.
 Several topologies are possible to build a structured multi-blocks mesh. The O-O
topology is quite easy to handle and it allows getting a grid that conforms the ship
contour and has non-distorted cells around the hull. But its major drawback regarding
the free surface simulation, is that the gridlines are not flow aligned, on the contrary of
an H topology aligned in the longitudinal flow direction. Considering a free surface
tracking method such as the VOF model, the O-O topology has been observed to
lead to an unphysical numerical wave field which tends to fit the curved O gridlines,
whatever the grid density considered. This topology is not a good choice for waves
modelling.
 Grid dependence of free-surface viscous-flow computations needs to be checked not
only for a hull wave profile, since errors in wave propagation properties (numerical
dispersion and damping) are hardly observed in the hull wave profile. A comparison
of longitudinal wave cuts at a distance, and a centreline wave cut aft of the ship, is
required.
4.4.4.6. Sinkage and trim

 Including dynamic trim and sinkage is important for the resistance. However, for all
medium-Fn applications done so far, viscous effects on these were negligible.

Therefore, often trim and sinkage can be found from potential-flow computation and
do not introduce additional scale effects.
Figure 4-21: Pitching oscillations due to startup
 To include trim and sinkage in the computation, it is possible to couple the RANS
solution with the body motion equations and solve for the time-dependent behaviour
until a steady result has been obtained. Typically, this introduces pitching/heaving
oscillations due to the startup, and these delay the convergence to a steady result. If
only the steady result is the objective, it is recommended not to try to model the time-
dependent physics. Rather, one should avoid the pitching and heaving oscillations, by
modifying the body motion equations (e.g. by introducing artificial damping terms), or
by using static equilibrium equations and iterative adjustment. Examples of
calculations with and without damping in the body motions equations are given in
Figure 4-21 and Figure 4-22.

Figure 4-22: Suppression of pitching oscillations due to artificial damping terms in body
motion equations (CNRS)
4.4.4.7. Wake flow

The wake field behind ships can be accurately predicted provided a careful setup of the
computational model and grid is done.
Geometry definition: To capture the flow details in the wake it is important to have an accurate
hull definition. All details ranging from the main characteristics of the bare hull (shape and
smoothness) down to the small details of the shape of the propeller hub are important; not
only the appendages upstream of the propeller should be considered, but also appendages in
the near region behind and above the propeller.
Computational model: The influence from the free-surface on the wake is small in most cases
and can be treated as a symmetry plane. Main exceptions are ships with relatively large stern
wave due to stern geometry, Froude number or small draft. The effect of the dynamic sinkage
and trim is usually small.
The restrictions for the computational domain size are normally determined by the
requirements for the resistance prediction rather than the wake prediction.
The choice of turbulence model is important for the wake prediction. The two most usual
choices are the EASM model and the k- SST eddy-viscosity model. In wake flows with
strong longitudinal vorticity, the EASM model is able to capture the turbulence anisotropy,
leading to an increase of the longitudinal vorticity compared to the SST model, which is in
better agreement with available experiments. Nonetheless, the k- SST eddy-viscosity model
is also a standard choice and there are empirical corrections available to improve its
performance in flows with strong longitudinal vorticity.
Grid requirements are to a large extent program dependent when it comes to topology, cell
aspect ratio, cell size jumps etc. Common for all methods are requirements for the resolution
of the boundary layer region, requiring about 30 cells in the normal direction. However, this
number depends on the type of boundary conditions applied at the ship surface and the
numerical uncertainty may also be affected by the choice of turbulence model. The use of wall
functions reduces significantly the number of grid nodes to resolve the boundary-layer, but it
affects the predicted flow field at the propeller plane, i.e. in the near-wake. Provided that the

cells are concentrated to the region close to the hull, in particular around the stern region,
about 2 million cells should provide a good wake prediction. Local grid refinement in the stern
region is a useful technique to increase the accuracy with a limited number of cells.
It has been found that, dependent on the case, there can be significant differences in the
wake field predicted with wall functions and without. For a full hull form, this was caused by
differences in the extent of a flow separation bubble just upstream of the stern contour, as
illustrated in Figure 4-17. In such cases, the solution without wall functions is better, provided
that it has sufficient resolution.
As mentioned above, the wake field can be significantly affected by the modelling of the
propeller shaft and hub. If the hull is cut off and closed at the location of the shaft, and the
computation resolves the recirculation region aft of it, this can have a serious effect on the
wake field, also outside that recirculation region. A proper discretisation of the local geometry
with H-O grids is very hard to achieve resulting in a non-negligible numerical uncertainty for
the near-wake predictions. Body-fitted grids with local refinement lead to a more accurate
wake distribution across the whole propeller disk. However, the quantification of the numerical
uncertainty in such grids is not trivial.
The proper modelling is that which corresponds with the desired situation, i.e. a streamlined
cap (for comparison with nominal wake measurements) or the propeller hub, for the actual
situation. (Illustration: a picture from T132).
The wake can be accurately predicted at model scale. The same is true for full scale provided
that the grid is redesigned to maintain the resolution of the thinner boundary layer. However, it
should be kept in mind that the experience from full scale is more limited. This is also true for
separated and unsteady flows. In these cases the use of RANSE becomes questionable,
because its purpose is to obtain a statistically steady solution with all the effects of
fluctuations included in the turbulence model.
4.5. Hull / Propeller Interaction
4.5.1. Introduction
As defined in Section 2.3.1.2, the computation of the flow around the hull without any
propeller modelling provides the resistance and the nominal wake field. A propeller behind the
hull changes the flow field and the pressure distribution on the hull, and therefore there is an
interaction between hull and propeller that can be represented in the computation in various
ways. The present section discusses the modelling and implementation, and gives provisional
guidelines.
4.5.2. Target flow variables
The objective is to obtain the flow field around the afterbody including the propeller suction;
the propeller thrust and torque, propeller rate of revolutions, the effective wake distribution
and the thrust deduction.
Total wake is the velocity field including the propeller action. The total wake differs from the
nominal wake by:
 the propeller induction;
 the effect of the propeller action on the flow around the hull,

but the split between these two can depend on the propeller model used and is, therefore, not
entirely unique.
Effective wake is the total wake, minus the velocity induced by the propeller; so it includes the
second contribution but not the first. Thereby it depends on the split between both effects and
on the propeller model. The effective wake is defined such that the propeller operation in the
flow field behind the ship can be approximated by that in an infinite domain with an inflow
equal to the effective wake field.
Thrust deduction (t) is the fraction of the propeller thrust that is lost due to the propeller
suction on the hull:
RT = (1-t) T (64)
where RT is the total resistance for the case without propeller. t can be computed from the
increase of the longitudinal (pressure + friction) force acting on the hull as a result of the
propeller action.
The effective wake fraction we is usually deduced from thrust identity. This means that from
the thrust of the propeller at a given RPM, as measured or computed, the value of K T is
calculated
T
KT 
 D4 n2 (65)
Next, from the open-water diagram the value of the advance ratio J = Ve/n.D is found that
gives that KT value, and we can be calculated.
we = 1 – Ve/Vs (66)
The relative rotative efficiency ηR is the ratio of the torque of the propeller in open water at
that J-value, and the torque actually measured or computed for the propeller behind the ship.
4.5.3. Selection of mathematical model

In VIRTUE, two main procedures have been developed:
 Method I: a RANS solution of the flow around the propeller behind the ship, as part of
the computation for the flow around the hull;
 Method II: a potential-flow solution for the propeller, iteratively coupled with the RANS
computation for the flow around the hull.
In the first case, the propeller flow is computed in a grid block that rotates with the propeller,
and is coupled with the ship-fixed grid by sliding interfaces.
In Method II, various ways of coupling have been applied in VIRTUE. In general the propeller
is modelled by any potential-flow representation, such as a vortex lattice method, a lifting line
method, or a panel method. The mathematical model thus is a combination of potential and
viscous flow, which has some conceptual difficulties but turns out to be effective. We present
the steps to be made in such a coupling below.
4.5.3.1. Representation of the propeller

In the computation of the viscous flow around the ship hull, the effect of the propeller is
usually represented by the forces it exerts on the flow through the propeller disk. These forces
are added as given ‘body forces’ in the right hand side of the momentum equations; for the
part of the flow domain where the actual propeller is located. At least the axial force

component (thrust) needs to be incorporated. The radial and tangential forces exerted by the
propeller may be incorporated or omitted.
4.5.3.2. Determination of the force distribution

The forces applied in the RANS computation should correspond with those exerted by the
propeller and can be derived from a propeller computation, for the propeller operating behind
the hull. As the propeller rotates through a spatially variable (but supposedly steady) inflow
field, this needs to be an unsteady propeller computation. The blade force distributions are
thus found as a function of the blade position. In a ship-fixed (non-rotating) coordinate frame,
for each point in the propeller disk the forces are in principle periodic with blade frequency.
While a full coupling thereby would again be unsteady, and require a time-dependent
computation of the flow around the hull that would have a blade-frequency variation, this is
generally neglected in Method II. Therefore, the time-varying forces for each point in the
propeller disk, in the ship-fixed frame of reference, are time-averaged to provide the forces
(and their spatial distribution) to be imposed in the steady computation of the ship flow. There
is no need for circumferential averaging in the coupling approach described and applied here.
4.5.3.3. Determination of effective inflow

As the propeller affects the flow around the stern, it also affects its own inflow. Therefore, an
iterative procedure needs to be followed, e.g.:
 first a computation of the flow around the hull, without propeller;
 then, a propeller computation in that flow field, producing a first approximation of the
force distribution;
 next, a second hull flow computation with that body force distribution imposed,
leading to a new estimate of the propeller inflow;
 a new propeller computation for that improved inflow; etc.
The general experience is that this procedure converges quickly.
The inflow field computed with the body forces present (the total velocity field) contains an
effect of the body forces (or the suction of the propeller), so-called induced velocities. These
must be subtracted from the inflow, because otherwise the propeller computation would not
give the right answer. Usually, the induced velocities are computed by the propeller code, and
are found from the previous propeller computation.
By either the full-RANS approach (Method I), or the RANS/potential flow combination (Method
II), the complete interactive problem can be solved for a propeller with given rotation rate.
However, the resulting thrust force may not equal the resistance plus the ‘thrust deduction’
interaction force (the increase of the resistance force acting on the hull due to the propeller
suction). If a ‘self-propulsion’ computation is desired, iteration is needed to balance the axial
forces by adjusting the propeller RPM. From that final result, in principle the power and RPM
follow.

4.5.3.4. Selection of boundary and initial conditions
Most of the boundary conditions are chosen the same as for a case without propeller. The
exception is the symmetry condition at the ship’s centreplane. The computation of the
propeller action in a wake field generally leads to a port/starboard asymmetric force
distribution. Thus the part of the flow that feels an effect of the propeller also is asymmetric.
To include this without further approximation, the domain needs to cover both sides.
If tangential forces are omitted, a further approximation is possible by imposing the average of
the port- and starboard axial forces. There is no systematic experience yet on the error in the
result caused by this.
In Method II, generally the flow field found without propeller action is used as an initial
condition.
In Method II, for the propeller computation the inflow velocity distribution is to be prescribed.
As explained above, this is to be the effective wake field. Most propeller models expect an
inflow velocity given in the propeller plane itself, but it may be impossible to compute the
induced velocities in that plane; so the effective inflow cannot be computed. In that case, the
options tried are:
 to compute the effective inflow at a slight distance upstream of the propeller, but
impose it unchanged as ‘the’ inflow field;
 to use an alternative propeller model or representation that allows to compute the
induction in the required plane.
Experience so far suggests that the choice of the upstream plane in the first option does
affect the results.
4.5.3.5. Selection of grid topology and time step

In Method I, when performing numerical propulsion simulation with rotating propeller by a full-
RANS approach, a two-grids system should be used: The grid of the fixed ‘ship block’ as
usually used in RANS computations, and the grid within the propeller block is rotating with a
constant rotational speed. The three sliding interfaces are the front and the rear boundaries
and the cylindrical outer surface. The key point of the propulsion simulation with rotating
propeller is that the consistency between the computed velocities and pressures in all
interface cells of the propeller block and the ship block must be established. At least
according to the experiences with the COMET code, this may be achieved by applying a
much more strict convergence tolerance within each time step, two orders lower than usually
used in RANS computations. It is recommended to limit the number of the so-called ‘outer
iterations’ for velocity-pressure correction to 20. This means that even if the strict
convergence criterion can not be fulfilled, the velocity-pressure correction should be
performed 20 times within a time step. Experiences have shown that this has improved the
computation results considerably. The angular movement of the blade of 2 degrees per time
step should be chosen. The corresponding time step can be found for the give rotational
speed. Computations for at least 5 turns of the propeller should be performed, until the
expected periodic behaviour of the computed thrust and torque of each blade is established,
which indicates the convergence of the propulsion simulation.
In Method II, the force distribution representing the propeller needs to be deduced from the
propeller computation, and imposed in the hull flow computation. This requires an
interpolation or distribution of the forces towards the ship grid. Without further precautions that
ship grid is usually not really appropriate for representing those forces. This can be improved
by locally adjusting the grid to match the propeller as in Figure 4-23. For the interpolation

itself, some use an intermediate axisymmetric grid block to improve the implementation.
Refining the grid in the propeller area may be needed to resolve the force distribution.
Figure 4-23: Grid view in the propeller zone
In Method II, to represent accurately the force distribution, the grid density in the propeller
area should be large enough. This may mean that the grid density in the propeller region
should be at least comparable to that used in the propeller code (vortex lattice or panel
method) for unsteady calculations with respect to the radial and angular increments.
4.5.4. Accuracy of modelling

In Method II, the effective wake field is defined as:
Ve = Vtot(RANS) – Vind(prop) (67)
Here, the computed Vtot(RANS) can be considered to be the effective wake plus the
induction caused by the body force distribution. Therefore, the correctness of the effective
wake field hinges upon the equality of the induction from the propeller model on the one hand,
and the induction from the body force distribution on the other hand. Any difference between
both introduces an error in the effective wake.
Usually, the body force distribution represents the lift on the propeller blades only. If the
induction Vind(prop) also includes the effect of the blade thickness, a difference between the
two induced velocity fields results. Also other inconsistencies can cause differences.
Therefore, it is recommended to compare the induced velocity fields in a simple open-water
configuration. From a propeller calculation, the load distribution and induced velocity field at
the coupling location are computed. Next, a RANS computation is made for that distribution of

body forces in a uniform onset flow. Any difference of the flow velocity from the uniform inflow
is then the induction from the body force distribution, which can be compared with that from
the propeller model. If there is a difference, one should figure out (and eliminate if possible) its
cause.
It is remarked here that the induction caused by a certain body force field depends on the
inflow speed. A given force distribution in a wake field causes a different induction from that in
open water.
In a method that uses a coupling plane just upstream of the propeller, it has been found to be
important to impose the propeller forces over the volume swept by the rotating propeller
blades, with the correct longitudinal distribution. If the forces are concentrated in a single
transverse grid plane, or distributed uniformly over a number of grid planes, the induced
velocity field agrees less well with that from the propeller model. Therefore, for best accuracy
a 3D interpolation of the propeller forces towards the ship grid is desired.
Together with the requirement to resolve the force distribution in the crossplane, depending
on the fineness of the grid this leads to some 2000 to 15000 cells in the ‘body-forces’ block
in practice.
In some propeller models, the blade loading is attributed to a single line for each blade. For
the propeller behind the ship, this single line must be a reasonable representation of the blade
location, e.g. the mid-chord line; not a generator line that can deviate from this significantly for
skewed propellers. Such deviations would mean that the forces are imposed at the wrong
circumferential location.
4.5.5. Error sources

There are several possible error sources in these propeller/hull interaction calculations.
 The computed wake field. If the average (nominal) wake fraction deviates from the
experimental value, the propeller thrust at given RPM deviates.
 The propeller model. The accuracy of the model can be easily determined for open
water, but not for a propeller behind the ship. Most potential-flow methods will cause
deviations at high and low propeller loading.
 Differences in the induced velocity field, in Method II; see above.
 Dependence on the location of the coupling plane, in Method II in some approaches.
 Any other approximations, such as neglecting some components of the propeller
forces, or port/starboard averaging, in Method II.
 Incomplete matching of velocity and pressure field across sliding interfaces, in
Method I.
 For self-propulsion simulations, errors in the predicted resistance.
 Various numerical errors, due to discretisation, incomplete convergence, interpolation
errors for forces and induced velocities, etc.
Some of these error sources may fortuitously compensate each other, so it is important to try
to distinguish and control the various errors separately.

4.6. Optimisation of ship hulls

The performance of numerical methods for prediction of the flow around ship hulls has
improved over the past few years. The accuracy of the results is now at a level where the
computed results in many cases can be used to rank different design alternatives. A next step
is therefore to include the numerical analysis in an optimisation procedure where the shape of
the hull is modified based on the computed results. A well validated numerical model is of
outmost importance for a successful optimisation. The steps described in section 4.6.1, 4.6.2
and 4.6.3 must therefore be followed in order to secure that the flow physics is captured and
that the numerical accuracy of the computation is good enough.
The weak part of an optimisation system is the accuracy of the numerical computations. The
optimisation methods will “always” work if they have access to accurate numerical results.
NOTE!!! The outcome of an automatic optimisation depends to a very large extent on the set-
up. Do not expect too much if you are not familiar with the different parts of the optimisation
system. An optimisation system can never be used as a “Black box”.
Experience based on optimisation in Virtue indicates that the most crucial elements of a CFD-
based hull form optimisation process are:
 The hull form variation method, which should be a trade-off between generality and
limitation of the number of parameters.
 The accuracy of the numerical computations
 Computer capacity. A large number of hull variants must be computed.
 The choice of optimiser is not too essential
4.6.1. Hull deformation scheme and design variables

The hull deformation scheme and the design variables used to control the hull deformations
are very important since they decide the shapes that can be generated during the
optimisation. The time for an optimisation depends to a very large extent on the number of
design variables. A flexible hull deformation scheme that can be controlled by as few design
variables as possible is therefore preferred.
It is however important to limit the deformations allowed in a particular optimisation. Variations
that change the flow features around the hull to a large extent should be avoided since the
accuracy of the numerical computation may vary for different hull types, (see validity
constraints in section 4.6.3).
It should be checked that tools for automatic grid generation or deformation that are used in
the optimisation process do not influence the computed trends. This can be done e.g. by
rerunning the computation for a final shape on a newly generated grid, and compare with the
result on the automatically deformed grid.
A three step approach is therefore suggested for the optimisation work.

1. Global variation: Here the main dimensions, displacement and geometry coefficients
like block coefficient and prismatic coefficients are determined thru manual or
automatic optimisation procedures. Important at this stage is also the transverse
stability and the longitudinal trim computations. The main particulars of the hull can
be used as design variables in a formal optimisation, but they are often introduced as
constraints due to limitations in restricted water such as canals, locks and shallow
water.

2. Regional variation: This step concerns the shape of the fore body, the midship
section aft body and appendages. There are several strategies for the regional
variation
 Traditionally Naval Architects have used basic curves such as sectional area
curve, design waterline, stem and stern profile as a base for the design of the
hull shape. The basic curves can be controlled by a few form parameters in
an automatic optimisation procedure, and the shape of the hull including
areas, volumes and moments can be computed. Basic curves can also be
used during the global variation step above.
 Another approach is to use deformation curves or deformation surfaces to
modify the shape of the hull. In this approach the deformation of an “external
curve or surface” is projected to larger or smaller parts of the original hull
form. The deformation of the external curve or surface is controlled by a
number of design variables.
 A method that introduces both global and regional variations of the hull shape
is the Lackenby shift. Here the prismatic coefficient, the LCB and the parallel
midbody of an existing hull are modified while LWL and BWL and the
displacement are unchanged.
 Deformation boxes
 Interpolation between basis hull forms
3. Local variation: The first two methods described for regional variations can here be
used over a smaller part of the hull in order to tune local flow features or to have a
smooth surface.
Some hull deformation schemes may guarantee smooth hull surfaces while it is necessary to
introduce a lines fairing step together with other schemes.
The procedure above can be considered as a Geometry deformation approach. The geometry
for each variation of the hull is used for generation of the computational grid.
An alternative is a Grid deformation approach where the computational grid for a basic hull
shape is deformed using an “external curve or surface” controlled by a number of design
variables. The Grid deformation approach calls for a grid smoothing step which ensures a
good grid quality in the computational domain after a deformation of the grid at the hull
surface.
A last but very important step is to decide the range (minimum and maximum values) for the
design variables. Each min and max value must be tested to ensure that “realistic” shapes will
be generated and that the quality of computational grids will be good enough to maintain the
numerical accuracy of the solution. A careful investigation will help to avoid interruption during
the automatic optimisation. It will also save computing time since computations with bad grid
quality can be avoided.
4.6.2. Objective functions

From a ship owner’s point of view the ship should carry a high payload as fast as possible at
the lowest possible cost. This measure of merit can be expressed in a number of different
quantities like the Required Freight Rate or the Ship Merit Factor. Fuel consumption,
propulsion efficiency or resistance are explicitly or implicitly included in these measures of
merit for a ship. The discussion below concerns the resistance and propulsion contribution
only.

Optimisation may be used both in the design phase and during operation of a ship. In the
design phase optimisation can be used to suggest hull shape modifications with respect to
resistance, wake quality, propulsive efficiency, seakeeping and manoeuvring. During the
operation optimisation can be used to find optimum trim and draft for both full load and ballast
condition for a fixed hull shape.
The objective function Resistance is easy to define but difficult to compute accurately. Care
has to be taken that all computations are converged to such a level that the iterative error
does no longer influence the computed trends. Since iterative errors are in general
unsystematic, they can easily disturb the computed trends. Therefore again the iterative error
should be lower than the discretisation error. Furthermore, the above-mentioned
recommendations should be taken into account: for example, it must be ensured that the size
of the computational domain and the formulation of the boundary conditions do not influence
the computed trends. Therefore, extensive investigation is necessary prior to carrying out an
optimisation.
Wake quality which will influence propulsive efficiency and cavitation is another objective
function which requires a very accurate well validated numerical model as described for the
resistance objective. It is much more difficult to define an objective function for the wake
quality. The dependence between the design parameters and the wake objective is not
obvious since the velocity distribution in the wake is influenced by many upstream flow
structures. It is suggested that the wake objective should measure at least the circumferential
velocity variation and the wake fraction. This objective should be representative for the
aspects that play a role, such as cavitation; but at the same time should preferably not be too
sensitive to numerical parameters. The details of the formulation is however important and it
is important to validate that the wake objective function gives the right indications for the
range of the design parameters.
The effect of a working propeller is very important for both the resistance and the wake. But
including a working propeller in the computational model makes it even more difficult to obtain
an accurate numerical solution. It is therefore suggested to start with an optimisation without a
working propeller.
It is also of importance to find a trim that minimise the resistance for different drafts. Both full
load and ballast condition must be considered. But it is also important to find an optimum trim
during a voyage. The optimum trim may change due to consumption of fuel and/or by partly
unloading the ship.
If more than one objective is of interest during an optimisation study it must be decided if the
optimisation will be carried out as a single or a multi objective optimisation. If a single
objective optimisation is used for a multi objective problem, it is important to find a weight
function distribution that corresponds to the relative importance of the different objectives. For
a multi objective optimisation combinations of objective functions are plotted in a diagram. All
solutions are plotted and points that are optimal in at least one objective are connected by a
curve, the Pareto front. Along this curve, a gain in one objective gives a loss in the other
objective. It is then up to the designer to decide which objective that is the most important.
4.6.3. Constraints
In order to prevent that the result of the optimisation process is not practically acceptable,
care has to be taken that the set of constraints is comprehensive and all hull forms that satisfy
the constraints are considered acceptable.
Three types of constraints can be identified for an optimisation:
 Design constraints which concerns main dimensions, displacement, LCB, space for
cargo and engine

 Geometrical constraints that apply to the curvature of the surface, retaining the flat of
side / flat of bottom, respecting a certain tip clearance, not cutting out the contour too
much. . These are important for the production cost.
 Hydrodynamic constraints: the ship must have adequate manoeuvrability and
stability.
 Validity constraints concerns limits for the numerical model. An initial step before an
optimisation is to validate the numerical model within the range for the design
variables. The numerical model should not be used outside the range for the
validation due to the risk of computing the wrong trend during the optimisation.
4.6.4. Optimisation algorithm
There are a number of optimisation methods that can be used for the hull shape optimisation
problem. The methods can be divided into two types: stochastic and deterministic. The
stochastic methods cover the design space and make it possible to find a global minimum.
Only function evaluation is necessary. The Genetic algorithm which follows the principle of
survival of the fittest is a stochastic method which can be used for ship hull optimisations. The
Genetic algorithm is easily used for multi objective optimisations. The deterministic methods
check the area around the present design point and find the direction towards a better design.
Only local minima can be found by the deterministic methods, unless they are applied with
many starting points. For cases where it is known that there is a single minimum the
deterministic methods are efficient, but in most cases it is not known if the design space
includes multiple minima. In that case it is recommended to use stochastic methods.
Systematic variations of the design variables can also be used as an alternative to the
automatic optimisation algorithms. Such systematic variations more clearly show the trends
and help identifying desired directions of hull form changes. On the other hand, systematic
variations are hard to apply for many parameters at a time.
Therefore, the typical way of working is different:
 while for automatic optimisation a substantial number of variation parameters is to be
specified in advance, and the optimiser finds its way in the design space,
 for systematic variations the number of parameters varied simultaneously is small
(maximum 5), and based on the observed trends new parameters are added and
some others dropped, aiming at a stepwise refinement of the design.
 Restart from the numerical solution of an initial shape may reduce the computation
time for the new variants.
It is recommended to check the constraints before a variant is computed. If the variant
violates the constraints there is no need to evaluate the candidate.
4.6.5. Investigation of the design space

As an initial step before the optimisation it is necessary to determine the range for the design
variables in order to ensure that realistic shapes will be generated by the surface
representation and that the numerical method can handle the shapes. Check if the shapes
are realistic. If not change the limits for the design variables or modify the definition of the
variables. It is also very important to do a grid dependence study to ensure that the same
trends are predicted at different refinement levels.
In order to reduce the computing time it can be efficient to introduce an approximation of the
design space. After an initial set of computations the design space is approximated by
polynomial surfaces, Krieging or Artificial Neural Networks. The basic idea is to replace the

numerical simulation by an evaluation of the approximation surface. This will reduce the
computation time considerably.
If an optimal hull form has been computed, the grid dependence in the computed gains in the
objective functions should be checked afterwards using at least one mesh that is significantly
finer in all directions than the meshes used in the optimisation process.
4.6.6. Automatic optimisation

Three blocks are needed for an automatic shape optimisation based on numerical
computations.
1. A hull generation/deformation tool or alternatively a mesh deformation tool.
 Shape variation based on values of a number of design variables provided by
the optimisation tool.
 Most of the constraints can be evaluated and checked.
 Infeasible hull variants can be removed.
2. A flow solver.
 Provide values for the objective functions.
 Provide constraints which are dependent on the computed flow.
3. An optimisation tool.
 Define the objective function(s).
 Define the design variables.
 Select optimisation method.
Modern optimisation tools can handle the data transfer between the blocks. The set up of the
optimisation system is easy. The accuracy of the values for the objective function is the
critical point in an optimisation.
The optimisation methods are robust and will suggest improved designs provided that
accurate values of the objective functions are submitted to the optimisation method.

5. Application 2: Numerical Sea keeping tank
5.1. Theoretical aspects of sea keeping
5.1.1. General theory of sea keeping
The main objectives of the sea-keeping analysis during a standard ship design process are
related to:
 The platform’s motions in waves have an influence on a number of design issues
including;
o For passenger vessels, comfort and safety in a sea-way,
o For cargo vessels – cargo behaviour, including sloshing of liquids and shifting
of bulk solids,
o For naval vessels – green water and operations of equipment, helicopters
etc.,
o Behaviour in both intact and damaged conditions, with flooded compartment
and water on deck, from the viewpoint of stability.
 The additional resistance due to waves for ship with forward speed and the sea-state
conditions (wave period, height and direction). Recommendations on ship speed and
direction could be derived to improve safety and to reduce fuel consumption
 The wave loads induced on the hull, appendages and deck equipments: pressure
distribution, slamming loads and induced whipping, local wave impact on the hull,
green water impact on deck equipments.
Sea-keeping is mainly governed by in-coming wave conditions, with free surface / hull
interactions influencing both hydrostatic and hydrodynamic loads. Sloshing and slamming
effects can also occur. Even if inertia loads are predominant, viscous and drag loads are key
points for several critical aspects of the response: drag and lift loads on appendages,
damping in motions (mainly roll), friction at the air / water transition.
Sea-keeping codes based on the linear wave diffraction theory have been used for 15 years
and give quite satisfactory results with respect to its basic assumptions. However:
 The application of potential flow theory does not predict viscous effects and vortex
shedding, which are mainly critical for roll motions close to the natural period.
Influence of hull friction and drag loads on appendages (bilge keels, rudder) are not
directly taken into account in the modelling.
 Wave steepness remains small according to the linear free surface condition for both
1st and 2nd order diffraction theory. Severe sea-states effects, such as breaking
waves, run-up along the hull can not be analysed. The model of the in-coming wave
is the 1st order Stokes formulation (Airy), imposing limitation for shallow water case.
 Ship motions are considered small compared to the main dimensions (length, breath
and draft) in order to use a linear kinematic condition on the hull. Large motions, if
predicted by linear diffraction calculations, could be contestable.
 Ship forward speed is small (Froude number) compared to the wave velocity in order
to simplify the free surface equation.
In linear theory, both fluid flow equations and ship motions equations can be solved in
frequency domain to compute transfer functions of loads and response. Response for

irregular waves can then be obtained through a spectral analysis using the wave spectra and
the wave scatter diagrams. The standard way to proceed is the following:
 Build a mesh of the ship hull from the CAD definition (about 1,000 to 10,000 surface
cells are generally used)
 Perform the 1st order diffraction and radiation analysis in frequency domain for a
range of wave periods and directions
 Solve the ship motions equations, including additional loads for viscous/drag effects
on hull and appendages by means of empirical formulations and coefficients. The
linearisation of wave loads and ship motions imposes that both hydrostatic loads and
additional drag loads are also linearised. Viscous and drag coefficients are derived
from model tests or from user expertise. At this step, only wave frequency
components of the motions can be issued.
 Derive the mean wave added resistance from the 2nd order diffraction calculations,
which depend on the 1st order motions.
Several approaches are available to include forward speed effects. Standard sea-keeping
codes propose the well-known “frequency of encounter” (where the ship speed is neglected in
the free surface conditions) and/or the “full” linear free surface condition (terms, with linear
and quadratic dependence of the ship speed, are included in the free surface conditions).
However, coupling effects between the steady flow and the diffraction flow are generally
neglected, which means that forward speed is taking into account directly in the free surface
condition (the well-known “slender ship assumption”).
Alternative could be used to eliminate some of these severe assumptions:
 Solve ship motions equations in time domain to include some non-linearities (as
hydrostatics, drag/lift on appendages, propulsion load, wind loads), keeping linear
wave loads, which is not quite consistent with the fluid flow assumptions and the
wave added resistance calculation. Most of the standard sea-keeping codes propose
a non-linear time domain approach for motions prediction. This approach could
include diffraction loads up to 2nd order in irregular waves, which is the case for ship
towing or ship mooring configurations.
 Solve the potential flow problem directly in time domain, using a full non-linear free
surface conditions coupled with the ship motions equations. However, limitations
subsist as viscous effects are neglected and in-coming waves are estimating by the
Stokes model, excepted if the concept of the numerical basin is used (wave-maker
and beach conditions).
Computing diffraction loads in frequency domain and then ship motions either in frequency or
in time domain leads to acceptable CPU time consuming with respect to screening analysis
required for ship optimisation. However, non-linear potential theory requires large CPU time
not far from those needed for viscous CFD calculations. This second alternative has been
tested in R&D projects but so far has not found any industrial applications.
Model tests in wave tank facilities are still extensively used as an alternative to numerical
simulations but also with major limitations:
 Scale effects for viscous and drag loads
 Dimensions of the tank: reflection on lateral walls, on beach and on wave-makers
 Stability of the waves avoiding long time tests needed for statistics
 Range of wave periods, amplitudes and energy distribution in direction
 Measurement of specific parameters as flow velocity field, run-up, water propagation
on deck, …

Duration and cost of model tests are also difficult to integrate at an early stage of the ship
design. Then, why do we need to develop and use CFD models if panel codes are so easy-to-
use? The main reason is in the drastic limitations of the standard sea-keeping codes which
can not take into account:
 In-coming steep wave and breaking waves
 Non-linear diffracted waves, local breaking waves, run-up along the hull and green
water on deck
 Large ship motions in waves, coupled with internal liquid motions (partially filled tanks
or flooded compartments)
 High level loads induced by ship slamming in waves and waves impacts
 Viscous and drag loads on hull and appendages
 Coupling of sea-keeping response with incident flow for ships in route or in
manoeuvres
5.1.2. CFD approach for viscous flow as a complementary tool for sea
keeping
As mentioned in the next section, a sea-keeping study requires a large number of calculations
to take into account the different loading cases:
 with ballast,
 fully loaded,
 with damaged conditions,
 manoeuvring in harbour channel
and when combined with met-ocean data, mainly sea-states conditions, can lead to several
hundreds loading cases to be analysed.
Even if non-linear free surface CFD codes were to be available, it is unrealistic to use them
directly today for full sea-keeping analysis. The basic approach is to use standard, linear, sea-
keeping codes through a screening study to identify the specific problems and the most
critical cases for which CFD codes are required instead of experimental model tests for areas
such as:
 the derivation of viscous and drag coefficients for damping estimation of resonant
motions (mainly roll): numerical decay tests or forced oscillations tests,
 estimation of waves run-up and impact through simulation in non-linear regular waves
or waves grouping (steep waves, freak waves, wave breaking) for specific time
windows previously identified
 estimation of added wave resistance in high steep waves (only short simulations
required)
 estimation of sloshing and/or slamming loads to be taken into account in standard
sea-keeping codes
The physics involved for each problem, which requires a CFD analysis, are different, mainly
concerning the timescale of the phenomena and the influence of the viscous effects.
However, in each problem, modelling of free surface effects and ship motions are required.
Methodology and algorithms must be adapted to physics. In Virtue project, four kinds of
needs have been identified and are briefly described hereafter.

5.1.2.1. Estimation of damping coefficients

Diffraction theory provides wave loads, added mass, radiated waves damping and wave
added resistance with a good level of confidence. Heave, roll and pitch motions of a ship are
generally taking place in the range of the in-coming wave periods leading to possible
resonance effects. Wave radiated damping is sufficient to predict correctly the motion
amplitudes, except for the roll. Ships are generally equipped with appendages inducing
additional damping through friction and vortex shedding (keel, rudder, bilge keel, foil).
Damping coefficients are generally derived from model tests with the problem of scale effect.
An alternative is to use a RANS code, proceeding as follows:
 selection roll amplitude and period range for which damping is required
 calculation of the roll moment imposing the roll motions during the CFD simulation
 estimation of the viscous damping coefficients associated to a given damping
formulation
 transfer of damping coefficients to a sea-keeping code through a data base tool
The estimation of the viscous damping coefficient needs to remove first the hydrostatic loads,
added mass and wave radiated damping from the calculated roll moment. The second step is
to select a roll damping formulation in coherence with formulation used in the sea-keeping
code. Derivation of damping coefficients needs accurate post-processing tools based on
Fourier analysis or time domain identification algorithm.
Calculations can be performed for different ship loading cases and appendages
arrangements to provide an extensive damping coefficients data base for a given type of
ships. Damping formulations can be used with a frequency or time domain approach. A semi-
automatic process has been developed in the VIRTUE project which is composed of:
 the RANSE code EOLE, to compute viscous damping loads from imposed motions
 the standard sea-keeping code DIODORE, to compute ship response in frequency or
in time domain
 ROLEX interfacing EOLE and DIODORE, including the damping coefficients data
base
During the VIRTUE project, experimental tests were carried out by DGA/BEC for a container
ship (HTC) equipped with bilge keels. The ship model was towed in still water with imposed
roll motion. Measured loads and damping coefficients have been compared with CFD
calculations (ECN/BEC and Principia) with a good agreement.
5.1.2.2. Ship response in severe seas

URANSE codes are now able to predict ship response in non-linear monochromatic waves.
Non-linear diffraction, local breaking waves, run-up and deck wetting can be modelled in the
same simulation such as shallow water influence, friction on hull and vortex shedding in the
wake of appendages. To reduce computing time, the following user methodology has been
proposed:
 Screening study performed with a standard sea-keeping code to select the most
critical sea-states within the design parameters: ship motions, wave added
resistance, green water, occurrence of slamming, etc.
 For each selected sea-states, identification of the time series sample including the
most critical waves group

 Simulations of the selected cases with an URANS code, imposing the initial
conditions given by the previous screening study
 Post-processing of the selected design parameters: ship motions, impact and
slamming loads, water flow velocity on deck, local acceleration and velocity, loads on
appendage, efficiency of rudder and/or keels
 Including internal liquid motions could easily be done by building a mesh of the
volume of the internal tank. Opening a compartment to sea will directly permit to
study the compartment flooding.
Today, the main limitation of using RANSE codes is that of the computing effort needed, and
the availability of multi-processors computers. This difficulty will probably disappear in the
next few years.
5.1.2.3. Internal liquid motions in tanks and flooded compartment

Several problems are considered in this section:
 Internal liquid motions in partially filled tanks of large size
 Fluid flow in flooded compartment in hull damaged conditions (breaches in the hull)
 Fluid flow on deck in severe seas
Estimation of internal liquid motions is a key point for the design of LNG carriers. Internal
waves impacts on the tank wall induce stress in the thermal insulation coats and in the tank
fixations. Two timescale have to be considered resulting in a two-stage analysis:
 Stage 1: The global sloshing inside the tank induced by the ship motions in waves.
An adapted sea-keeping code, as mentioned above, can be used to derive the
internal free surface motions and the location of impacts. The assumption of
incompressible fluid and an one-phase (liquid) flow is retained at this stage
 Stage 2: The local fluid impact on the wall and the resulting dynamic response of the
insulation coat are governed by the characteristics of liquid natural gas. The local
physics is quite complex (cushioning, wall vibrations) and a modelling of two-phase
compressible flow coupled with the structural response is recommended.
In the Virtue project, only the first stage has been considered in the validation step. But the
algorithms, VOF and SPH, developed in the CFD codes, could be used to simulate the
second stage as the physics are quite similar to those of local slamming (air trapping, local
structural deformation and vibrations).
The use of potential theory for modelling internal flow with free surface has been developed in
the same time as diffraction codes. However, limitations of this approach are highly drastic for
sloshing in partially filled tanks:
 natural periods of the internal standing waves are generally excited by the ship
motions which produce large free surface motions, wave breaking and wave impact
on the walls inside the tank
 In case of small filling ratio, non-linear progressive waves are taking place, not
predicted by linear potential theory
 Friction/drag effects on the wall and on internal equipments must be predicted
 In case of a flooded compartment open to sea, loss of pressure through the breach is
given by the turbulence of the flow

The standard methodology to analyse sloshing effects is:

 Sea-keeping simulations with standard software, considering the tank filled with rigid
mass. Effect of the free surface is only taking into account for stability and hydrostatic
loads
 Simulations of the internal liquid motions for the most critical using a CFD code to
provide internal global and local impact conditions on the tank walls
If tanks are of large size (new LNG carriers), model tests have shown that ship motions are
influenced by liquid motions. Recently, sea-keeping codes have been adapted to include
liquid motion in tanks modelled by potential flow theory coupled with linear free surface
conditions. It is also possible to simulate effects of a flooding compartment assuming a steady
response of the ship (no sinking). This approach gives quite good results for the ship motions
in moderate wave heights, but seems not to be so well conditioned for sloshing in more
severe sea conditions.
An alternative solution explained in the Virtue project consists to couple a time domain sea-
keeping with a CFD code where:
 The sea-keeping code solves the ship motion equations including wave loads, non-
linear hydrostatic loads, damping and other external loads. The ship motions are
transferred to RANSE code.
 At each time step of the sea-keeping simulation, loads induced by liquid motions
inside tanks are calculated by the RANSE code and transferred to the sea-keeping
code.
Validation of this approach has been done using existing experiments obtained for a simple
configuration composed of a rectangular barge moored in beam sea and equipped with
rectangular tanks. A large range of wave conditions and tanks filling ratios has been tested.
This approach could be also used for simulation of slow transient sinking if hydrostatic and
hydrodynamic loads can be recalculated to take into account the instantaneous ship loading
induced by the compartment flooding. A limitation of the approach is the occurrence failure of
compartment walls or hull failures leading to a fast sinking. In that last case, only full CFD
calculation is appropriated.
5.1.2.4. Slamming loads and hydro-elastic response

The notion of slamming covers different physical aspects of water impact on the ship hull and
equipments:
 The impact of the bow (bulb) and stern on water induced by the ship motions in
waves. The impact is mainly vertical and is governed by the local vertical velocity.
 The horizontal impact of steep waves or breaking waves on the ship hull governed by
relative ship/waves velocity
 The water impact on the ship equipments induced by water run-up and green water
on deck
The majority of research and development in this field has concentrated on the classic water
entry problem rather than wave impact or green water effects. Nevertheless, there is a
substantial literature on both subjects, and Classification Society Rules that define empirical
approaches that can be taken. There remains however, a significant amount of uncertainty in
the choice of impact coefficients, and CFD offers a potential route to understanding extreme
design wave loads of this kind.
Slamming loads influence the hull design in a number of different ways:

 Global forces inducing hull guider vibrations (whipping)

 Local loads and pressures inducing local deformation and vibrations of hull structure
(hydro-elastic response)
Taking the ship bow as an example, the slamming phenomena can be described in the four
stages:
 Slamming occurs when the vertical bow velocity at the bow water entry induced
heave and pitch motions is greater than a specific value. The Ochi criteria is generally
used:
V  0.093 gL (68)
As the ship motion is governed by inertia, water entry velocity is imposed. That
means that slamming loads have a small influence on the ship motions
 First stage of slamming is the hull impact on water associated to a very small
timescale ( 0.005 sec) corresponding to the occurrence of the maximum of local
pressure:
p  1/2ρ CP V2 (69)
Pressure is governed by local hull geometry, free surface deformation (jets), local
hydro-elastic deformation
 Second stage of slamming is the hull water entry associated to a larger timescale (
0.1 sec) corresponding to the occurrence of the maximum of the global force which
induces whipping. Slamming force is mainly governed by the fluid inertia loads
varying during water entry and is generally represented by a simple formulation:
FS (t)  ρCSSwetted V2 (t) (70)

Where CS is an impact coefficient derived from the force calculation.
CP and CS are not correlated excepted when the maximum of pressure and global force occur
at the same time. This is, for example, the case for a flat plate or shallow wedge impacting a
flat water area, and where air cushioning is not involved. For the more practical ship hull
impact problems, the flow field and geometries involved are complex. Thus far, no well
accepted computational methods have been developed and validated, and hence this was a
subject still requiring research within the VIRTUE project.
Within the Virtue project, a calculation process has been proposed in a similar way as for
sloshing:
 Use of a standard sea-keeping code to identify the occurrence and the conditions of
slamming: long time series for specific cases of the sea-states scatter diagram.
Selection of the most critical cases.
 Use of a CFD code to simulate hull impact with the imposed water entry conditions
selected in the sea-keeping screening study. Derivation of impact coefficients C S
associated to a slamming load formulation
 Use the slamming loads time series in a FEM model to estimate stresses induced by
whipping.
An alternative is to implement a hydro-elastic beam model coupled with the rigid body
equations in a standard sea-keeping code. Then, the effect of imposing slamming loads can
be combined with the wave frequency loads. This is the case for some sea-keeping codes
used in the Virtue project. Whipping response could then directly derived.

5.2. How to use CFD jointly with standard sea-keeping codes

As mentioned before it is recommended that CFD codes be used as complementary tools to
standard sea-keeping codes, for analysis of both viscous load contributions and highly non-
linear free surface effects.
5.2.1. Viscous damping in ship motions

Standard sea-keeping codes include roll-damping formulations. The coefficients are provided
directly by the user for each run. Some default values are generally proposed. Alternative with
CFD is to compute viscous loads and to derive the damping coefficient adapted to the
damping formulation (as described in the next section), for a range of ship configurations.
Standard sea-keeping codes are generally used both in frequency domain and in time
domain:
 Frequency domain to issue transfer functions (RAOs) and to perform spectral
analyses. Loads are assumed to be linear with the wave and ship motions
amplitudes, excepted for damping which could be quadratic
 Time domain to include non-linear effects associated with hydrostatic loads, drag /
damping loads and inertia loads, excepted wave loads which remain linear.
For frequency domain analysis, CFD code is used to built damping coefficient data bases
which is then called by the sea-keeping code. An example of the general process is given
hereafter, using the tools developed in the Virtue project for roll damping prediction.
Hull geometry
Diffraction
CFD calculations Mass

FO tests matrix Wave loads
hydrostatic
Damping
coefficients
Roll RAOs (FD)

Extreme prediction
Figure 5-1: General Process for roll damping predictions
A specific module uses the main characteristics of the ship (type of hull, main dimensions,
mass distribution, configuration of appendages) and a selected formulation to provide the
global damping coefficients corresponding to the generic linear-quadratic formulation. The
following steps are usually used for a given ship:

 Select the range of geometrical and mechanical parameters of the ship (loaded
cases) which define the “ship configuration”
 Select the range of motion parameters: roll frequencies and amplitudes, forward
speed
 Select damping formulations and the corresponding set of coefficients which will be
derived from CFD
 Carry out sensitivity studies to select numerical parameters: fluid domain size, grid
density, free surface model, turbulence model, time step, simulation duration, etc.
 Derive coefficients using the recommended identification technique. Here, a specific
post-processing module is needed which can also provide interpolation between
coefficients within the given ranges of hydrodynamic parameters (experimental
design and response surface method)
Before an automatic use of this approach for design, some specific studies have to be done:
 Selection roll-damping formulation retained for derivation of coefficients
 Assess CFD calculations: sensitivity and rules for numerical parameters, validation
for a standard range of hydrodynamic and configuration parameters
 Select the coefficient identification method for post-processing of CFD time series
A similar approach could be proposed in the time domain:
CFD calculations
FO tests matrix
Hull geometry
Diffraction
Mass Damping loads
time series
Damping
coefficients
6 dof motion (TD)

Extreme prediction
Figure 5-2: General Process for roll damping predictions in time domain

5.2.2. Ship motions in wave

In conventional linearised frequency domain seakeeping codes, incoming waves are
simulated using linearised Stokes theory, and the global flow including interaction with the
body is solved after separation of incident and perturbed flow based on the assumption of
linearity. In such scheme, there is no problem of wave generation, as the incident wave is
modelled using a simple analytical model, without any correlation with the numerical scheme
and discretization used for solving the perturbation flow.
When a non linear formulation of the boundary value problem is solved in the time domain, as
it is the case for free surface CFD codes or fully non linear potential flow codes, simple linear
combinations of incident and perturbed flow do not apply any more. Usually, the complete
flow is simulated without separation between incident and perturbed flow, and waves are
introduced from the external boundary of the computational domain, through ad hoc
modifications of the boundary condition. Another possible approach is to combine potential
flow with CFD approach: the wave field is initialized with an undisturbed incident wave field,
deduced from an analytical solution, and the diffracted wave field is obtained by resolution of
Navier-Stokes equations coupled with a free surface method.
The calculation of ship response to wave excitation is treated in conventional naval
architecture by the use of either strip theory (a particularly convenient and simple approach
based on slender body assumptions) or three dimensional diffraction theory. These two
methods consider numerous assumptions in order to apply potential flow modelling in either
frequency or time domain.
The principle shortcomings of these methods can be summarised as the following:
1. They rely on linear first order potential flow theory and have limited validity when non-
linear factors dominate in design, for example:
a. In case of steep or breaking waves interacting with the hull.
b. In case of wetted area of the hull varying significantly over the wave cycle.
c. When waves “overtop” the deck edge – so called green-water problems.
d. When interactions between complex structure and waves affect motions or
loads.
2. They require assumptions to be made with respect to viscous damping or drag effects
which are not taken into account in potential flow theory.
A more accurate alternative, but much more CPU consuming, is a CFD/wave model coupled
with a rigid body prediction model allowing simulating the 6 DOF free floating response of a
ship in waves.
The coupling “fluid/structure” model is considered in each inner iteration of the numerical
scheme in order to have the fluid flow and the induced ship motions convergence during the
same instant of the simulation.
5.2.3. Influence of Sloshing, Compartment flooded and Water on deck

Internal liquid motions could be highly non linear, with waves breaking even if ship motions
remain small. Then linear potential theory is not appropriated to predict internal loads.
Alternative is to use CFD codes. Two approaches can be followed:
 Use an interface between a CFD code, computing internal liquid motion, with a
standard sea-keeping code, computing ship motions (time domain method)
 Use a fully coupled approach directly in the CFD code as the wave/ship interaction is
available

The second approach is highly time consuming and could be used only for the most critical
waves group. The first approach is illustrated hereafter:
Ship Characteristics
Sea states distribution
Hull Geometry, Structure
Wave spectra
Tanks description
Seakeeping calculations (FD or FT)

Screening study
selected critical events

CFD Calculations
Sloshing in tank
Seakeeping simulations
Short time series
Figure 5-3: General Process for ship motions in waves
For compartment flooding conditions, some additional functions have to be considered:

 in the sea-keeping code, calculations of hydrostatic loads corresponding to the ship
response at each step must provide a new reference frame for CFD calculations
 in the CFD code, one imposes external fluid flow conditions at the breach location
corresponding to the relative position of ship within waves.
5.2.4. Slamming loads prediction

The first step is to apply a methodology based of the standard approach used for design:
 Identification of the sea-sates conditions to consider
 Estimation of the ship motions in waves in frequency domain to cover a large range of
conditions. Time domain approach can be used to obtain more accurate results
 Estimation of the slamming occurrence within the waves and the part of hull
concerned in. The Occhi criteria, based on the local relative vertical velocity, is used
 Estimation of the slamming conditions and calculations of the local slamming loads
using an external routine based on CFD calculations instead of semi-analytical or
empirical formulations
 Loads are imposed on a FE model of the hull structure to provide dynamic response.
Some sea-keeping models integrate directly a simple beam model for estimation of
static and dynamic shear stress and bending moment, which is the case for Diodore.

In this methodology, the main assumption deals with:

 The estimation of the slamming conditions which are governed by the non linear local
waves field and by the non linear ship response in case of severe seas.
 The estimation of the local slamming loads, modelling of the local free surface
deformation, jets evolution and cushioning effects, vortex generation in the water
entry phase
The CFD application will be focused mainly on this key point to propose an alternative to
existing approximations. Coupling with ship motions is immediate if the concerned wave time
history is imposed in the CFD code. But ship hull kinematics provided by a standard sea-
keeping code could be also imposed to model forced impact. For the two approaches local
loads are computed and imposed on a FE model of the structure. Full hydro-elastic coupling
seems not required as high frequency vibration of the ship do not influenced the ship motions
and the local kinematics. This is true only for standard ships (majority) but could be discussed
for a “flexible” hull.
The methodology is to solve both global sea-keeping response and impact loads in the CFD
codes, even if elastic hull guider is not modeled in the CFD code. The following steps are:
 Identification of the more critical sea states and the corresponding more critical waves
sample using a standard sea-keeping code
 Simulation of sea-keeping response with the CFD code imposing the selected wave
time histories and the hull velocities. CFD code will provide the time series of the
global and local loads on the hull guider
 Calculation of whipping response using a FE model (simple beam or detailed hull
model) on which loads time series issues from CFD calculations are imposed.
The following scheme illustrates the recommended methodology issued from Virtue analysis.
Ship Characteristics
Sea states distribution
Hull Geometry, Structure
Wave spectra
Seakeeping Analysis (FD or FT)

Slamming occurrence and conditions
selected slamming events
CFD Calculations Segmented model

Local slamming loads Elastic hull guider
Seakeeping simulations
Short time series
Figure 5-4: Methodology to predict slamming loads and induced whipping response

5.3. Guidelines for numerical sea-keeping applications
5.3.1. Definition of target variables

Standard sea-keeping codes compute fluid flow in frequency domain. The time evolution of
the flow parameters (velocity and pressure) is imposed through a harmonic dependence. A
steady solution is obtained, only depending on hull mesh refinement. Time dependence is
considered when solving motions equations.
CFD codes solve the fluid flow equations coupled with the rigid body equations directly in time
domain. Numerical performance is highly depending on the algorithms selected for free
surface:
 Eulerian methods such as Volume Of Fluid (VOF) method with one-phase flow or
two-phase flow assumptions,
 Arbitrary Eulerian Lagrangian method (ALE) for which the free surface is one of the
boundary of the computational domain able to move during time,
 Particles methods: SPH (Smooth Particles Hydrodynamics).
Then numerical parameters governing the simulation are:
 Size of the fluid domain and 3D mesh refinement to capture the local phenomena
 Time step adjusted to capture the highest frequency mode of the analysed problem
 Duration of the simulation to perform confident statistics and post-processing
 Choice of initial and boundaries conditions.
Depending on the initial conditions, a transient part needs to be removed before analysis of
the time series. This operation is easier when the fluid flow converges to a steady solution. In
case of unsteady response, it could be difficult to separate the transient part coming from
numerical effects and the physical part.
Unfortunately, computer capabilities are limited. That imposes an optimisation of mesh
refinement and time step. User experience is needed to select the good balance between
accuracy of results and CPU time consuming which is depending on the physical problem.
Adequate use of a standard sea-keeping code jointly with CFD is also a key step in this
optimisation.
The results of sea-keeping analysis are mainly concerned in global parameters as loads on
hull parts and ship motions. Pressure and velocity fields or free surface elevations could be
analysed only to understand local effects. So local inaccuracy of the results could have little
influence on the final results.
An additional key point is the level of accuracy required in sea-keeping analysis. Main
objectives are to provide hydrodynamic loads for the ship structure design and to predict
maximum ship motions to define limits of operational conditions. As design rules impose to
apply safety coefficients from calculations, required accuracy level for the numerical results is
on the order of 10% to 15% (which is quite large compared to the 1% required for still water
resistance prediction). This is why little work has been done on the sensitivity to numerical
parameter for sea-keeping applications

5.3.2. Selection of adequate mathematical model for the flow physics
5.3.2.1. Viscous damping in ship motions

Contribution of viscous and drag loads, so the model has to comprise:
 Mesh refinement close to the hull
 Boundary layer and Turbulent models
 Free surface algorithm: mesh adaptive method or VOF method if influence of vortex /
free surface interaction
 Initial conditions: imposed motions of the hull in still water for various amplitude and
frequency are recommended instead of decay tests simulation. One DoF is preferred
in a first time instead of multi-DoFs expected if a high level of coupling effect is
anticipated concerning the fluid flow (vortex shedding conditions)
5.3.2.2. Ship response in waves

Same conditions as for damping loads
 Free surface algorithm: VOF is preferred to capture breaking waves, run-up and
water on deck
 In-coming wave: non-linear monochromatic waves (Rieneker and Fenton formulation)
generated by the SWENSE method. Boundary conditions impose the in-coming
waves characteristics out-side of the fluid domain. In shallow water the RF model
could be replaced by a more appropriate semi-analytical formulation
 If direct comparison with wave tank tests is needed, waves could be generated
modelling the wave-maker and a numerical beach with the adequate reflection
coefficients.
5.3.2.3. Sloshing
 VOF or SPH are required to capture breaking waves
 Two-phase flow formulation (liquid-gas) is recommended for LNG carrier tanks
 Initial conditions: imposed tank motions derived from a preliminary ship motions
analysis
 Boundary conditions: no-slip conditions on the tank wall
 Coupling with ship motions could be neglected in a first time, or could be included in
the imposed motions
5.3.2.4. Compartment flooded or water on deck

 VOF or SPH are required to capture breaking waves
 Initial conditions: imposed tank motions derived from a preliminary ship motions
analysis if the mean water quantity on-board is assumed constant
 Coupling with ship motions will be required to simulate “time to sink” conditions
 Boundary conditions: wall conditions

 Turbulent flow conditions to model pressure loss trough the breaches

Alternative would be to consider the compartment as an internal tank with an unsteady
pressure loss condition on a wall part. That means that flow taking place through the breach
does not influence the external flow close to the ship hull.
5.3.2.5. Slamming
 VOF or SPH are required to capture local jets, but BEM method could be sufficient if
slamming loads is only governed by the water entry phase
 Compressible fluid and two-phase flow could be required if cushioning effect is
expected
 Initial conditions: imposed time dependant impact velocity derived from a preliminary
ship motions analysis. Coupling with ship motions is neglected
 In-coming wave fields could be imposed through the SWENSE method but still water
conditions will be preferred in a first time
 Boundary conditions: still water flow outside of the fluid domain, friction on the hull
assumed rigid in a first step
5.3.3. Selection of computational domain, adequate grid topology and

spacing
The mesh is built considering the following objectives:

 In order to obtain a correct simulation of the hydrodynamic loads induced by frictional
effects on the hull (damping for a hull in pitch motion for example), the size of the wall
cells has to be chosen according to the thickness of the boundary layer (depending
on the Reynolds number). One has to distinguish (turbulent) models with or without
wall function.
 A no-dimensional distance is introduced:
 w U*y
y  y 
  (71)
This criterion uses the local value of the friction w and the distance y of a cell from
the wall. The size of the wall cell must be smaller than the boundary layer thickness
so that the first point can be calculated by the classical logarithmic law (k- model).
Typically, the value of

y (of the wall cell) must be in the range: [15,50]. Without
wall function, it is necessary to strongly refine the mesh close to walls in order to have
a small value of
y  . For the k- SST model for example, the boundary condition for
 must be imposed for

y  3.
 Far from the hull, the refinement must be sufficient to simulate a correct diffracted
wave field.
 The mesh is built avoiding cells having too small angles (<20°) or too flat angles
(>160°).

 The size gradients between 2 adjacent cells need to be not too large, especially in the
direction where significant local gradients of the flow occur. The corresponding ratio
must be less than 2. This ratio can be larger in the area where the flow gradients are
weak.
 Several topologies are possible to build a structured mesh. The O-O topology is quite
easy to handle and it allows getting non-distorted cells around the hull. But its major
drawback is that the gridlines are not flow aligned, on the contrary of an H topology
aligned in the longitudinal direction. Considering a free surface tracking method such
as the VOF model, the O-O topology leads to unphysical numerical wave field which
tends to fit the curved gridlines.
 The more horizontal the mesh lines are, the best the free surface tracking is. This is a
real constraint for moving mesh (attached to the structure) in pitch motion. This
problem can be avoided with specific deforming mesh methods such like the spring
method, where the mesh moves only near the moving structure.
 When a multi-block mesh is considered, non-matching boundaries conditions
between blocks can be used, but must be avoided in strong gradient areas (where
vortices occur for example).
 Due to CPU constraint (unsteady flow), one have to be aware that the number of cells
must not be too high. For example, a calculation on classical multi-processors
computers (4 processors to 16 processors) can take more than a week CPU time for
meshes containing a few million cells.
5.3.4. Selection of turbulence model

Classical two equations turbulence models can be considered for sea-keeping simulation.
The k- model is widely used because of its simplicity and economical aspect. However this
model can induce too large diffusion of the flow especially in areas where strong opposite
pressure gradients occur. A k- model, better adapted in this situation, can be preferred.
Note that for the k- model, the diffusion can be controlled introducing a filter based on a
characteristic length scale l of the turbulence.
k2
Knowing that the eddy viscosity is  t  C , the idea is to filter  with the length scale l

k 3/ 2
(  min  C t
3/ 4
), so that does not overtake a maximal value.
l
For this improved k- model, the value of the turbulence length needs to be correctly
considered, not too high to avoid diffusion problems (sometimes encountered with the
classical k- model), but not too small otherwise the larger turbulence scales would not be
represented.
Otherwise, and as the example calculations given later on the CALM buoy demonstrate, a
particular form of the k- model – the so called Menter SST model with flow curvature
correction terms, appears to give the simplest reliable form of eddy viscosity model for
prediction of separated viscous flows. It could not be considered ideal in all cases, and users
are advised to experiment, but it is certainly the least computationally intensive when
compared with alternatives such as Reynolds Stress (RSM) or Large Eddy Simulation (LES)
models.

5.3.5. Selection of boundary conditions and initial conditions

A typical characteristic of sea-keeping is the existing of an unsteady flow in the whole fluid
domain which imposes the boundary conditions and initial conditions:
 In the absolute frame, the no-slip condition of the flow on the hull means to impose
the flow velocity on the ship.
 On the outer boundaries, imposed preferably at a minimum of 2L from the ship, the
flow is expected as undisturbed so with no velocity.
 Non-reflective conditions must allow to fully outlet the diffracted and radiated waves
induced by the hull and avoid possible instabilities coming from the boundary.
 If a one–fluid model is used, the atmospheric pressure is imposed on the free
surface.
 If a two-fluid model is considered, there is no boundary condition on the free surface.
The atmospheric pressure is imposed on the upper boundary of the air domain.
5.3.6. Selection of time step/transient flow/simulation duration

Selection of time-dependent parameters is directly related to the physics of the phenomena to
be captured with the constraint that CPU time consuming needs to be realistic.
5.3.6.1. Time step

The time step must not be too large. It needs to be adapted with respect to a possible CFL
criteria (including velocity and size of cells) and according to the characteristic time of the
considered unsteady flow induced by the structure motions. For example, for forced pitch
motion, an alternative up and down periodic force is imposed at each cell of the mesh, as high
as the cell is far from the rotation point, inducing a very quick shift of the local velocity flow
(inversion of the rotational movement, strong variation of the acceleration). The phenomenon
is as much pronounced as the period of the hull motion is small and so the frequency of the
periodic force is high.
For forced periodic motions, the time discretisation must be chosen as a fraction of the motion
period, with an optimal rate. Based on many applications on periodic moving bodies, the time
step must be on the order of T/N with N≈100 minimum which is a sufficient value to pick up
possible higher frequencies (T/n) (although this is algorithm dependent and some solvers may
need smaller time steps than this).
5.3.6.2. Transient flow

Using an unsteady model, the transient free surface flow needs to be correctly simulated. If
the simulation is carried out in the absolute frame, so the ship moves in a fluid at rest. The
velocity of the ship must be introduced progressively from the start of the calculation in order
to avoid too large initial deformation of the free surface which could lead to unphysical
behaviour of the transient flow. For the same reason, the time step must not be too large.

5.3.6.3. Simulation duration

The initial conditions of a sea-keeping calculation correspond to a fluid at rest, so the transient
phase of the flow can be quite long. Therefore, a forced oscillation calculation must be run on
a minimum duration corresponding to approximately 5T, with T the period of the forced
motion. At this time only monochromatic waves or wave grouping conditions seems to be
realistic. Imposing irregular waves means that a statistical analysis of the output results will be
done, requiring too large simulations.
5.3.7. Selection of method for free surface prediction

The free surface can be simulated with an Eulerian tracking method such like a VOF (Volume
Of Fluid) method, or a classical adaptive method where the upper gridline follows the free
surface. Basically, the standard VOF method has a larger validity domain because it allows to
represent very complex shapes of interface, up to breaking waves.
However, this model needs refined grid to obtain a correct accuracy and to avoid unphysical
drop of wave energy and loss of liquid volume, especially for weak dynamic flow so for slow
motion of the free surface. Note that improvements of the VOF model exist in certain CFD
codes which allow getting a more accurate description of the free surface evolution whatever
the flow conditions.
SPH method is well adapted to capture fast transient flow variation as water impact and local
wave breaking. However the method needs still more research to model viscous 3D flows
even if some very encouraging results have been obtained. Implementation in a standard
URANS code is not possible and specific codes are currently developed.
5.3.8. Selection of wave models

Various models for non linear waves are possible, from classical models using boundary
conditions to introduce the nonlinear wave kinematic, to advanced potential/CFD coupled
models.
Several techniques exist such as for example those which consist in introducing the wave at
the boundaries of the fluid domain by means of:
 Analytical solutions given by potential theory: the main difficulty is to manage
propagating the waves in the Navier-Stokes domain (compatibility between different
types of solutions).
 Moving boundaries for simulating the action of a wave-maker at one of the extremity
of the domain.
Nevertheless, difficulties can appear:
 Long computational time for waves establishment
 Conservation of the water average level may be not respected.
 Diminishing of the amplitude of the propagated wave.
 Waves reflection on the selected boundary conditions
In potential/CFD coupled models, the whole field is split into a non-perturbed incident flow and
a diffracted flow. The incident flow is known by an analytical formulation, for example based
on the stream function in the case of a regular monochromatic wave (Rienecker & Fenton).
The generation from a domain boundary is the basic approach for the generation of waves in
a numerical model. Waves are generated by the imposition of a perturbation applied to a part

of the external boundary of the computational domain, acting in the same way as a physical
wave maker in an experimental wave tank. Very fine meshes are required for a correct
modelling of wave propagation in the domain in order to avoid loss of wave energy.
For the wave generation, an inlet velocity boundary condition is used to provide velocity
components based on a standard linear wave formula. The height of the water, and hence
the volume fraction, at the inlet boundary is calculated using a second order relationship that
takes into account of the total pressure. Thus the capability of the system is limited to the
generation of mono-chromatic Airy waves.
The waves generated by the above inlet condition are propagated with a free surface tracking
method such as the VOF one. A drawback of this methodology is that the wave amplitude
decays with distance along the numerical tank unless a very refined mesh is considered (so
long CPU time). However, improvements can be made, for example by the use of a moving
grid according with the propagating wave.
5.3.9. Selection of Discretization schemes

For conventional CFD codes that use a pressure-velocity coupling of some form, or the more
direct coupled solvers, the choice of discretisation scheme is critical to stability and accuracy.
For sea-keeping problems, the most sensitive part of the calculation is that of the free surface
wave propagation. In the Virtue project, it has been found that the following general guidance
is a practical starting point for all such calculations:
 That the level of grid resolution should be no less than 100 grid cells per wave
length,
 That the time step should be no fewer than 100 time steps per wave period,
 That the numerical scheme should be at least both second order in time and
second order in space.
Schemes that use first order discretisation in space will be too artificially diffusive no matter
what the level of grid resolution, and first order time discretisation loses wave energy as the
simulation progresses.
It should also be noted that all equations being solved should be of the same order, as should
any sub-models, for example for the Volume of fluid fraction (VOF) representation of he free
surface, and any grid movement algorithms.
If using general purpose CFD tools (for example CFX, Fluent or STAR-CD), these options do
exist and users should ensure that they are used or advise sought from the code vendors if
any particular code settings need to be employed in order to achieve this minimum level of
accuracy.
5.3.9.1. Pseudo-compressibility solver method

Several kinds of pseudo-compressibility methods are used for solving the Navier-Stokes
system. An example of this method is given hereafter.
The continuity and the momentum equations are coupled by a pseudo-compressibility
scheme, associated to pseudo-density integration (dual time stepping).
This is done by the introduction of pseudo–time derivation terms in the flow equations,
involving a pseudo-density which is related to pressure by a pseudo-compressibility function.
This function is chosen so that the inviscid part of the flow equations is hyperbolic in pseudo-
time.

This dual time stepping method enables to iterate over the non-linearities without an explicit
linearization. It is then extended to all subsets of equations (for the other transport equations:
turbulence, VOF, rigid body motions equations), with possible different pseudo-time steps per
subset, since no accuracy in pseudo-time is searched.
The time derivatives in the complete set of equations are expressed at the end of the time
step using a retarded finite differences scheme with an accuracy of second order.
The incompressible Navier-Stokes equations are discretised in finite volume form using a
centred scheme to compute fluxes. Second and fourth order artificial viscosities are
introduced to stabilize the algorithm and to prevent the tendency to odd-even decoupling.
The integration in pseudo-time of the continuity and the momentum equations is done using a
Runge-Kutta multi-stages scheme, with local time stepping. An implicit residual smoothing
operator is also applied following each stage of the Runge-Kutta scheme to improve stability.
The terms at the left-hand side of the set of equations are treated implicitly in pseudo-time.
The dependent state variables are updated in pseudo-time until an asymptotic state in
pseudo-time is reached.
5.3.9.2. Free surface VOF method

There are two main VOF methodologies:
 Eulerian: the VOF evolution (so the free surface evolution in the eulerian mesh) is
solved from a transport equation of the VOF function. Due to the discontinuous nature
of the VOF, the fluxes are computed from a specific well-known algorithm called
donor-acceptor algorithm (of the authors Hirt and Nichols). The method can handle
with very complex shapes of free surface. Its main drawback is related to a quite poor
accuracy (0 order) due to the hypothesis to be considered for the free surface
orientation: necessarily parallel at the cell interfaces.
 Lagrangian: the VOF interface tracking is achieved using the well-known concept of
VOF Piecewise Linear Interface Calculation (PLIC) based on Lagrangian advection of
the segments (plans in 3D) representing the free surface. In this method the real
orientation of the free surface in each cell is considered and there is no need to solve
a transport equation of the VOF function.
5.3.9.3. Free surface SPH method

The governing equations for fluid flow are the mass and momentum conservation for fluid
particle.
The lagrangian nature of SPH (Smoothed Particles Hydrodynamics) means that changes in
density and flow morphology are automatically accounted for without the need for mesh
refinement or other complicated procedures.
In general, SPH calculation is very computationally demanding, both in memory and in CPU
time. It usually involves a large number of particles to be geometrically enough to model the
deformation of fluid body. In 3D cases, the SPH model may involve several millions of
particles.
The SPH method is a very promising method, but it needs further validations before being
used as a predictive tool for industrial applications.

5.3.10. Selection of iterative convergence criteria

In order to follow convergence rate for Navier-Stokes equations, the root mean square of
residuals written in dimensionless form is defined with dependent variables (typically pseudo-
density or pressure, velocity components, turbulent kinetic energy and its dissipation rate)
For unsteady simulations, and using the dual time stepping method, the convergence of the
solution at each time step is practically obtained when the ratio between initial residual and
final residual has decreased at least of a factor equal to 100. It is generally found that the
mass (pressure) residual provides the simplest indicator. Although this appears significantly
higher than would be the case for a steady state solution, it has been found to provide a good
balance between calculation time and accurate solution.
The choice of time step size is therefore critical, since the larger the time step, the more
computational work is needed at each step of the calculation. This appears to be problem
dependent, and so some experimentation is necessary at the start of each analysis to find the
right balance.
For coupled solvers on unstructured grids, the penalty in terms of excessive computational
time can be considerable if inefficient choices of convergence criteria and time step are made.
We have found that it is often worth limiting the number of iterative sweeps of such solvers to
no more than 10 per time step if practical calculations are to be undertaken.
5.3.11. Post-processing
5.3.11.1. Viscous damping in ship motions / hydrodynamic coefficients

CFD simulations or model tests of forced oscillation tests provide time series of the total
forces and moments on the hull and, if required, local loads on specific parts such as
appendages (bilge keels, rudder, foils etc.). A methodology is required to identify the roll
damping coefficients from these loads time series:
Contrary to model tests, loads given by CFD simulations include hydrostatic K components
which can be separated before post-processing. Model tests include hull inertia.
A transient part of the loads time history must be removed and the necessary number of
periodic oscillations is selected within the remaining part. Filtering the time series can be
needed.
When “clean” time series of the hydrodynamic loads are obtained, a method can be selected
to identify the added inertia Ia and the damping components BL and BQ from the roll equation
for given I and K:
M(t)  (I  Ia)(t)  BL(t)  BQ (t) (t)  K(t)

(72)
An identification technique has to be considered. Classical methodologies can be used:

 Identification in frequency domain: separation of components in phase with the roll
acceleration (added mass) and in phase with the roll velocity (linearised damping) by
Fourier analysis.
 Identification in time domain: each term of the roll equation is identified by a least
squared method.
When the identification technique is selected, three options can be followed:
 The linear damping component is assumed to be zero, then only the added inertia Ia
and quadratic damping BQ are derived from the roll equation:

M(t)  I(t)  K(t)  Ia(t)  BQ (t) (t)

(73)
 The linear damping component is assumed to be given by potential flow theory and
only the added inertia Ia and quadratic damping BQ are derived:
M(t)  I(t)  K(t)  Brad(t)  Ia(t)  BQ (t) (t)

(74)
 Without no assumptions, the added inertia Ia and both the linear and quadratic
damping BL and BQ are derived, but this option is only available with least square
method:
M(t)  I(t)  K(t)  Ia(t)  BL(t)  BQ (t) (t)

(75)
The next step is to introduce CFD coefficients in standard sea-keeping codes.

 Added inertia coefficient: Although viscous and rotational effects are taken into
account in CFD approaches, CFD added inertia coefficient may be of the same order
than that provided by BEM codes.
 Damping coefficients: If radiation damping is undertaken by the sea-keeping standard
code, as for example for roll motion, CFD linear damping can be introduced in
addition of CFD quadratic damping which is representative of damping provided by
vortices, and viscous effects.
5.3.11.2. Ship motions in wave

For each simulation, and for motions and loads:
 Remove the transient part (typically 10 wave periods)
 Statistics on time series: mean, max, min, r.m.s., zero-up crossing period (mean)
 Spectral analysis: power spectrum, peak period(s) (if several), identification of sub or
super-harmonics
 If need, Fourier analysis to provide the main frequency components
Additional results could be required:
 Local pressure and velocity fields close to the appendage and equipments (rudder,
bilge keels, foils)
 Occurrence and conditions of run-up and deck-wetting
 Occurrence and conditions of slamming
5.3.11.3. Sloshing, Compartment flooded and Water on deck

 Loads induced by the “internal” fluid flow on the tank walls and ship equipments
 Local pressure distribution and free surface elevation

5.3.11.4. Slamming
 Time series of the slamming loads through the pressure integration on the
instantaneous wetted hull
 Maximum loads and corresponding time step and identification of slamming
coefficient CS
 Maximum local pressure and corresponding location and time step
5.3.12. Estimation of numerical accuracy of the solution

If total resistance of ship moving with a forward speed has been studying by CFD approach
for more than 15 years, sea-keeping problems of ships have been investigated more recently.
The state of the art in sea-keeping with CFD codes only began some years ago. Objectives of
VIRTUE sea-keeping work-package are not being able at the end of the project to estimate
numerical accuracy of the solution of CFD models. If the accuracy problem has been clearly
identified within the resistance work-package (Chapter 4), it is not at the moment the case for
sea-keeping for which numerical modelling involves complex problems such like,
unsteadiness of the flow, different types of coupling with free surface tracking method, or rigid
body motions equations. As mentioned before, accuracy required for sea-keeping calculations
is lower than for resistance prediction in still water. The main goal is to provide loads and ship
motions in severe seas conditions for which standard sea-keeping codes are limited.
Nevertheless, validation still remains a fundamental objective, and comparisons with either
analytical or experimental results have been carried out as often as possible, quantifying the
differences and trying to improve models for minimizing them.
5.4. Application Cases
5.4.1. Case 2-a: HTC hull with forward speed in roll motion
The basic benchmarks done with the viscous CFD codes concerns the prediction of roll
damping. A typical containership (called “Hamburg Test Case – HTC” in the Virtue project)
has been tested in the 600m length towing tank of the Bassin d’Essais des Carènes (DGA –
France) at a scale of 1:24. Constant forward speed has been imposed jointly with roll
amplitudes (tests matrix is provided in the Table hereafter). The HTC was equipped with bilge
keels. But tests were also carried out removing the bilge keels.
Forward
Roll Roll angle
TEP file Config. comments Froude speed Vm
period(s) (°)
(m/s)
V2 To=3S Rudder and bilge
RUN6 0.20 3 15 1.6
A3 keels
V0 To=3s
RUN32 bare hull 0.00 3 5 0.0
A1
V0 To=3s
RUN33 bare hull 0.00 3 10 0.0
A2
V0 To=3s
RUN34 bare hull 0.00 3 15 0.0
A3
Table 5-1: Selected runs from the model test matrix

Two types of computations have been performed with the RANS codes available: with and
without appendages.
5.4.1.1. Multi-block meshes

For case with rudder and bilge keels, the mesh comprises three blocks with non-matching
interfaces. The additional blocks are used to provide fine grid near the bilge keel and rudder.
The radiation problem is simulated in time domain. The computational domain size is (-1.07L,
1.69L), (-0.52L, 0.52L) and (-0.52L, 0.10L) in x-, y- and z-directions, respectively. The mesh is
constituted of about 1 million cells.
For the bare hull cases, the structured grid topology is built with O-grid topology around the
hull form. The model is symmetrical about the centre plane of the ship.
Near the free surface, the more horizontal the mesh lines are, the best the free surface
tracking is. This is a real constraint for moving mesh in roll motion (the mesh is attached to
the hull during its motion).
5.4.1.2. Simulation setup

 Navier-Stokes laminar or turbulent (k-ε) model
 VOF implicit free surface model
 Imposed motion:
o On the ship surface mesh: forced roll oscillations with or without forward
speed
o The global mesh moves with the ship in the absolute frame
 Simulation duration : 6*T
 Time step = 1/200 T
 All the calculations concern the HTC ship model at scale 1/24.
 Initial conditions :
o fluid at rest
o position of the ship: starboard at roll top position
 Open boundary conditions:
o The open boundary conditions must allow to fully outlet the radiative waves
induced by the ship motion.
o The position of the outer boundary must be fixed according to the ship:
 not too close to avoid reflective effects,
 not too far to avoid a too high grid distortion
 Note that the mesh can be relatively coarse in the boundary area
 A last constraint for the choice of the numerical domain dimensions is to check that
the retained one (length, free surface position=filling rate) has its natural periods out
of the range of the considered exciting periods. The natural periods of the numerical
domain can be calculated with a potential model.

The comparison of numerical results with model tests concludes to a fair agreement even
some difficulties remain:
 Total roll moment is well predicted by CFD but derivation of coefficients encountered
some accuracy problems: how to remove hydrostatic loads, what kind of filters used
in the time series post-processing both in model tests and CFD, robustness of the
derivation method. These problems are not directly related to CFD validation
 Some calculations do not well converge and some transient behaviour can be
observed, depending on the code used
Figure 5-5: View of the HTC hull
Figure 5-6: Example of CFD/experiment comparison – time evolution of the roll moment –
imposed roll motion and forward speed

5.4.2. Case 2-b: Wave propagation w/o structure

Two types of in-coming waves are available:
 Linear mono-chromatic (so-called Airy waves or 1st order Stokes wave):
o Wave generation: An “inlet” or upstream velocity boundary condition is used
to provide velocity components based on a standard linear wave formula.
The height of the water, and hence the volume fraction, at the inlet boundary
is calculated using a second order relationship that takes account of the total
pressure
o Wave propagation: two improvements were made. Firstly, the CFD grid was
made to move according to the expected, undisturbed, propagated wave
shape with grid clustering local to where the interface is expected to be.
Secondly, a High Resolution Interface Capturing (HRIC) was implemented as
an alternative VOF scheme. Both contributed to significant increases in
accurate wave propagation
o Wave beach: The wave beach model consists of a combined grid stretch and
linear wave damping zone within the mesh model. This has been proven with
respect to the monochromatic wave problem, but further work must be
investigated for the extension to bi-chromatic and irregular waves
 Non linear mono-chromatic waves:
o Wave generation: model provided by Rienecker and Fenton theory
o A non-linear regular incident wave train is considered. To compute it, an
algorithm based on the stream function theory (Rienecker & Fenton, 1981)
has been implemented. This algorithm has been chosen as it can generate
the incident field for a wide range of depth, amplitudes and wavelengths (in
the limit of breaking waves). It is based on potential theory and gives the
numerical solution of steadily progressing periodic waves on irrotational flow
over a horizontal bed. No analytical approximations are made.
o Wave propagation: The approach, named SWENSE (Spectral Wave Explicit
Navier-Stokes Equations) in based on undisturbed ambient waves which are
modelled using fully non linear potential flow theory, while the interaction with
a ship or floating structure is accounted for in a modified Navier-Stokes
formulation in which the forcing from ambient waves is accounted for
explicitly. This method is coupled with VOF method
In a first stage, validation of wave propagation, without ship hull perturbations, consists only to
check that the perturbation flow field is nil in term of free surface elevation, pressure and
kinematics.
Then the validation will be done including a body in the flow field in different conditions
starting with simple case:
 A fixed vertical cylinder of circular cross section in weak steep waves for which an
analytical formulation exists for the diffraction field
 A fixed truncated vertical cylinder in more severe conditions : small draft / diameter
ratio, high steep waves for which model test (captive) exists
 A calm buoy in forced oscillations tests (heave, pitch) including a skirt allowing
damping effects
 Real ships in captive test of free floating and with or without a constant forward
speed.

For each configuration, validation has been performed not only on global loads induced by
waves on the hull but also on the local characteristics of the flow: velocity, pressure, vortex
field.
5.4.2.1. Monochromatic wave propagation:

 Characteristic of the generated waves:
o Crest to trough height: 0.1m
o Wavelength: 1.56m
o Period: 1s
 Initial conditions: Free surface elevations, velocity components and pressure are
obtained with the initial solution provided by Rienecker & Fenton.
 Boundary conditions:
o lateral boundaries : slip condition
o inlet and outlet boundaries : periodic condition. This condition allows to have
reasonable size of the fluid domain. The only constraint is that the domain
dimension in the propagation direction must be proportional to the wave
length.
 Mesh characteristics: The mesh includes (157*5*71) cells in (X, Y, Z).
 Time domain simulation
o Time step : Δt=0.02s
o Simulation duration : 500Δt=10s=10T with T=incoming wave period
Results: the free surface remains perfectly sinusoidal during the simulation, with amplitude
strictly included between –0.05m and 0.05m (initial condition). There is strictly not any loss of
the wave amplitude after 10T (500 Δt). These results validate the coupling of the Navier-
Stokes-VOF algorithm with the SWENSE method. The free surface remains without any
perturbation during the simulation.
Z
Y X
Y X
t=0.0s
0.05m
0.2
0
-0.05m
-0.2 6
Z
450dt = 9T
-0.4 460dt = 9.2T
4
470dt = 9.4T
-0.6
0.8 2 480dt = 9.6T
0.6 X 490dt = 9.8T
15dt = 0.3T
0.4 0
Y
0.2
0 -2
Figure 5-7: VOF-Swense model – validation of the wave propagation without structure

5.4.2.2. Wave diffraction on a cylinder:

The cylinder is supposed vertical and fixed on the seabed.
Characteristics of the test case

Crest to trough height 0.1 m
Wavelength 1.56 m
Period 1s
Depth 1m
Diameter 0.6m
Δt 0.02s
Simulation duration 300Δt=6s=10T
Table 5-2: wave and cylinder characteristics
Y X
Y
X
Z
0
-0.5
-1 5
6
4 0
Y X
2
-5
0
Figure 5-8: example of mesh (including 800000 cells)

t=3.6s
Pp
400
350
300
250
200
150
100
50
0
-50
-100
-150
-200
Figure 5-9: Diffracted wave and velocity fields (W component)
The figure below shows the time evolution of the drag load issued from EOLE and the
comparison with analytical results based on the potential theory.
The maximal intensity reached by the load peaks is comparable to the load value given by the
potential theory.
Diffractions loads on a vertical cylinder
EOLE Mac Camy et Fuchs

200
150
100
50
Fx (N)
0
0 1 2 3 4 5 6 7 8 9 10
-50
-100
-150
-200
Time (s)
Figure 5-10: Comparisons simulation-analytical formulation (Mac Camy & Fuchs) of the loads
applied on the cylinder

5.4.2.3. Wave diffraction on a truncated cylinder:

The truncated fixed vertical cylinder test case consists of the following properties:
Characteristics of the test case

Crest to trough height 0.237 m
Wavelength 5.15 m
Period 1.81 s
Water depth 5m
Diameter 0.5 m
Draft 0.5 m
Δt 0.02s
Simulation duration 300Δt=6s=10T
Table 5-3: Wave and truncated cylinder characteristics
Figure 5-11: View of the mesh - refinement in the free surface area
Figure 5-12: Diffracted wave and velocity fields (W component)


14
12
FFT of surge force
10
FFT surge force
Force (N)
4 /2 2
2
0
0 2 4 6 8 10 12 14 16
Frequency (Hz)
Figure 5-13: Main frequencies of the horizontal load time series
Figure 5-14: Comparisons with 1st order potential
For the vertical truncated cylinder with a diameter close to the wavelength, the comparison
st nd
with 1 order diffraction theory (panel method) is fairly good, 2 order component (2) is
nd
observed as expected but comparison with diffraction theory up to 2 order has to be done,
as for drift load (mean component)

5.4.3. Case 2-c: Calm buoy at zero-speed

The Calm Buoy configuration (CBU) is composed of a large truncated vertical cylinder of
small draft, with a circular hole at its center (so called “center-well”). The calm buoy is
equipped with a “skirt” (horizontal disk) at its bottom, which induces large vortices. Tests
concerned forced oscillations tests in still water (heave and pitch), fixed hull in waves
(captive) and free floating moored hull. Loads, wave field and motions have been compared
with existing experiments.
The figures shown below give an idea of this geometry as a full scale system and at the
model scale following tests carried out at Sirehna in EXPRO-CFD.
Figure 5-15: Girasol CALM Buoy Figure 5-16: Scale Model at 1:35
The CALM buoy geometry at model scale is given hereafter:
 Diameter: 0.575m
 Draft: 0.1175m
 Water depth: 5m
 Skirt Diameter: 0.675m
 Skirt Depth: 0.0925m
 Mass: 18.98 kg

2
Mass Inertia (XX&YY): 0.313 kgm
In these best practice guidelines, some simple cases that represent the fundamental flow
regimes around the buoy in forced heave and pitch are presented, since for these cases, high
quality experimental results provided by Sirehna from the EXPRO-CFD project do exist.
5.4.3.1. Forced Heave Motions

Simulations have been done with each CFD model and experiment for the following
parameters, and reported in VIRTUE deliverable D2.3.4:
Case Heave
2.1a 2.1b 2.2a 2.2b
Freq (Hz) 0.973 0.973 0.843 0.843
Amplitude 0.01 0.035 0.019 0.035
Case 2706OT22 2706OT9 2706OT17 2706OT11
Table 5-4: Heave motion parameters for the calm buoy

Calculations for forced heave were carried out using both the Fluent and the CFX11 CFD
codes. In both cases, the assumption was made that a 2D axi-symmetric geometry could be
applied. The grid used for Fluent is shown below.
Figure 5-17: 2d axi meshing distribution

The mesh in the region of the buoy was uniform and rectangular, with cell dimensions of 2mm
square throughout. Sliding boundaries were inserted at the skirt edge and at the junction with
the central column within the well. A total of 134,638 cells were used in the 2D domain for the
calculations shown here.
The following are the key features of the solution scheme used in Fluent:
 Pressure based segregated solver

 Unsteady with 1st order implicit time marching scheme
 Multiphase VOF model with surface tension (CICSAM for volume fraction
discretisation scheme)
 Turbulence Models available: RKE, SST, RSM
 Discretisation scheme
o PRESTO (Pressure Staggering Options) for Pressure discretisation scheme
o RKE and SST : QUICK scheme for all variables
o RSM : 2nd order upwind scheme for all variables
 Dynamic Mesh with Layering method

With CFX11, the key parameters were similar except that:
 A second order time marching scheme was used,

 Mesh motion algorithm is based on grid distortion.
 Solver uses a coupled pressure-velocity algorithm.
The following plots show a comparison of the CFX and Fluent simulation results for case 2.2b
above (amplitude = 35mm, frequency = 0.843Hz). The reaction force plotted represents the
net variation in the fluid reaction in heave following deduction of the time varying hydrostatic
force, and so represents only the effects of wave radiation, inertia and viscous effects mainly
arising from the presence of the skirt.
Heave force - 2706OT11
CFX11
45 experiment
Fluent
35
25
15
5
Force (N)
5 6 7 8 9 10
-5
-15
-25
-35
-45
Time (s)
Figure 5-18: Comparison of CFX11 and Fluent simulations with experiment.

It would appear that both CFD codes give good quality CFD predictions. The similarity in their
levels of grid resolution, numerical schemes, and turbulence modelling approaches is likely to
be the chief explanation, and it is comforting to find that both codes behave well for these
cases. A comparison of the FFT for the heave forces again confirms that both CFD codes
can provide good agreement as shown in Figure 5-19 below.
Heave force FFT analysis - 2706OT11
100
CFX11
experiment
Fluent
10
1
0 1 2 3 4 5 6
Magnitude
0.1
0.01
0.001
Frequency (Hz)
Figure 5-19: Comparison of Heave Force FFT for CFX11, Fluent and Experiment.

5.4.3.2. Forced Pitch Motions

As with the heave example given above, we have chosen a relatively simple test case with
which to illustrate current best practice and the state of the art. In this case, the geometry was
represented in CFX11 by using a full 3D model representing 180 degrees of arc. The cells
closest to the buoy surface were 2mm square, and in this case a total of 860,778 cells were
used. The domain was 5m maximum in the radial direction, and the water depth is 2.1m as in
the forced heave case.
The standard CFX11 coupled solver (the only option) was used throughout. The spatial and
time discretisation schemes are both second order, and the free surface VOF method uses an
interface compression algorithm similar to HRIC and CICSAM. We again used the Menter
SST with curvature correction to model turbulence. Mesh movement was handled through a
mesh distortion approach which can be summarised as:
 The grid cells adjacent to the buoy, out to the edge of the skirt, move with it,
 The grid cells in the far-field (in this case 0.5m from the centreline), remain fixed,
 The grid nodes between these two regions are moved according to a simple linear
interpolation algorithm.
The CFX solver accounts for grid movement through adjusting the conservation equations for
the cell face velocities. It is important to note that the calculation of the derivatives of these
velocities must be to the same order as the spatial discretisation scheme used in the
momentum and mass conservation equations.
In VIRTUE, a range of frequencies and amplitudes were used in the calculations, the full
results from which can be found in D2.3.4, and shown below:
Freq (Hz) 0.973 0.973 0.843 0.843
Amplitude (degrees) 3.0 5.5 3.0 5.5
Table 5-5: Pitch motion cases definition
The results for the frequency of 0.973 Hz and 5.5 degrees amplitude of motion are given
below.
Pitching torque - 2506OT3
CFX11
5 experiment 8
Angular displacement
4
6
4
2
1
Torque (Nm)
2
Angle (deg)
0
2 3 4 5 6 7 8 9 10 11 12 13
0
-1
-2
-2
-3
-4
-4
-5 -6
Time (s)
Figure 5-20: Example of comparison CFD/experiment – evolution of torque – pitch 5.5° -

Freq=0.973 Hz

The agreement between experiment and computation in these calculations appears to be of

high quality, and certainly sufficient given the experimental uncertainties to conclude
positively on the validation of the method. The plot below shows clearly the evolution of
vortex pairs from the skirt.
Figure 5-21: Velocity vectors on the centreline
The solution illustrates clearly the role of the skirt in generating and shedding vortices as a
damping mechanism.
5.4.3.3. Summary of CALM buoy best practice

These two calculations have illustrated how for two simple test cases, and when high quality
experimental data is available, CFD tools can be used to calculate with good precision the
hydrodynamic reaction forces due to forced motions.
This test example has been used in VIRTUE for the purpose of benchmarking other CFD
tools, and for the calculation of both wave diffraction (with the buoy fixed), and free floating
response in regular waves. The free-floating case represents some challenges that take this
example out of the domain of the fundamental flow regimes and precise validation shown
above, and so is not used as an example of best practice. However, readers wishing to try
the free-floating cases may refer to Virtue deliverable D2.3.4 for more details.

5.4.4. Case 2-d: LNG Carrier at zero-speed

Model tests were performed by MARIN for SBM on a moored LNG carrier (a part of Virtue).
CFD test cases concern wave diffraction on the fixed ship (captive tests) and free floating
conditions, including the “linear” mooring system. Loads, wave fields and motions are
compared.
This case considers the motions of a model scale LNG carrier in the condition that it is
moored in shallow water and subject to wave action. Experimental test data are available
from work carried out at MARIN, giving detailed time histories of response in regular head
seas at the equivalent of around 1.0m full scale amplitude. The key aspects of this test case
are:
 The vessel has a relatively complex geometry that is typical: a bulbous bow, above
water bow flare, and transom stern with a bulbous propeller shaft housing.
 It has a low stiffness mooring system for surge and sway motions that presents a
challenge with respect to slow drift motions and long natural periods.
 Moored in shallow water, there are sea-bed interaction effects and wave kinematics
modelling issues that represent a good test for the CFD calculations.
 There are extensive experimental data available for vessel motions, and also wave
elevations in the proximity of the bow region.
The main characteristics of the vessel at full scale are:
 Length BP: 274 m
 Beam: 44.2 m
 Depth: 25.0 m
 Draft: 11.0 m
 Displacement: 99,210 tonnes
 LCG: 136.10 m
 KG: 16.3 m
 Transverse GM: 4.80 m
 Roll Radius of Gyration: 15.03 m
 Pitch Radius of Gyration: 68.29 m
 Yaw Radius of Gyration: 68.15 m
 Roll Natural Period: 15.90 m
The mooring system consists of (at full scale) four identical pre-tensioned lines. Each line
length is equal to 297m, with a stiffness of 128kN/m and with a pre-tension of 2967kN. The
fairlead and anchor positions relative to the aft perpendicular and with a datum at the keel
level are given in the table below.
Line Fairlead Anchor

X Y Z X Y Z
1 283.23 0.0 25.0 356.20 -210.0 14.0
2 283.23 0.0 25.0 356.20 210.0 14.0
3 -9.93 0.0 25.0 -356.20 210.0 14.0
4 -9.93 0.0 25.0 -356.20 -210.0 14.0
Table 5-6: Mooring systems for the LNG carrier

Test cases have been computed with four different CFD codes:
Code User Mesh Wave model Free surface

CFX Atkins Unstructured Wave-maker VOF
EOLE Principia Structured Swense VOF
ISIS ECN unstructured Imposed kinematic VOF
ICARE ECN structured Swense Mesh deformation
Table 5-7: LNG carrier simulation methods used by different participants
5.4.4.1. LNG - captive test

The figures shown below illustrate the LNG carrier model in the wave field, and a comparison
of the wave forces in deep and shallow water as calculated using a potential flow diffraction
model (Diodore) as comparison. The plot also shows the surge force for a single frequency
calculated using the Eole CFD code.
Figure 5-22: wave field – LNG captive test

180
Surge force - 1st harm onic am plitude
160
140
120
Force (N)
100
Diodore - H=30m
80 Diodore - infinite H
Eole
60
40
20 Shallow water
0
0,5 1 1,5 2 2,5 3 3,5 4
Period (s)
Figure 5-23: wave loads - comparisons with 1st order potential theory
Comparison of 1st order component, issued from a Fourier analysis, with diffraction theory
appears fairly good even if shallow water effect is present.

5.4.4.2. LNG – free floating in waves

Results have been also compared for the LNG moving in waves. The expected transient low
frequency component (natural frequency of the mooring) needs a larger time than in model
tests to disappear and depends on the solver. So the motions time series are filtered in order
to remove the low transient frequency component.
The test case chosen as a benchmark involved regular waves at a full scale period of 17.95
seconds and wave amplitude of 1.0m. The following plots show the time histories of surge,
heave and pitch motions for the LNGC in these waves.
0.04
CFD-Eole/Principia
Surge at model scale (m) CFD-CFX/Atkins
0.03 Experiments
CFD-ISIS/ECN
0.02
0.01
0.00
10 15 20 25 30 35
-0.01
t (s)
-0.02
-0.03
0.020 Experiments / SBM

Heave (m)
CFD-CFX / Atkins
0.015 CFD-Eole/Principia
CFD-ISIS/ECN
0.010
0.005
0.000
5 10 15 20 25 30 35
-0.005
t (s)
-0.010
-0.015
2
Experiments
Pitch at model scale (°) CFD-Eole/Principia
1.5
CFD-CFX/Atkins
CFD-ISIS/ECN
1
0.5
0
0 5 10 15 20 25 30 35
-0.5
t (s)
-1
-1.5
Figures 5-24: surge, heave and pitch motions at wave frequency comparison with
experiments (MT-heave, MT-WF: wave frequency component of experimental motions)

There are a number of observations that can be made:

 The surge transients as noted above take some considerable time to decay and can
have an influence on the heave and pitch response. This is shown in more detail in
D2.3.4, where some strategies to add additional surge damping were applied.
 All of the CFD codes tended to under predict motions, with the exception of Eole
which over-predicted pitch, although making good predictions of heave response.
 All codes reproduced amplitudes of motion that were within 20% of the experimental
values.
It is expected that further improvements to accuracy could be made in the case of CFX11
which used around 1.6 million nodes, and Eole which was of around the same size of model.
It should also be noted that predictions made using conventional linear theory within a time
domain simulation, also tend to under predict responses to about the same degree as the
CFD simulations by comparison with experiment. The experiments themselves contain some
degree of uncertainty in relation to the calibration of the wave tank and the expectation that
the ship model experiences the prescribed wave heights. Measurements in the MARIN tank
suggest that this uncertainty could be around 10% in wave height. We conclude from this that
the absolute level of accuracy of CFD for these simulations is difficult to quantify exactly, but
is in the range 10 to 20% in this case.
The LNGC hull form, along with the test conditions, can be made available for those wishing
to repeat the above calculations by accessing the Virtue web site.

5.4.5. Case 2-e: HTC with forward speed in waves

The HTC at model scale is considered.
 The forward ship velocity is 1.89 m/s.
 The time step is 0.01 s.
 The wave characteristics are:
w / Lpp  1
 w / Lpp  0.008
with the ζw amplitude of the wave (between the undisturbed free surface at rest to the
maximum wave height).
5.4.5.1. Captive test case
Figure 5-25: HTC in waves at t=8.5s (top) and t=8.8s (bottom) – captive test – forward speed
1.00E-01
8.00E-02
6.00E-02
4.00E-02
2.00E-02
Exp
X (-)
0.00E+00
EOLE
1 2 3 4 5 6 7 8
-2.00E-02
-4.00E-02
-6.00E-02
-8.00E-02
-1.00E-01
Time (s)
Figure 5-26: comparison of loads – HCT in waves – captive test – forward speed

The numerical model allows getting reasonable results in amplitude. However a difference on
the mean values can be observed. It probably comes from the hydrostatic part of the signals.
EOLE time series appear more symmetrical than the experimental signal.
5.4.5.2. Free-floating model with forward speed

The hull is free to move in pitch and heave.
X
1.00E-01
8.00E-02
6.00E-02
4.00E-02
2.00E-02 Exp
EOLE
X (-)
0.00E+00
2 3 4 5 6 7 8 9 10
-2.00E-02
-4.00E-02
-6.00E-02
-8.00E-02
-1.00E-01
Time (s)
Figure 5-27: comparison of loads – HTC free floating – forward speed
The load amplitude is quite correctly reproduced by the numerical model but the computation
put into evidence a second order frequency, in addition to the main frequency, which is not
visible in the experiments.
Note that, disregarding the peaks related to this second frequency, both signals show a
similar slope: a progressive increase followed by a sharp decrease.

6. Application 3: Numerical Manoeuvring tank
6.1. Theoretical aspects of Manoeuvring
6.1.1. Introduction
Ship manoeuvring is the controlled change or retention of the direction of motion of a ship and
its speed in that direction. Assessment of the manoeuvring performance of a ship can be
done using full-scale trials, model tests or by simulations. In most cases, these assessment
studies consist of comparing properties obtained through performing standard manoeuvring
with requirements posed by ship owners, shipyards or other criteria, such as those posed in
IMO Resolution MSC. 137(76). The standard manoeuvres comprise:
 Zig-zag manoeuvres, used to determine the initial turning ability and the yaw
checking and course keeping abilities;
 Turning circle manoeuvres, used to determine the turning ability;
 Stopping manoeuvres, used to determine the stopping ability.
The present guidelines apply to predicting the manoeuvring performance using CFD based
simulations only. For guidelines regarding model tests or manoeuvring simulations in general,
see e.g. the ITTC Recommended Procedures and Guidelines, Testing and Extrapolation
Methods, Manoeuvrability, in particular the Captive Model Test Procedures (7.5-02-06-02)
and the Free Running Test Procedures (7.5-02-06-03).
The aim of the Virtual Manoeuvring Basin within VIRTUE is to develop and utilise state-of-the-
art tools that can numerically simulate approaches otherwise followed during model tests.
Therefore, a brief outline of the test procedures and approaches is given.
6.1.2. The approaches of manoeuvring model tests

There are two main types of manoeuvring model tests. One is captive model tests, where the
model is restrained. The motion of the model is prescribed and the loads are measured
through load balances. Based on the loads, a mathematical model is derived which is used to
simulate the manoeuvring behaviour of the ship. The other type of manoeuvring tests is the
free running model test where the model is driven by operating propellers and the rudders or
other control devices are acting to control the direction of motion. The geometric and
kinematic characteristics are recorded through special instruments.
6.1.2.1. Captive model tests

These tests are conducted by forcing the model in predefined motions and measuring the
hydrodynamic loads. Based on the tests, a mathematical model is derived. This mathematical
model, consisting of a set of equations of motions, describes the loads on the ship as a
function of the instantaneous position (attitude) and velocity. The mathematical model is
implemented in a simulation program in which the equations of motions are used to calculate
the accelerations of the ship in time. By integrating the accelerations velocities are obtained
and by integrating the velocities, the actual position of the ship is found. The evolution of the
position in time results in the actual manoeuvre.

The most popular captive model tests include: straight line or drift tests, rotating arm (RAT) or
circular motion (CMT) tests and planar motion mechanism (PMM) or computerised planar
motion carriage (CPMC) tests.
In straight line tests, the velocity dependent derivatives at any draft and trim can be
determined. The dynamometers measure the forces and moments experienced by the model
at different drift angles.
In rotating arm tests, a yaw velocity r is imposed on the model by rotating it in a circle around
a fixed axis. The model can be oriented such that the local drift angle at the reference point is
zero (pure yaw test) or such that a combination of yaw and drift is experienced by the model
(combined test).
In PMM tests, oscillatory sway and oscillatory yaw motions are executed. This can be done
without drift (pure sway or pure yaw) or combined with a drift angle (combined test).
All tests can be conducted for the bare hull but also be conducted for the appended hull
combined with variation of e.g. the angle of the rudders or with variation of the propeller RPM.
For ships with large changes in attitude during the manoeuvres (e.g. high-speed craft or ships
with low GM values), part of the test program needs to be repeated for the expected range of
attitudes (heel or trim angles and sinkage values) in order to obtain accurate coefficients for
these conditions as well.
Since the acceleration of the centre of gravity of the model during RAT, CMT, PMM or CPMC
tests is non-zero, the measured forces have to be corrected for the inertial forces in order to
obtain the hydrodynamic loads on the model.
After the tests, the hydrodynamic loads are analysed, see below, in order to obtain the
hydrodynamic derivatives for the mathematical model that is needed to simulate the
manoeuvres. To establish the mathematical model, the loads as a function of e.g. the drift
angle, yaw rate, sway acceleration, yaw acceleration or combinations thereof are required.
There are different possibilities to obtain these relations, such as for example:
 Using combined tests in which several drift angles are tested at a constant yaw rate
(or yaw rate amplitude during oscillation tests), the influence of the drift angle on the
loads can be derived.
 Using pure drift tests, the influence of the drift angle on the loads can be derived.
 Using pure sway oscillation tests, the influence of the sway velocity amplitude (and
subsequently the drift angle amplitude) can be derived.
The advantage of the first option is that a combined motion in general can be closer to the
condition experienced during the actual manoeuvre than the pure drift motion which is hardly
encountered (except when external forces such as wind act on the ship or during crabbing
operations). The third option has the advantage that besides the influence of drift angle also
the influence of the sway acceleration (sway added mass) is obtained. The second option is
however easiest to conduct and analyse.
6.1.2.2. Free running model tests

The widely used free running model tests include turning tests and zigzag tests. During these
tests, the configuration of the prototype ship (loading condition, appendage configuration) is
directly reflected in the ship model.
In turning tests, the trajectory of the model is recorded after its control device (e.g. rudder,
pod, thruster) is deflected and held at a fixed angle.
In zig-zag tests, the rudder or other control device is deflected to a pre-selected rudder angle
and held until a pre-defined change of heading angle is reached. Then, the control device is

deflected to an opposite angle and held until the heading check angle on the opposite side is
reached. This is repeated several times.
During the free running tests, the time traces of the position and velocity of the model and the
control settings are recorded. These time traces are analysed in order to establish the
manoeuvring characteristics of the model.
6.1.3. Virtue manoeuvring towing tank as a complementary tool

In the VIRTUE Virtual Manoeuvring Basin the same distinction between captive manoeuvres
and free running manoeuvres is made. These approaches are designated as follows:
6.1.3.1. Derivative approach:

In the derivative approach, the hydrodynamic loads and derivatives are calculated for
predefined manoeuvring motions, either steady or harmonically varying. The following basic
motions are emulated:
 Steady oblique motion (pure drift)
 Steady turning motion (pure yaw)
 Oscillatory sway motion (pure sway)
 Oscillatory yaw motion (pure yaw)
 Combined motion (yaw rate and drift)
6.1.3.2. Direct numerical manoeuvring simulation

In this task, free sailing simulations will be realised by solving RANSE coupled with ship
motion. Two environment conditions are considered:
 Unsteady manoeuvring ship simulation in calm water
 Unsteady manoeuvring ship simulation in waves
6.1.4. Coordinate system and nomenclature

Generally, the origin of the right-handed system of axes used in manoeuvring studies is
located at the intersection of the waterplane, midship and centre-plane, with x directed
forward, y to starboard and z vertically down. The forces and moments in the three directions
are designated as the forces X, Y and Z and moments K, M and N. The ship's total velocity in
the centre of origin is designated Vs and velocities in x, y and z directions are designated u, v
and w. Rotations around these axes are designated respectively p, q and r.
The drift angle  is defined as = -atan v/u and the non-dimensional yaw rate  as =
r·Lpp/Vs.
6.1.5. Equations of motion

Basic manoeuvring studies consider the equations of motions of a manoeuvring surface ship
ship in four degrees of freedom: surge, sway, roll and yaw. For ships with large changes in

running trim, also equations for heave and pitch need to be included. For ships with very large
GM values or ships sailing at very low speeds, the roll motions may be neglected.
The pertinent equations of manoeuvring motion in the ship fixed reference system could be
written as below, assuming symmetry with respect to the centreplane of the ship (y G = 0) and
neglecting pitch and heave:
X  m  u  vr  xG r 2  zG pr 
Y  m  v  ur  zG p  xG r  (76)
K  I xx p  mzG  v  ur 
N  I zz r  mxG (v  ur )
in which X, Y, K and N are the hydrodynamic forces or moments acting on the ship and I xx
and Izz are the moments of inertia around the x and z axes respectively. The hydrodynamic
loads can be obtained either by derivative approach or direct numerical simulation.
Sometimes, it can be beneficial to divide the hydrodynamic loads into separate components,
such as components acting on the bare hull (subscript H) and on the separate appendages
(e.g. propeller with subscript P and rudder with subscript R), as follows:
X  XH  XP  XR (77)
It is assumed that interactions are included in the separate components. When correctly
modelled, the influence of a change in e.g. rudder size can be studied without changing the
descriptions of the bare hull forces and propeller forces. Furthermore, this approach allows a
hybrid combination of using CFD to estimate components of the mathematical model while
the other components are estimated using e.g. empiric formulae.
Sometimes the equations of motions are written in non-dimensional form to allow for easier
comparison of the hydrodynamic coefficients between different ships.
Different methods are utilised, but within VIRTUE the following non-dimensionalisation is
used:
 the longitudinal force X and transverse force Y are made non-dimensional using
1
2
Vs 2 Lpp T (78)
 the vertical force using

1
2
Vs 2 Lpp B (79)
 the heeling moment K by

1
2
Vs 2 Lpp T 2 (80)
 the pitch moment M by

1
2
Vs 2 Lpp 2 B (81)
 the yaw moment N by

1
2
Vs 2 Lpp 2 T (82)
In this section, pressure components are indicated with subscript p and friction components
with subscript f.

6.2. Applications Cases: Derivative Approach

The following sections provide application cases representative for manoeuvring studies
when using the derivative approach. The examples are provided by MARIN.
6.2.1. Introduction
Numerical calculations of the flow around a ship in steady motion using a RANS solver are
presented. The loads on the ship are predicted and the corresponding hydrodynamic
derivatives are derived.
All calculations were performed with the MARIN in-house flow solver PARNASSOS, which is
based on a finite-difference discretisation of the Reynolds-averaged continuity and
momentum equations, using fully collocated variables and discretisation. The equations are
solved with a coupled procedure, retaining the continuity equation in its original form. The
governing equations are integrated down to the wall, that is, no wall functions are used.
For all calculations, the one-equation turbulence model proposed by Menter (1997) was used.
The Spalart correction of the stream-wise vorticity was included (Dacles-Mariani et al. 1995).
Appendages were not present during the bare-hull tests and therefore were not modelled in
the calculations. The calculations were conducted without incorporating free-surface
deformation. Based on the speeds used during the tests for these ships and the range of drift
angles studied, the effects of speed and free-surface deformation on the forces on the
manoeuvring ship are likely to be small.
The application case considered is the bare hull Hamburg Test Case (HTC) at model scale.
The experiments were conducted by HSVA. During the experiments, the loads on the model
were measured. Furthermore, the flow field around the hull at two different cross sections was
obtained during a CPMC test through Particle Image Velocimetry (PIV).
6.2.2. Grid
The results presented in these guidelines were all obtained on structured grids with H-O
topology, using grid clustering near the bow and propeller plane. The six boundaries of the
computational domain were as follows: the inlet boundary was a transverse plane located
upstream of the forward perpendicular; the outlet boundary was a transverse plane
downstream of the aft perpendicular; the external boundary was a circular or elliptical cylinder
for the drift cases and doughnut shaped for the rotation or combined motion cases; the
remaining boundaries were the ship surface, the symmetry plane of the ship or coinciding
block boundaries, and the undisturbed water surface.
The grid consisted of an inner block and an outer block, see Figure 6-1. The inner block was
the same for all calculations and the outer block could deform to allow for the drift angle, the
rotational motion of the ship, or both. Therefore, grids for various manoeuvring motions could
be made efficiently.

Figure 6-1: Impression of inner and outer blocks (coarsened for presentation)
for the drift angle case
To incorporate the manoeuvring motion of the ship, the inner block is rotated around the
vertical z-axis over the desired yaw angle. Then the outer block is generated around the inner
block. The cell stretching used in the inner block is automatically applied to the outer block as
well. Matching interfaces between the blocks are used so that the inner and outer blocks can
be merged. The size of the outer blocks is chosen such that the rotated inner block can
smoothly be incorporated in the outer grids. This means that increasing drift angles and/or
yaw rates will result in wider domains. The size of the domain is based on the use of a solver
for potential flow to calculate the velocities in the inflow and external planes. Before starting
the calculations, the separate blocks are merged into one block for the port side of the ship
and another block for the starboard side of the ship.
The domain sizes (multiples of Lpp) and variation in the number of grid nodes in the stream-
+
wise, normal and girth-wise (n, n and n) directions are given, together with the maximum y
+
value for the cells adjacent to the hull, designated y2 , that was obtained during the
calculations.
+
Re  xin xout |y|max zmin n n n points y2
6
0° -0.70 0.75 0.16 -0.16 329 71 51 1.2×10 0.56
6
2.5° -0.71 0.76 0.35 -0.31 377 95 51×2 3.7×10 0.74
6 6
6.3×10 5° -0.71 0.79 0.38 -0.31 377 95 51×2 3.7×10 0.81
6
10° -0.72 0.91 0.45 -0.31 377 95 51×2 3.7×10 0.90
6
15° -0.73 1.11 0.52 -0.32 377 95 51×2 3.7×10 0.90
Table 6-1: HTC, Ta=Tf = 0.429 m, steady drift

At the ship surface the no-slip condition is applied directly and the normal pressure derivative
is assumed to be zero. The undamped eddy viscosity, the variable in Menter's one-equation
model (1997), vanishes at a no-slip wall.
Symmetry conditions are applied at the undisturbed water surface and on the ship symmetry
plane (for the zero-drift conditions). At the inlet boundary, the velocity profiles are obtained

from a potential flow solution, which also determines the tangential velocity components and
the pressure at the external boundary. At the outlet boundary, stream-wise diffusion is
neglected and the stream-wise pressure derivative is set equal to zero.
The lift generated by the hull form is modelled in the potential flow solution by applying a
vortex sheet on the symmetry plane of the ship. At the stern of the ship, the Kutta condition
(the flow leaves the trailing edge smoothly) is applied, which allows the solution of the
unknown vortex strengths on the sheet. Since the only purpose of the potential flow solution is
to set the boundary conditions for the viscous flow solution at the inlet and external
boundaries, vortex shedding from the bilges of the ship is omitted.
6.2.4. Case 3-a: Steady oblique motion

The calculated transverse forces and yawing moments are presented in Figure 6-2 and Table
6-2: . The good agreement between the CFD estimations and the experimental values is
clearly seen.
Figure 6-2: Comparison between experiments and predictions, HTC steady drift.
Y transverse force, N yaw moment, exp experimental results, cfd results based on the viscous
flow calculations, sb results based on semi-empiric slenderbody method (Toxopeus, 2006 or
Toxopeus and Lee, 2008), cfd fit results based on the mathematical model
 1
Re [-] Fn [-] Partner T.m. Xp Xf X Yp Yf Y Np Nf N
[deg]
6 Exp
6.3×10 0.132 0 - - - -0.014 - - 0.000 - - 0.000
(HSVA)
6
6.3×10 - 0 MARIN MNT -0.002 -0.012 -0.014 - - - - - -
6
6.3×10 - 2.5 MARIN MNT -0.003 -0.012 -0.014 0.008 0.000 0.009 0.006 0.000 0.006
6 Exp
6.3×10 0.132 5 - - - -0.015 - - 0.020 - - 0.012
(HSVA)
6
6.3×10 - 5 MARIN MNT -0.003 -0.012 -0.015 0.017 0.000 0.017 0.013 0.000 0.013
6 Exp
6.3×10 0.132 10 - - - -0.017 - - 0.048 - - 0.023
(HSVA)
6
6.3×10 - 10 MARIN MNT -0.003 -0.012 -0.016 0.043 0.001 0.044 0.024 0.000 0.024
6
6.3×10 - 15 MARIN MNT -0.003 -0.013 -0.016 0.079 0.002 0.081 0.037 0.000 0.037
6 Exp
6.3×10 0.132 20 - - - -0.015 - - 0.134 - - 0.048
(HSVA)
6 Exp
6.3×10 0.132 30 - - - -0.008 - - 0.244 - - 0.073
(HSVA)
Table 6-2: HTC, Ta=Tf=0.429 m, steady drift
The pressure distribution on the bow and stern areas is given in Figure 6-3. The stagnation
area with high pressure at the bow is clearly visible, as well as the low pressure areas at the
bow and stern where the flow curves around the convex hull form.

Figure 6-3: Pressure distribution on the hull, HTC, =10°
The axial velocity field at the aft perpendicular (x=-0.5Lpp) is given in Figure 6-4. The vortical
structures generated upstream at the propeller gondola and bilge keels are clearly visible.

Figure 6-4: Axial velocity contours at x=-0.5Lpp, HTC, Re=6.3×10 , =10°

6
6.2.5. Case 3-b: Steady turning motion

The calculated transverse forces and yawing moments are presented in Figure 6-5 and Table
6-3:. The good agreement between the CFD estimations and the experimental values is
clearly seen for the yaw moment. The CFD prediction of the transverse force is reasonable,
but slightly lower than the experiments. The magnitude of the Y force during pure rotation is,
however, very small and is of less significance than the other force or moment components.
Figure 6-5: Comparison between experiments and predictions, HTC steady yaw.
Y transverse force, N yaw moment, exp experimental results, cfd results based on the viscous
flow calculations, sb results based on semi-empiric slenderbody method (Toxopeus, 2006 or
Toxopeus and Lee, 2008), cfd fit results based on the mathematical model

1
Re [-] Fn [-]  [-] Partner T.m. Xp Xf X Yp Yf Y Np Nf N
6 Exp
6.3×10 0.132 0 - - - -0.014 - - 0.000 - - 0.000
(HSVA)
7 Exp
1.1×10 0.238 0 - - - -0.014 - - 0.000 - - 0.000
(HSVA)
6
6.3×10 - 0 MARIN MNT -0.002 -0.012 -0.014 - - - - - -
7
1.2×10 - 0 MARIN MNT -0.002 -0.011 -0.013 - - - - - -
7 Exp
1.2×10 0.238 0.1 - - - -0.014 - - 0.003 - - -0.004
(HSVA)
6
6.3×10 - 0.1 MARIN MNT -0.002 -0.012 -0.014 0.002 0.000 0.002 -0.003 0.000 -0.003
6
6.3×10 - 0.15 MARIN MNT -0.002 -0.012 -0.014 0.004 0.000 0.004 -0.005 0.000 -0.005
7 Exp
1.2×10 0.238 0.2 - - - -0.015 - - 0.008 - - -0.009
(HSVA)
6
6.3×10 - 0.2 MARIN MNT -0.002 -0.012 -0.014 0.006 0.000 0.006 -0.007 0.000 -0.007
7
1.2×10 - 0.2 MARIN MNT -0.002 -0.011 -0.013 0.005 0.000 0.005 -0.007 0.000 -0.007
6
6.3×10 - 0.3 MARIN MNT -0.003 -0.012 -0.014 0.009 0.000 0.009 -0.012 0.000 -0.012
7 Exp
1.2×10 0.238 0.4 - - - -0.017 - - 0.016 - - -0.018
(HSVA)
6
6.3×10 - 0.4 MARIN MNT -0.003 -0.012 -0.015 0.012 0.000 0.013 -0.018 0.000 -0.018
7
1.2×10 - 0.4 MARIN MNT -0.002 -0.011 -0.013 0.012 0.000 0.012 -0.018 0.000 -0.018
7 Exp
1.2×10 0.238 0.56 - - - -0.019 - - 0.024 - - -0.029
(HSVA)
6
6.3×10 - 0.56 MARIN MNT -0.003 -0.012 -0.015 0.015 0.001 0.016 -0.029 0.000 -0.029
Table 6-3: HTC, Ta=Tf=0.429 m, steady yaw
During the measurements, the axial velocity field at the x=-0.48Lpp plane was measured. The
experimental results and the predicted flow fields are given in Figure 6-6.The vortical
structures generated upstream at the bow, propeller gondola and bilge keels are clearly
visible. Reasonable agreement is found between the calculations and the experiments,
although some quantitative differences can be seen.

Figure 6-6: Axial velocity contours at x=-0.48Lpp, HTC, Re=1.2×10 , =0.4

7
6.2.6. Obtaining the hydrodynamic coefficients

Several different approaches exist to obtain the hydrodynamic coefficients from the
experimental data set. These comprise for example:
 Curve fit through the complete data set using a predefined mathematical model
 Separate curve fits through the results of test series (pure drift, pure yaw, combined
motion) using predefined shapes for the influence of drift, yaw and combined motion.
 A step-wise approach in which derivatives are obtained successively.
With the first approach, the total error between the experimental values and the fit can be
minimised. However, it is difficult to ensure that specific coefficients are determined with the
required accuracy. For example, the traditional linear coefficients can become inaccurate
when the complete data set consists of a large number of tests in which non-linear effects
play a significant role.
With the second approach, more control is available, while still obtaining reasonably small
errors between the fit and the experimental values.
With the third approach, complete control is available. However, sufficient experience of the
analyst is required to obtain a good balance between the control over how to derive the
coefficients and the accuracy of the fit. Furthermore, if the linear coefficients are derived for
example by only using small drift angles, the error in the measurement setup might influence
the error in the coefficients considerably.

For illustration purposes, steps that can be taken to derive the hydrodynamic coefficients
based on the pure drift and pure yaw tests using the third approach are given in this section.
For simplicity, only the bare hull components of the transverse force Y and yawing moment N
are dealt with.
6.2.6.1. Step 1: linear derivatives for drift

Forces and moments can be divided into linear part and non-linear parts, for example:
YH  YH,lin  YH,nonlin (83)
The linear coefficients for drift should represent the slope near =0° or v=0 m/s accurately.
Some publications present linear derivatives in dimensional form by:
Y H,lin  v   Yv  v (84)
Note: sometimes, one sees the non-dimensional form:

Y H,lin   Y  (85)
which is an approximation for the dimensional form for small .
The correct non-dimensional form is:

Y H,lin   Y  sin  (86)
v
since sin    (87)
Vs
One should also be careful that due to the definition of the drift angle, the signs of Y v and Y
are opposite.
Although this formulation is mathematically correct and works well for standard manoeuvres
such as zig-zag manoeuvres or turning circle manoeuvres, problems may arise when the
mathematical model is used in a simulator for e.g. harbour manoeuvres: for large drift angles,
large linear contributions will be modelled, while physically, this linear contribution should
vanish. Furthermore, the linear contribution does not depend on the forward speed.
This can be realised by adopting the following representation:
 dimensional
Y H,lin u,v   Yuv  uv (88)
 non-dimensional
Y H,lin   Y  cos  sin  (89)
In both formulae, the linear contribution vanishes for →90°.
In simulators, astern speeds may be encountered by the ship. In these situations, it is

especially important that the sign of the forces and moments is correct. This required further
modification of the mathematical model, resulting in the following non-dimensional linear
contributions for the transverse force and yawing moment:
Y H,lin   Y  cos  sin  (90)

and
NH,lin   N  cos  sin  (91)
To derive Y and N, the measured forces and moments obtained during the steady drift
calculations are divided by cossin and plotted against , see Figure 6-7. The intersection of
the fit through the calculations with =0° represents the value of the linear coefficient. It is
noted that some scatter exists and that for Y the curve changes considerably as a function of
 compared to the curve for N. This already indicates that for Y a considerable non-linear
contribution can be expected, which can also be seen in Figure 6-2.
The scatter in the data will introduce uncertainty in the obtained value of the linear
coefficients.
Figure 6-7: Deriving the linear coefficients for 
6.2.6.2. Step 2: linear derivatives for yaw rate

Similar considerations apply to the linear contribution of the yaw rate. These considerations
lead to the following mathematical models:
Y H,lin     Y    cos  (92)
and
NH,lin     N    cos  (93)
Division of the calculated results for steady turning motion by leads to the graphs presented
in Figure 6-8. Again, the change of Y as a function of  is more non-linear than the change in
N. In this case, the scatter is also less for N than for Y, resulting in a relatively large
uncertainty in the linear coefficient Y.

Figure 6-8: Deriving the linear coefficients for 
6.2.6.3. Step 3: non-linear derivative for pure sideway motion or zero-speed

rotation
In order to be able crabbing manoeuvres (e.g. moving sideways during berthing), so called
cross-flow drag coefficients are required. For the transverse force, this can be modelled as
follows:
YH,nonlin   90  Y   sin  sin  (94)
For turning on the spot, the non-dimensional yaw rate will become infinite due to the velocity
approaching zero. Therefore, this should be handled in dimensional form, through:
NH,non-lin,dimensional  V  0   Nr r  r r (95)
In the present example, calculations for pure sideway motion and turning on the spot have not
been performed due to the unsteady nature of these manoeuvres. In such cases, the
coefficients can be determined using empirical relations, such as for example given by Hooft
(1994).
6.2.6.4. Step 4: non-linear derivatives for drift

The non-linear coefficients for drift can be determined using the results of the steady drift
calculations, using a mathematical model similar to
YH    Y  cos   sin   Y   sin   sin   Yab  cos y   sin y   signsin 

a b
(96)
and
NH   N  cos   sin   Nab  cosan   sinbn   sign  cos   sin   (97)
in which Y, Y|| and N have been determined in steps 1 and 3. The coefficients Yab and Nab
are determined in this step and describe the non-linear behaviour of the force and moment as
a function of the drift angle.
The powers ay, by, an and bn of the cosine and sine functions are determined based on the
calculation results and on which shape of the force/moment distribution for large drift angles is
desired. In general, these values range between 1 and 3. Using integer values is
recommended. The powers ay and by (or an and bn) cannot simultaneously have a value of 1,
since then these terms describe the linear contribution of the drift angle.

The terms Yab and Nab can be determined by subtracting the linear terms and the pure
sideway motion terms from the calculation results and subsequent fitting of the data.
Figure 6-9: Final fit for pure drift
6.2.6.5. Step 5: non-linear derivatives for yaw

The non-linear derivatives for yaw are determined similar to the non-linear coefficients for
drift. However, for the transverse force, a non-linear coefficient does not appear to be
necessary in the present case. For the yaw moment, the following mathematical model can
be used:
N    N'  cos     N'uc  cos    cn  sign   N'       (98)
in which N and N|| have been determined in steps 2 and 3. The power c n of the cosine
function is determined based on the calculation results and on which shape of the
force/moment distribution for large yaw rates is desired. In general, c n ranges between 2 and
3. Using integer values is recommended. The power c n cannot have a value of 1, since then
this term describes the linear contribution of the yaw rate.
The term Nuc can be determined by subtracting the linear terms and the zero-speed rotation
terms from the calculation results and subsequent fitting of the data.
Figure 6-10: Final fit for pure yaw (left: without Nuc, right: with Nuc, cn=2)

6.2.6.6. Step 6: other non-linear derivatives for combined motion

Having determined the linear and non-linear coefficients for pure drift and pure yaw motion,
the complete mathematical model including combined motion terms can be specified as
follows:
YH  Y  cos   sin   Y  cos     Y   sin   sin 
a b
(99)
 Y      Yab  cos y   sin y   signsin 
and
NH  N  cos   sin   N  cos     Nuc  cos    cn  sign   N     
 
 N    N    signcos   
 Nab  cosan   sinbn   sign  cos   sin  

(100)
Depending on the ship and computational results, it may be necessary to include different or
additional cross-terms.
In the present case, the coefficients Y||, N and N are obtained by regression of the data
after subtracting all known linear and non-linear contributions.
Figure 6-11: Final fit, including combined motion
6.2.7. Other available cases suitable for application and validation

In literature, several other concise datasets regarding steady oblique motion are available,
such as Series 60 (Longo J.F., 1996), KVLCC2M (Kume et al., 2006), DARPA SUBOFF
(Huang et al., 1992) and DARPA SUBOFF (Roddy, R.F.,1990).

6.2.8. Case 3-c: PMM Simulations - Unsteady forced motion

This part presents firstly the analysis and methodology used in the treatment of Planar Motion
Mechanism (PMM) simulation to evaluate linear coefficients introduced in mathematical
manoeuvrability models. Then numerical results on PMM simulation of forced sway and yaw
motion will be compared with experiments.
6.2.8.1. General analysis and methodology

PMM simulations and experiments processes consist in imposing an oscillatory motion to the
ship on only one degree of freedom in order to identify with harmonic analysis of the efforts
some linear coefficients used in mathematical models of manoeuvrability.
The basic dynamics of manoeuvring can be described and analyzed using Newton’s
equations of motion. Basic equations in the horizontal plane can be considered first with
reference to one set of axes fixed relative to the earth, and a second set fixed relative to the
ship.
Writing Newton’s law in the inertial (non–accelerating) reference frame (x0, y0) we have:
mx0  X 0 ,
my0  Y0 ,
I z  N ,
(101)
where:
X0 Total force in the
x0 direction,
Y0 Total force in the

y0 direction,
N Turning moment around the

z0 axis,
m Ship’s mass,
Iz Mass moment of inertia of the ship about the
z0 axis,
 Yaw angle measured with respect to the

x0 axis.
In spite of the apparent simplicity of equations (101), the motion of a ship is more
conveniently expressed when referred to the (x,y) system of coordinates fixed with respect to
the moving ship. The ship fixed reference frame is a right hand frame, with the x–axis pointing
in the longitudinal direction, the y–axis positive starboard, and the z–axis positive down. We
can transform coordinate systems using:

X  X 0 cos  Y0 sin ,
Y  Y0 cos  X 0 sin ,
x0  u cos  v sin ,
y0  u sin  v cos ,
(102)
Differentiating and substituting into equations (101) we obtain, using the fact that   r ,
mv  mur  Y , Sway equation (103)
I z r  N , Yaw equation (104)

Where v and r are the sway, and yaw velocities of the moving ship.
Considering little amplitude motions, we may decompose Y and N in first order Taylor series
with independent variables position and ship’s velocity. For simplification reasons we consider
only parameters who have the larger effects on the efforts, that is to say v , v , r and r .
Thus, we obtain:
Y  Y0  Yv v  Yv v  Yr r  Yr r
N  N0  Nv v  Nv v  N r r  N r r
(105)
A
Az 
With in general, z
Substituting into equations (103) and (104) we obtain:

Yv v   m  Yv  v   mu  Yr  r  Yr r  Y0
 N v v  N v v  N r r   Iz  N r  r  N 0
(106)
Forced motions experiments consist in imposing oscillatory motion to the ship about one
degree of freedom and identifying terms of the efforts Taylor development.
x  x0 sin t 
Considering , an imposed motion about one degree of freedom, we obtain by
using simple velocity and acceleration derivatives:
x  x0 sin t 
x  x0 cos t 
x   x0 2 sin t 
(107)
Doing a k-order harmonic decomposition of efforts due to forced motion at  frequency, we
obtain for any component F:
F   Fk sin  kt  k 
k (108)

Taking into account only the first order and using sine and cosine decomposition, we have:
F  F0  F1 sin t  cos 1   F1 cos t  sin 1 

(109)
F1x  F1 cos 1  F1x  F1 sin 1 
Considering and terms in phases respectively with
acceleration and velocity, we obtain:
F  F0  F1x cos t   F1x sin t 

(110)
6.2.8.2. Case of the sway forced motion

Sway forced motion is obtained imposing oscillatory sway motion:
y  y0 sin t 
(111)
We can easily determine sway velocity and acceleration:
v  y0 cos t 

v   y0 2 sin t 
(112)
Using the first order linearisation of the sway and yaw equations and the first order Taylor
series decomposition of the sway and yaw efforts, we have in case of a pure sway motion
(r r 0)
Yv v   m  Yv  v  Y0
 Nv v  Nv v  N0
(113)
And with the first order harmonic decomposition of

Y0 , we have:
Yv y0 cos t    m  Yv  y0 2 sin t 

= Y1v cos t   Y1v sin t 
 N v y0 cos t   N v y0 2 sin t 
 N1v cos t   N1v sin t 
(114)
By identification of the sine and cosine terms respectively in right and left part on the
equation, we can determine linear coefficients for sway motion.
Y1v N1v
Yv   Nv  
y0 y0
Y1v N1v
Yv  m Nv 
y0 2 y0 2
and (115)

We may notify that linear coefficient of sway motion Yv is composed with a term from
Y1v
y 2
hydrodynamics ( 0 ) and the mass m.
6.2.8.3. Case of the yaw forced motion

Yaw forced motion is obtained imposing oscillatory yaw motion:
   0 sin t 
(116)
We can easily determine yaw velocity and acceleration:
r   0 cos t 
r   0 2 sin t 
(117)
Using the first order linearisation of the yaw equation and the first order Taylor series
decomposition of the yaw efforts, we have in case of a pure yaw motion ( v v0)
 mu  Yr  r  Yr r  Y0
 N r r   Iz  N r  r  N 0
(118)
The first order harmonic decomposition of is written:
 mu  Yr  0 cos t   Yr 0 2 sin t 

 Y1r cos t   Y1r sin t 
 N r 0 cos t    Iz  N r  0 2 sin t 
 N1r cos t   N1r sin t 
(119)
By identification of the sine and cosine terms respectively in right and left part on the
equation, we can determine linear coefficients for yaw motion.
Y1r N1r
Yr    mu Nr  
 0  0
Y1r N1r
Yr  Nr   Iz
 0 2  0 2
and (120)
Y N
Linear coefficients of yaw motion r and r respectively depend of the mass m and the
inertia about axis z, Iz. These terms larger than hydrodynamics component in the measured
results from experiments introduce some difficulties when extracting the hydrodynamics
efforts from the global efforts.

6.2.8.4. Application to Hamburg Test Case for sway PMM simulation

Sway forced motion experiments consist in towing a model in a tank with a constant velocity
and imposing oscillatory motion in the ship’s lateral axis direction. These experiments have
been carried out at HSVA for the European Project VIRTUE on the Hamburg Test Case
(Vogt, 2007).
The forced motions imposed during experiments are exactly reproduced in the computations.
The longitudinal velocity u is constant and the lateral velocity v follow the oscillatory law
v  t   v '.U cos t  . The other degrees of freedom are fixed, so the ship has only the wanted
motion (sway). The parameters used for the computations are: v '  0.15 and
1
u  1.89 m.s .
The picture below shows the trajectory obtained for these parameters.
Figure 6-12: Trajectory of the ship for the sway PMM test
The charts below present some examples of the results performed with a 0.3 millions nodes.
X' Y' N'

Harmonic Amplitude Phase(Rad) Amplitude Phase(Rad) Amplitude Phase(Rad)
0 -0.935 0.00 -0.066 0.00 -0.012 0.00
1 0.013 1.76 2.370 1.13 0.768 0.15
Experiments 2 0.019 0.67 0.026 0.13 0.013 0.90
3 0.011 1.30 0.033 3.00 0.013 -0.02
4 0.000 1.99 0.000 -3.04 0.000 -1.88
0 -1.108 0.00 -0.005 0.00 0.001 0.00
1 0.002 2.63 2.645 1.24 1.164 0.15
Computations 2 0.040 0.02 0.001 1.13 0.000 -0.98
3 0.001 1.09 0.055 2.41 0.018 1.39
4 0.009 2.68 0.000 1.02 0.000 0.56
Figure 6-13: Third order Fourier analysis of the efforts and moments in the horizontal plane,
v’=0.15

0.2
0.0
0 1 2 3
-0.2
Amplitude X'
-0.4
-0.6 Experiments HSVA

Computations
-0.8
-1.0
-1.2
Harmonic order
3.0
2.5
2.0
Experiments HSVA
Computations
Amplitude Y'
1.5
1.0
0.5
0.0
0 1 2 3
-0.5
Harmonic order
1.4
1.2
1.0
Experiments HSVA
0.8
Computations
Amplitude N'
0.6
0.4
0.2
0.0
0 1 2 3
-0.2
Harmonic order
Figure 6-14: Comparison of efforts and moments amplitudes in the horizontal plane from
computations with experimental values
These results show firstly that effort amplitudes and phases of the harmonic 0 or 1 are well
predicted by computations.

The pictures below present iso-velocity contours that show vortex generation along the hull
during the unsteady computations.
T0 T0+1/4T
T0+1/2T T0+3/4T
Figure 6-15: Iso-velocity contour at four instants (T0, T0+1/4T, T0+1/2T, T0+3/4T)
We must point out that these comparisons are difficult because of uncertainties due to
experiments. An analysis of the different sources of uncertainties in manoeuvrability
experiments has been studied by Yoshimura (Yoshimura, 2002).
Efforts signals are now compared to experimental results. The pictures below show for two
computations cases with different mesh densities and time steps a comparison between
computational results and experiments.
mesh 0.5M, tau 0.01s

2
Exp. HSVA
0
Y'
-1
-2
0 2 4 6 8 10
T
Figure 6-16: Time signal of lateral effort Y’


Exp. HSVA
1
N' 0.5
-0.5
-1
0 2 4 6 8 10 12
T
Figure 6-17: Time signal of moment about z axis N’
Observed differences are to be confronted with dispersion of the experimental results. The
pictures above presents the brut signal from HSVA in spite on the fact that they seem to be
already treated with a low pass filter. A fine identification of the amplitudes and phases for
high order harmonics is therefore quite difficult.
6.2.8.5. Application to Hamburg Test Case for yaw PMM simulation

The yaw forced motion is performed with an imposed velocity always tangent to the trajectory.
That condition is equal to a null lateral velocity in the ship’s reference frame.
Figure 6-18: Trajectory of the ship for the yaw PMM test
Experiments conducted at HSVA for the European Project VIRTUE (Vogt, 2007) and
simulations have been conducted with these parameters: motion period 32 seconds, ship
L
velocity 1.886 m.s and amplitude for the non-dimensional tactical diameter r '  t    t 
-1
R
equal to 0.4.
The tables below report the Fourier analysis from time efforts and moment in the horizontal
plane for computations and experiments.

X' Y' N'

0 -1.979 0.00 -0.013 0.00 -1.029 0.00
1 0.000 0.00 0.539 0.12 -0.017 -0.01
Experiments
2 -0.016 -0.03 0.010 0.00 -0.002 0.02
3 0.001 0.00 0.002 -0.01 0.002 0.00
0 -2.139 0.00 -0.009 0.00 -0.877 0.00
1 0.002 -0.24 0.648 0.27 0.338 0.20
Computations
2 0.731 0.16 0.006 -0.14 0.001 0.10
3 0.002 -0.12 0.215 0.14 0.080 -0.08
Table 6-4: Third order Fourier analysis of efforts and moments in the horizontal plane, r '  0.2
X' Y' N'

0 -2.132 0.00 -0.015 0.00 -0.978 0.00
1 0.007 0.00 1.068 0.24 -0.010 -0.01
Experiments
2 -0.068 -0.07 -0.015 0.01 0.013 0.07
3 -0.006 0.00 -0.030 -0.03 0.011 -0.01
0 -2.352 0.00 -0.003 0.00 -0.864 0.00
1 0.006 -0.22 1.177 0.19 0.107 0.21
Computations
2 0.282 0.11 0.000 -0.05 0.004 0.18
3 0.002 -0.10 0.349 0.10 0.170 -0.31
Table 6-5: Third order Fourier analysis of efforts and moments in the horizontal plane, r '  0.4
The graphs below present results for r '  0.4 in order to compare numerical results to
experiments more easily. Amplitudes are given in function of the harmonic number.
0.5
0.0
0 1 2 3
-0.5
Amplitude X'
-1.0
Experiments HSVA
Computations
-1.5
-2.0
-2.5
Harmonic order
1.4
1.2
1.0
Experiments HSVA
0.8
Computations
Amplitude Y'
0.6
0.4
0.2
0.0
0 1 2 3
-0.2
Harmonic order

0.4
0.2
0.0
0 1 2 3
-0.2
Amplitude N'
-0.4
Experiments HSVA
-0.6 Computations
-0.8
-1.0
-1.2
Harmonic order
Figure 6-19 : Amplitudes in function of harmonic order, numerical results compared to

experiments, Efforts and moments in the horizontal plane, r '  0.4
These results show quite good prediction for the amplitudes of efforts and moments in the two
conditions. Furthermore we obtain some differences on phases quite small in absolute values.

6.3. Application Cases: Direct numerical manoeuvring simulation

Solution of nonlinear Navier-Stokes equations including viscous and turbulence terms are
now commonly used in ship resistance predictions by almost all research institutes and few
industrial parties (Chapter 4 - Application 1: Numerical Towing tank). More recently, their field
of application as been extended to hull-form optimisation (Tahara et al., 2000, Jacquin et al.,
2004, Campana et al., 2004) and to simulations of unsteady ship manoeuvring (Di Mascio et
al., 2004, Wilson et al., 2002, Jacquin et al., 2006). Computations of steady and unsteady
ship manoeuvring using forced motion as PMM - Planar Motion Mechanism (Cura-Hochbaum,
2006) represent a first approach for unsteady simulations before performing self-propelled
fully free model computations.
Thus, the application field of these numerical tools is incredibly wide, especially for unsteady
simulations where the model can sail in all directions of a manoeuvring ship. The interest of
such fully unsteady numerical simulations for manoeuvring-performances evaluation is to take
into account the viscous flow around the hull or appendages, the wake in the propeller plane
and complex nonlinear effects such as hull / propeller / rudder interactions.
This application case presents next steps in the simulation of an unsteady ship manoeuvring,
considering computations of a 6DOF free model in turning motion on calm water and in
regular waves. A hull form optimisation process will be used to improve the ship manoeuvring
performances during unsteady turning circle.
6.3.1. Case 3-d: Unsteady manoeuvring ship on calm water

The goal is to simulate a fully free self-powered ship with moving appendages within a
RANSE solver.
6.3.1.1. Appendages simulation

The final aim of the simulation of appendages in manoeuvrability is the simulation of moving
appendages with some command laws in order to compute IMO tests as turning circle, zig-
zag or pull-out.
As explained previously, the complete simulation has to be validated step by step with more
and more complex levels. Three levels of different difficulties can be emphasized. The rudder
will be firstly simulated with a constant force obtained from experiments or calculated with
empirical laws, functions of the surface and the angle of the appendages. Then we will use
meshes with meshed rudder but with non deforming mesh and fixed angles. Finally
computations will be conducted with deforming meshes and command laws. Two methods for
mesh deformation using either spring method or interpolation of existing meshes will be
compared.
6.3.1.2. Propeller simulation

Using Navier-Stokes computations to predict the real manoeuvring ship behaviour is fully
achieved only if the propeller is taken into account in the simulations. We have to consider
here the interactions of the propeller with the hull and appendages. The hull’s drag usually
increases because of the pressure field modification and also appendages drag since they
are located in the propeller flow.
Several approaches can then be considered. The first one consists in directly modelling the
propeller in the computation with its own mesh. The propeller rotation is taken into account

with the rotation of the mesh in a cylinder. These computations are one of the aims of the
numerical naval hydrodynamics but needs a huge mesh density and so very long CPU time.
The second approach, which has been developed in the present work, consists in simulating
the propeller effect on the flow with external forces in mean momentum equations. This
solution called actuator disc method allows getting, in a way more or less complex, suction
and wake effects induced by the propeller and with a reduced CPU time. That solution was
developed with RANSE by Jacquin (Jacquin et al., 2007).
The method used for the development of the simulation of the propeller is quite similar to the
one used for appendages. The first step is to simulate the thrust of the propeller using a
constant force equal to the ship resistance at the speed we want to simulate. Then, a more
complex model using actuator disk will be used before using finally rotating meshed propeller.
6.3.1.3. Global configuration

The table below sums up the different steps in the simulation of ship manoeuvring considering
or not interactions of the propeller and of the appendages.
Propulsion Appendages Interactions

Turning
Constant Constant Meshed
Actuator disk meshed
effort effort appendages
rudder
X X No
X X Hull/Propeller
X X Hull/Propeller/appendage
X X Hull/Propeller/appendage
Table 6-6: Step by step increasing complexity for the manoeuvring simulation
The pictures below illustrate the previous table giving some example of configurations.
Figure 6-20 : from top to bottom, constant effort for propulsion and constant effort for rudder
lift, actuator disk and constant effort for rudder lift and actuator disk and meshed rudder

6.3.1.4. Application to Hamburg test case

Computations will be compared to experiments performed at ECN for the European Project
VIRTUE. These experiments consisted in IMO turning circle manoeuvre (as described in IMO
MSC/Circ. 1053) on Hamburg Test Case for different rudder angles.
Computations have been performed with constant efforts simulating rudder lift and propeller
thrust from experimental values of thrust and rudder lift. A light mesh of 50.000 cells was used
in order to reduce drastically computational time. Nevertheless we will study the impact of the
mesh density to validate our results. Since a constant force was used to simulate the rudder
lift we will only consider the turning circle and not the transition between straight road and
permanent gyration. First computations were conducted only with 3 degrees of freedom
considering the horizontal plane. Special difficulties especially instabilities due to pitch
stiffness were observed during 6 degrees of freedom computations.
The results presented below used the following parameters: rudder angle ± 25°, velocity 0.5
-1
m.s .
The computation is performed with different steps. In the first one the ship is accelerated fixed
to her velocity. Then the surge, pitch and heave are free and the ship is propelled with a
constant force. When the equilibrium is obtained all degrees of freedom are free and the ship
turns under the effect of the constant force simulating rudder lift.
The next table and picture present the steady turning diameter compared to experiments.
Steady turning diameter

Rudder angle Experiments ICARE
3 DOF Error 6 DOF Error
+25° 2.52·Lpp 2.51·Lpp 0.42% 2.66·Lpp 5.73%
-25° 2.78·Lpp 2.79·Lpp 0.19% 2.90·Lpp 4.31%
Table 6-7: Steady Turning diameter obtained by computations compared to experiments,

rudder angle ± 25°, 3 and 6 DOF

Experiments
Rudder Angle : - 25° ICARE 3DOF
6 ICARE 6DOF
-2
Y
-4
-6
-8
-10
-12
-5 0 5 10
X
Figure 6-21: Tactical Diameter obtained by computations compared to experiments, rudder

angle ± 25°, 3 and 6 DOF
Results are quite good compared to experiment in the permanent gyration period but we
obtained different transitions for 3 degrees of freedom and 6 degrees of freedom that could be
explained by hull effects due to heel.
A view of the free surface elevation during the turning circle manoeuvre is presented below.

Figure 6-22: Free surface elevation and iso-velocity during turning circle manoeuvre

6.3.2. Case 3-e: Unsteady manoeuvring ship in Waves
6.3.2.1. Preliminary results

The case of a DTMB Model 5415 with forward speed in head waves is presented here. This
model has been chosen as a test case for the CFD workshop in Gothenburg 2000 (steady
case) and Tokyo, 2005 (steady and unsteady case). Here added resistance in waves is
compared with experimental data (Gui et al, 2001).
Computations have been done on a half domain discretised by a O-O grid of 212784 nodes.
A grid convergence study has been undertaken to verify that this grid was enough refined.
Parameters of the computations are the Froude Number (Fn), the wave steepness (Ak) and
the nondimensional wavelength (/L) with:
U 2 A
Fn  , Ak  (121)
gL 
(U is the ship forward speed, L length between perpendiculars, A the wave amplitude and 
dimensional wavelength). In the following CT ,0  is the steady component of the total resistance
FX t 
coefficient CT t   where Fx(t) is the total drag resistance (S wetted surface of the
0.5 U 2 S
ship).
Added resistance due to waves refers to ocean waves and is not to be confused with wave
making resistance. Ocean waves cause the ship to expend energy by various effects: for
example the wetted surface area of the hull increases so causes added viscous resistance.
This component of resistance can be very significant in high sea states and has a very
nonlinear behaviour.
The added resistance in head waves is usually obtained by subtracting the still water
resistance from the measured mean total resistance at the studied speed. That implies that
two computations have been done with the same conditions for trim and heave at the same
Froude number for each steepness at three wavelengths (one in still water and a second one
with waves). Table 16 and figure 42 resume the results obtained for different wavelengths and
steepnesses at Fn=0.28. The error is computed as ((num - exp )/exp )100 and the last column
represents the amplitude of the added resistance comparing to the total mean resistance. As
expected it appears that the added resistance in head waves is positive and increases
following the waves amplitude from 0.2 % to 31 % of the mean total resistance. It varies with
the squared wave amplitude for all of the cases presented in this table (if the numerical
accuracy is sufficient).
CT ,add
Fn Ak  Lpp Num. Exp. Err.(%) CT ,add CT ,0 (%)
0.28 0.025 0.5 0.00002 0.000013 35 0.2
1.0 0.00022 0.00015 46 2.6
1.5 0.00042 0.00040 5 4.5
0.05 0.5 0.00022 0.00017 29 2.5
1.0 0.00085 0.0008 6 8.8
1.5 0.00179 0.00178 1 17
0.075 0.5 0.00043 0.00041 5 4.5
1.0 0.00182 0.00192 5 17
1.5 0.00392 0.00401 2 31
Table 6-8: Added resistance compared to experiments at Fn=0.28

For a steepness Ak  0.025 agreement between computations and experiments is quite poor
but in this case the amplitude of the added resistance is very small compared to the total
mean resistance (about 2-5 % of CT ,0 ). Computations at higher steepnesses (0.05 and 0.075)
show accurate predictions of the added resistance (error below 5 %) as the ratio CT ,add CT ,0 is
up to 10 %.
Pictures below represent an example of comparison of computation with experiments.
Figure 6-23: Comparison between computation (up) and experiments (down) for total wave

pattern at two time steps,  1.5 ,
L

6.3.2.2. Application to ship manoeuvrability of DTMB 5415 in waves

The computation of a DTMB 5415 in turning circle manoeuvre in waves presented below is a
demonstrative case. The simulation has been performed considering a velocity of 18 knots
using a constant force as a propeller and a constant force simulating rudder lift. Those
parameters are the same for all different cases. The changing parameters concern waves
parameters. A previous computation without waves was performed in order to compare the
influence of the sea state on the ship trajectory. Then three amplitudes and three wave
periods have been computed to evaluate the impact of these parameters on the ship’s
trajectory.
The charts below present velocity, roll and pitch in the ship’s reference frame as an example
of the turning circle manoeuvre in waves.
These first computations show a similar behaviour with manoeuvring ship, with the prediction
of snap-roll phase when rudder is turned, and speed decrease when ship is turning.
6
Velocity (m/s)
0
100 200 300
Time (s)
Figure 6-24: Velocity during turning circle manoeuvre in waves’ simulation.
-2
Roll (°)
-4
-6
100 200 300

Time (s)
Figure 6-25: Roll during turning circle manoeuvre in waves’ simulation.

0.5
Pitch (°)
-0.5
-1
-1.5
100 200 300
Time (s)
Figure 6-26: Pitch during turning circle manoeuvre in waves’ simulation.
Figure 6-27: DTMB 5415 in turning circle manoeuvre without wave

Figure 6-28: DTMB 5415 in turning circle manoeuvre in waves during the snap-roll phase.
Figure 6-29: DTMB 5415 in turning circle manoeuvre in waves.

6.4. Guidelines for numerical manoeuvring applications

The following sections review the numerical aspects of accurate manoeuvring simulation.
These guidelines serve as guidelines specific to manoeuvring applications, next to the
general guidelines presented earlier in this document.
6.4.1. Definition of target variables

The target variables required for the evaluation of the ship manoeuvring performance are in
general the longitudinal force, transverse force and yaw moment. For ships manoeuvring at
high speeds or ships with low metacentric height, the roll moment should also be included.
For semi-displacement or planning ships the heave force and trim moment will also be of
interest.
Depending on the application, the wave elevations around the ship, the hull pressure
distribution and the wake in front of the propeller(s) and rudder(s) can be regarded as target
variables.
When using the derivative approach as described in section 6.1.3 and 6.2, an adequate set of
calculations should be defined. These calculations should cover the range of anticipated
conditions that the ship will meet during the manoeuvres that are to be assessed. Information
about the range can be obtained from previous experience, manoeuvring information of
similar ships or by using (reliable) fast-time simulation programs based on empiric
manoeuvring coefficients.
6.4.2. Selection of adequate mathematical model for the flow physics

To estimate the forces and moments acting on a manoeuvring ship an adequate turbulence
model is required to close the RANS equations. Turbulence models generally adopted are
solver or user dependent and generally range between one equation models (Spalart Almaras
or Menter) or two equation models (k- or k-). In some cases considered within VIRTUE,
more advanced methods were used, but their superiority could not be clearly demonstrated.
Depending on the ship, the influence of the turbulence model on the forces and moments can
be large, in particular for cases with ill-defined flow separation lines. In those cases, k-
models appear to perform well. In case of doubt, it is advised to re-do the calculation with
another well-established turbulence model to ensure the reliability of the results.
The differences between calculation results with and without wall functions were small for the
cases considered within VIRTUE. However, in cases where large flow separation takes place,
using wall functions may lead to erroneous results. This can be the case for extreme
manoeuvres or full-block ships.
In some cases, especially for semi-displacement or planing ships, the forces on the ship can
be speed dependent. This is mainly caused by free-surface deformation. To simulate these
effects, the deformation of the free-surface should be modelled in the calculations. To obtain
sufficient accuracy of the computational results, this will lead to a considerable increase in the
number of grid nodes or cells, compared to double-body calculations.
During the VIRTUE project, it was found to be preferable to conduct a double-body calculation
with good accuracy than to conduct a calculation including free-surface modelling with low
accuracy.

6.4.3. Selection of computational domain

Compared to resistance calculations, a larger domain is required for manoeuvring studies:
due to the generation of lift on the hull, the pressure field around the ship and the generated
wake aft of the ship will be more pronounced for a manoeuvring ship. For increasing drift
angles and yaw rates, the domain size should increase as well.
Several options exist to choose the domain layout:
 Rectangular (brick-like) grid with adaptation of the boundary conditions to facilitate
the drift angle and/or the yaw rate. In this case, the undisturbed flow will not be
aligned with the outer boundaries of the domain. One grid can be used for all
calculations.
 Aligned grid, in which the ship geometry is rotated to facilitate the drift angle and the
outer boundary is curved to facilitate the curvature of the flow due to the yaw rate. In
this case, the undisturbed flow is aligned with the outer boundaries of the domain. For
each new drift angle or yaw rate, a new grid needs to be generated. To reduce the
time spent on grid generation, a (rotated) ship geometry aligned inner-grid in
combination with a deforming outer block can be used (see the application example
for the Derivative Approach).
 Large-size spherical shaped domain with application of undisturbed flow at the outer
boundary. Due to the large size of the domain, undisturbed flow can be assumed at
the outer boundary. The cells at the outer boundary are large and introduce numerical
diffusion such that the wake of the ship will disappear at the outer boundary. In this
case, the undisturbed flow will not be aligned with the outer boundaries of the
domain. One grid can be used for all calculations.
6.4.4. Selection of boundary conditions and initial conditions

The boundary conditions differ from motion to motion. In general, no special measures are
required for manoeuvring application, unless unsteady motions are simulated. For unsteady
motions, the boundary conditions may need to be adjusted for every time step.
Calculating a ship in steady oblique inflow, does not impose special treatment of the inflow,
initial and boundary conditions. Steady rotational motion, oscillatory motion or direct
numerical calculations do: here either the grid containing the hull form should move in space
corresponding to the motion to be simulated (moving grid approach) or a non-inertial
reference system (body force approach) should be adopted. Within VIRTUE, it has been
demonstrated that both approaches can result in practically identical results, without
preference to one or the other.
6.4.5. Selection of adequate grid topology/spacing and time step

In general, the use of structured meshes is preferred when this can facilitate the use of
higher-order discretisation schemes. However, unstructured meshes with (adaptive) local grid
refinement can also be used to obtain good results. Using tetrahedral cells near the wall is not
recommended.
The most common block-structured grid topologies are O-O and H-O type meshes. Both
mesh topologies can be used to obtain good results. In O-O grids, the grid lines follow the
contour of the geometry, but this can lead to largely distorted cells, e.g. in the area of bulbous
bows. For H-O grids the topology is violated at the bow and stern contours. However, the
quality of the cells can be better compared to O-O type grids and clustering of grid lines near
the bow and stern may reduce the error due to the distortion of the geometry.

When calculating moderate manoeuvring motions (e.g. || < 20°, || < 0.5), the influence of
instationary flow on the integral or local values will be small. In these cases, steady solution
processes can be used to calculate the flow. For some solvers or cases, using time stepping
may lead to the (same) steady solution, but with improved robustness of the iterative
procedure. In these cases, the time stepping is used as under-relaxation. Although the effect
of the time step size is solver dependent, a good first estimate of the time step is t=tref/20,
with tref=L/V and L an appropriate length in the flow direction (e.g. L pp).
In cases where more accuracy in time is required, for example cases with unsteady flow
separation, vortex shedding or oscillatory motion, considerable smaller time steps and higher-
order accuracy are needed. In these cases, the time step should be smaller than roughly
t=tref/100.
6.4.6. Selection of discretisation schemes

For manoeuvring applications, the discretisation scheme should be selected using the general
guidelines.
6.4.7. Selection of iterative convergence criteria

For manoeuvring applications, the iterative convergence criteria should be selected using the
general guidelines. Furthermore, the user should verify the convergence of the integral values
by examining the evolution of each value in time or during the iterations. For oscillatory
motions, sufficient oscillations (at least two full periods) should be simulated to ensure
appropriate convergence of the harmonic components.
For manoeuvring applications, the post processing should follow the general guidelines.
Especially, consider whether the interpretation of the results and any conclusions drawn
based on the interpretation are within the accuracy of your computation. E.g. if your
uncertainty of your solution is 10% and you want to investigate the difference between two
hull forms, you can only conclude whether one is better than the other when the difference is
larger than the uncertainty.
Furthermore, make sure that the coordinate system and symbol definitions comply with the
specifications applicable for the reporting of the results. For example, the coordinate systems
usually adopted by CFD specialists will be different from the coordinate systems used by
manoeuvring experts.
6.4.9. Estimation of numerical accuracy of the solution

For manoeuvring applications, the estimation of the numerical accuracy should follow the
general guidelines. Since the use of CFD for manoeuvring applications is a relatively new field
of expertise, conducting verification of the calculations is highly recommended. Fortunately,
empiric estimations of the manoeuvring forces can be found in literature, such as those
published by Inoue or Clarke. It is advised to verify the magnitude of the results of the
calculations with these empiric estimations.

6.4.10. How to introduce CFD derivatives in manoeuvring simulator

Same principles and methodologies could be applied to introduce CFD derivatives in
manoeuvring simulator as physical model tests. Reference is made to the ITTC guidelines for
further reading.

7. Application 4: Numerical Cavitation tank
7.1. Special issues relating to propeller flow modelling

The viscous flow analysis by RANS or LES of a propeller is quite distinct from a similar
analysis of a ship’s hull for several reasons:
 Force computation
The drag of a ship is usually dominated by the frictional drag and a viscous flow
analysis is mandatory for reliable resistance predictions. Potential flow calculations,
although useful for wave resistance and for general qualification (comparing two ship
hulls and deciding which one is the best of the two), do not provide an alternative.
Thrust and torque of a propeller, on the other hand, can quite reasonably and at a
fraction of the costs, be computed with an inviscid flow code, because frictional forces
are small relative to pressure forces. This means that the extra computational effort of
a viscous flow analysis for propellers is useful and justified only when the quality of
the results is substantially better or when flow details like tip or blade-root vortex
formation can be adequately modelled. Obviously that better quality comes with high
demands on grid quality and resolution.
 Complex geometry and gridding
The geometry of a propeller must be considered as more complex than the geometry
of a ship’s hull. Propeller blades show locally very high curvatures (at the leading and
trailing edges and at the tip and the root)) and are nearly flat in other areas (near
midchord). Because a propeller is a rotating device, the blade is torsioned and high
skew is rather the rule than the exception. The generation of a good quality grid is
therefore a major task, and deviations from orthogonality tend to be higher than in
grids around ships, particularly for blade-periodic structured grids.
 Cavitation
It is sometimes argued that a non-cavitating propeller cannot be the optimum
propeller. True or not, the flow around a propeller is often a cavitating flow. Since
cavitation is important for the hydrodynamic efficiency of the propeller, for propeller-
induced pressure fluctuations (being the most important source of ship vibrations),
and for other nuisances like erosion and noise, there is often a practical demand that
the viscous flow analysis should include cavitation. It is true that two-phase flow
modelling is sometimes applied to the flow around ships also, viz. when the free
surface is included, but the interface between the phases is then smoother. Indeed,
for ships single-phase flow computations with free-surface fitting offer an excellent
alternative, while for propellers such an approach would have serious shortcomings.
 Rotating object
The propeller is a rotating object. It implies that the viscous flow must either be
computed in an inertial reference frame but with solid grid motion, or must be
computed in a blade-fixed rotating (non-inertial) reference frame requiring Coriolis
and centrifugal forces to be added to the mathematical model. If in the analysis the
propeller is combined with non-rotating bodies like for example the hull or a rudder,
rotating and inertial frames have to be combined, requiring sliding interfaces with or
without overlap of grids.
 Unsteady flow
With the exception of the non-cavitating flow around a propeller operating in open
water (uniform inflow along the direction of the propeller axis), the flow around a
propeller is an unsteady flow. Blade-periodicity of the flow is lost, so next to additional

efforts for time-accurate solutions, a substantial increase of the grid size is involved.
The flow around a ship, unless operating in incoming waves, can usually be treated
as a steady flow.
 Diversity of space and time scales
Ultimately, the computation of the flow around a propeller should reproduce the
essential phenomena related to propeller cavitation, i.e. propeller induced pressure
fluctuations and erosion due to cavitation. In other words, for propellers one is to
solve a two-phase flow problem aiming at capturing both the large scale influence of
cavitation (hull pressure fluctuations) and the small scale effects related to cavitation
(erosion assessment).
All the items listed above, make the computations for propeller more demanding than for
ships.
7.2. Guidelines for predicting cavitating flows around 2D and 3D

stationary foils
7.2.1. Type of simulation
Before taking up a new numerical flow simulation, it is good practice to carefully consider
what kind of flow is to be simulated, what physical phenomena will dominate, and what kind of
results the computations are expected to produce. This will to a large extent determine the
set-up of the entire calculation process, including pre and post processing.
7.2.1.1. 2D or 3D
For 3D foils it is of course evident that also the flow simulation must be done in 3D, but for 2D
foils there is still a choice. Reasons to favour the more demanding 3D flow simulation might
be:
 the results of the computation are to be compared with experimental data, and the
flow in an experiment on a nominally 2D foil can never be fully 2D, because 3D
effects inevitably occur at the junction with the tunnel walls;
 if LES is applied it seems unnatural to limit the eddy structures to 2D;
 even if the wetted flow is 2D, cavitation may still have a 3D appearance and cause
the flow to become 3D.
7.2.1.2. Inviscid (BEM or Euler) or viscous (RANS or LES)

If the boundary layer on the foil is thin and flow separation does not occur, it is well-known
that inviscid flow simulations do a very good job in calculating the lift and drag characteristics
of the foil at a fraction of the costs of a viscous flow simulation. On the other hand inviscid
flow simulations give no information about the boundary layer behaviour, which may actually
be what you want to know. Also, cavitation and boundary layer behaviour are strongly related.
It is therefore relevant to make a choice based on the purpose of your simulations.

7.2.1.3. Stationary or transient

The flow around stationary foils at small or moderate angle-of-attack in a steady onset flow
can be expected to be steady, except for turbulence. However, at high angle-of-attack a
steady flow solution may not be realistic. Moreover, with cavitation on the foil, the flow is
seldom steady; only for small cavities, occurring at cavitation numbers close to inception,
machine-accurate steady state solutions may be obtained.
7.2.1.4. Wetted or cavitating flow

Cavitating flows are prone to unsteadiness and have additional requirements on the gridding
(see below) compared to wetted flows for the same foil. If wetted flow and cavitating flow are
to be computed on the same grid, the latter determines the grid features.
7.2.1.5. Compressible or incompressible

The Mach numbers associated with the flow past hydrofoils are so low that compressibility
effects may safely be ignored. But for cavitating flows there are evidently compressibility
effects, notably in the collapse phase. If these are considered relevant (e.g. for erosion) you
may decide on a simulation with a compressible fluid. On the other hand, many aspects of
cavitating flows can be realistically computed with the assumption that the liquid and the
vapour phase are incompressible.
7.2.2. Domain size and meshing
7.2.2.1. Domain choice

For external flows the choice of the size of the computation domain is quite important
because exact boundary conditions are typically not available on the domain border. For the
simulation of the flow around the foil in a supposedly unbounded domain the radius of the
computation domain may need to be 30 to 50 chords to avoid effects on lift. If on the other
hand an experiment is simulated numerically, the dimensions of the tunnel test section
determine the lateral boundaries of the computation domain and only inlet and outlet planes
have to be chosen.
7.2.2.2. Grid
The grid around the foil must be surface-fitted and quadrilateral cells (hexahedra or prisms in
3D) are recommended near the foil. The use of triangular cells (tetrahedral in 3D) near a no-
slip surface degrades the accuracy of the results and/or leads to an excessive number of grid
cells.
If a structured grid is applied, a C-type grid is preferable for foils with a sharp-cornered tail,
while an O-type grid is more suited for foils with a tail of finite thickness.
Deviations from orthogonality of the grid should be minimised. The cell size should change
gradually, the cell size ratio of neighbouring cells staying preferably below 1.25.

7.2.2.3. Special requirements for cavitating flows

While the grid requirements for wetted flows are primarily driven by the resolution of the
boundary layer on the foil, for cavitating flows extra requirements come in. All methods
operating with a single mixture fluid of variable density are concerned with the sharp density
gradients at the liquid-vapour interface. Since the cavity is usually much thicker than the
boundary layer, high grid resolution is needed where it is not available in grids for wetted flow.
Solution-adaptive grids seem to hold the best prospects.
7.2.3.1. Foil surface

For inviscid flow simulations the impermeability condition (zero normal velocity) must be
applied on the foil surface. In viscous flows the no-slip condition (zero tangential velocity)
must be imposed in addition. The application of the wall function approach to set boundary
conditions on the foil surface is dissuaded, in particular for cavitating flows.
7.2.3.2. Inlet
On a domain boundary denoted as inlet boundary the velocity vector is to be specified.
7.2.3.3. Outlet
Conditions at the outlet boundary are harder to specify because the velocity field there is
disturbed by the foil and the pressure may not have recovered an undisturbed level.
Moreover, in time-dependent flow simulations the flow at the outlet is likely to be time-
dependent as well. Weak boundary conditions in terms of the normal gradients of the
variables are usually chosen.
7.2.3.4. Lateral boundaries

On lateral boundaries undisturbed flow conditions or free-slip boundary conditions are
typically applied. Attempts to include tunnel wall boundary layers in the simulations are often
omitted.
7.2.3.5. Initial conditions

For transient flow computations the initial solution must be specified for all dependent
variables. It is advised to choose the initial velocity field so that it satisfies the continuity
equation. While the average pressure level is immaterial for a wetted flow simulation, because
only pressure gradients appear in the flow-governing equations, in a cavitating flow it is
important in setting the cavitation number.

7.2.4. Turbulence modelling
7.2.4.1. Model choice

For inviscid flow simulations there is of course no need to select a turbulence model, but in
RANS-based computations it is a crucial step and can have substantial effects on the results,
while for LES a sub-grid turbulence model is needed, although sometimes omitted (so-called
implicit LES).
These days, the two-equation models are most often chosen, although the one-equation
models are applied as well. The k-ε model (in several variants) is not recommended because
of its poor performance in predicting boundary layers in adverse pressure gradient. A model
from the k-ω family is therefore a practically acceptable option, with an EASM model, being
slightly more advanced but more costly and more sensitive to numerical problems, as an
alternative.
7.2.4.2. Near-wall treatment

If the grid has sufficient resolution to allow the flow to be computed down to the wall, you must
make sure that your turbulence model is suited for low-Reynolds flows (it may require extra
damping functions). If wall functions are applied it is assumed that the flow is fully turbulent
and that the velocity profile has a part where the law-of-the-wall applies. This may not be true
everywhere and then the settings for the dependent variables appearing in the turbulence
model may be inappropriate.
7.2.4.3. Transition prediction

Transition, i.e. the change of the flow in a shear layer from a laminar to a turbulent state, is a
complex phenomenon. It can be due to Tollmien-Schlichting type instabilities (natural
transition), invoked by a trip wire or local roughness (forced transition) or caused by high
turbulence levels in the free stream (by-pass transition). In numerical simulations it is
sometimes ignored (if wall functions are applied, which assumes that the boundary layer is
turbulent everywhere), sometimes occurring as a result of the turbulence model (which has
not been tuned for that purpose) and sometimes by solving extra equations (e.g. for an
intermittency function or a momentum-thickness Reynolds number) in combination with
empirical data.
7.2.5. Numerical issues
7.2.5.1. Spatial and temporal accuracy

The spatial accuracy of a computation is governed by the grid density, the type of
discretisation schemes applied to the spatial derivatives and the level of convergence of the
iterative solution. The temporal accuracy is governed by the time step, the discretisation
scheme for the time derivatives and the level of convergence of the solution in each time step.
7.2.5.2. Convergence criteria

A good way to judge the convergence of the iterative solution is by monitoring the residuals of
the equations being solved. Typically the residual are considered in some norm, e.g. the L1,

L2 or L∞ norm, the latter being the most stringent. Other options are to monitor the changes in
the dependent variables between two successive iterations (make sure that these exclude
any underrelaxation) or the global forces on the foil.
7.3. Guidelines for predicting propeller flows including cavitation
7.3.1. Definition and planning phase
If we look at the situation in a very simple way, the numerical cavitation tunnel consists of
computer hardware, two phase flow-modelling software and the user(s) who is(are) controlling
hardware and software.
7.3.1.1. Tutorials
In order to guide the user through complex applications using advanced features of their
code, the suppliers of RANS codes provide tutorials. Tutorials are a very efficient way to get
into touch with new applications and to avoid errors and misinterpretations of the
documentation. For the propeller case one may find suitable tutorials covering rotating frames
of reference, moving meshes and multi phase flow in general. Additional work related to
numerical propeller analysis, which is clearly not covered by a tutorial, concerns strategy and
grid generation work.
7.3.1.2. Planning and sharing work

It is not unlikely that several specialists share the work on a RANS propeller project. In this
respect it is reasonable to reserve effort and time for a specification and planning phase.
Especially everybody has to be aware of the tools at hand. To avoid doing work in vain it is
proposed to approach according to the following sequence:
7.3.1.3. Characteristics of actual propeller flow problem

 Specify aim and focus of actual analysis (flow phenomenon of interest and
geometrical area concerned)
 Judge if cavitation modelling is to be invoked or fully wetted analysis is sufficient
 Judge if a steady analysis is suitable or if an unsteady computation is required.
 Judge if a single domain grid (for steady state analysis and some unsteady
configurations) or a multi domain grid (to account for a mesh motion via sliding
interfaces or overlapping grids) is required
 Specify results to be extracted in the post-processing
7.3.1.4. Grid generation items

 Specify a checklist for the quality of the blade surface description
 Specify if the grid is supposed to be structured or unstructured

 Prescribe y+ and deduce the height of the cell layer at the blade surface
 Specify resolution on blade surface and inside the flow
 Specify grid topology and dimensions
 In order to reduce the total number of cells consider solution-dependent local
refinements (if the solver allows it)
 Generate the grid
7.3.1.5. Pre-processing and settings for the solver control

 Specify the boundary conditions
 Specify the solver settings (equations to be solved, related schemes etc.)
 Specify if additional coding is to be linked to the standard code to support the analysis
(boundary conditions, grid motion, additional output of forces etc.)
 Decide about the post-processing tool and check output formats of the solver
 Decide about normalisation of quantities
 To capture flow details via the velocity field consider extracting the induced velocities
from the total velocity field.
7.3.1.7. Solver at hand

 Generally multiple frames of reference and moving grids are features that require
’acceptance’ by the solver. If the solver is capable of handling overlapping grids a lot
of work in the grid generation process may be saved. A near grid for the propeller
blade is then to be combined with a simple grid that just connects the outer
boundaries and ignores the presence of the blade.
7.3.2. Type of simulation
7.3.2.1. Aim of analysis

The reason for a numerical analysis may be that an existing propeller is to be examined and
sources for unsatisfactory behaviour are to be identified. A balance between effort and
required quality of the results should be considered and possible simplifications should be
kept in mind. Direct cavitation modelling usually needs additional effort in grid generation,
requires much more fluid cells and may also request additional tuning for the solution
process. An indirect assessment of cavitation can be set up via the fully wetted pressure,
which may be sufficient to specify cavitation inception.

7.3.2.2. Algorithms for cavitation

The solver at hand may not offer too many alternatives for direct cavitation modelling. Usually
the existence of a second phase will be accounted for by a transport equation for the vapour
volume fraction. Differences between cavitation models of that kind can be found in the
source term that defines the rate of vapour production or destruction. Sauer (2000) proposed
a source term based on bubble dynamics. Normally weak destruction or production terms
should increase the robustness of the numerical solution algorithm. Usually a cavitation model
provides parameters to control destruction and production rates. It is recommended to invoke
the cavitation analysis after a converged fully wetted solution is obtained.
7.3.2.3. Accounting for symmetries

Setting cyclic boundary conditions, if the solver provides such an option, helps to reduce the
size of the problem considerably. They could be applied for the open water propeller test case
treated in the 1st VIRTUE WP4 workshop (Streckwall and Salvatore, 2008). Generally a
violation of the cyclic nature of the flow around the blades (viewed from a rotating reference
frame) requests an unsteady analysis. Moreover cyclic boundaries cannot be used in this
case and the grid has to cover every propeller blade. The work on the grid and the required
computation time will be much larger than for a steady analysis. So first one should consider
whether the flow problem of interest could be covered with steady or approximately steady
flow around the blades.
7.3.2.4. Propeller and ship: simplified approaches

If the propeller is interacting with a stationary object like a ship hull an unsteady analysis is
needed. A rigorous approach would require a moving mesh part that includes the propeller
and a fixed grid part around the stationary object (like a ship hull). If there is no time or
funding for such a complex computational setup, the environment of the propeller may be
simplified. Modelling just the ’influence’ of objects in the vicinity, instead of including the
objects in the grid will save a lot of effort. In this manner a hull or any other wake generator
may be replaced by a prescribed flow at the inlet plane. Working out the details for this
approach one is left with two alternatives:
 Keep the grid fixed near the inlet and introduce a rotating sub-domain further
downstream that carries the propeller (using sliding interfaces or a range of
overlapping cells to establish a connection of the sub grids, (Figure 7-1).
 Keep the whole domain rotating as a single block (which is achieved by time
stepping). Then also the boundary elements at the inlet will be rotating, changing their
global coordinates accordingly. The wake at the inlet will be stationary if the reference
coordinate system is the global system (Figure 7-2, left). The wake will be counter
rotating in a reference system, which is attached to the (Figure 7-2, right).

Figure 7-1: Fixed and rotating parts: a) one cell layer consisting of the cells adjacent to the
Inlet-, Outlet- and outer wall boundaries (left); b) A disc including Inlet boundary and several
cell layers downstream (right).
Figure 7-2: Two alternatives: a) motion solely assigned to flow pattern at the Inlet (left); b)
whole domain in motion, i.e. every grid node rotating around shaft axis (right).
7.3.3. Domain size
If a ship hull is part of the analysis the size of the computational domain is to follow the
recommendations of WP 1 for viscous hull flow investigation.
7.3.3.1. Numerical tunnel, wake at inlet

If the ’influence’ of the hull is incorporated via a measured or computed wake field, the size of
the computational domain can be reduced considerably and a numerical tunnel section may
be considered to define the domain size. The distance between the Inlet and the propeller
plane is a crucial issue. It is desired to perfectly reproduce the wake at the propeller plane.
This should be checked within an additional analysis on an extra grid where the propeller

block is replaced by fluid cells. It is also important to consider induction effects of the propeller
at the Inlet. In the 2nd VIRTUE WP4 workshop on propeller cavitation this problem was
investigated in detail (Summary of results in Salvatore et al., 2009). A distance of 1.25
propeller diameter was found to be a fair compromise to arrive at an acceptable low propeller
induction combined with an acceptable low diffusion of the wake on the way between Inlet
and propeller plane.
7.3.3.2. Downstream and lateral extent

It is suggested to place the downstream termination of the computational domain at least 4-5
propeller diameters behind the propeller plane. The lateral surfaces of the domain (it may be
just one cylinder surface if the propeller was put in a circular tube) should be positioned at 2-3
diameters distance from the propeller axis.
7.3.3.3. Tunnel for propeller/rudder interaction

If a rudder is to be integrated in the analysis (Figure 7-3) the cylinder shape of the domain can
usually not be maintained, since the rudder is typically cut by the stern contour of a ship which
would then define the top boundary of the flow domain. Such cases have been studied by
Streckwall et al. (2007) and Luecke et al. (2009). The distances of sides and bottom for such
a set up could be chosen according to the above recommendations (which consider a
propeller in a tube) replacing the propeller diameter by the rudder height.
Figure 7-3: Sample showing a limited domain for propeller / rudder analysis with prescribed
flow (wake) at Inlet (simulation is to be run using a rotating sub-domain for the propeller as
outlined in Figure 7-1).

7.3.4. Meshing
For the propeller it is obvious to restrict the work on the grid by focussing on the domain
around one blade. The cyclic repetition of this single blade domain has to be assured, which
will lead to further restrictions and will affect the topology and probably also the cell quality.
7.3.4.1. Quality check

The quality of the propeller surface description has to be checked with respect to...
 leakages
 knuckles that are probably not in accordance with the intended geometry
 completeness of tip definition
 missing information (hub, shaft) or elements (fairing at the hub).
7.3.4.2. Structured/unstructured mesh

For a propeller the grid generation process can be time consuming. It is worthwhile to try an
unstructured grid first. A lot of time can be saved in the grid generation process, if such a grid
works. Unstructured grids may completely consist of tetrahedral cells or completely of
hexahedrals or a general mixture of polyhedral cells. Tetrahedral cells are to be avoided at
highly resolved wall boundaries. The Import-functionality of your RANS solver(s) may not
allow for a free choice of grid structures. Check this first, probably by a small test grid.
7.3.4.3. Near wall mesh

The grid at the surface has to reflect the intended near wall resolution (to run the solver either
with wall functions (assuring that the normalised wall distance for the first cell centre reads
y+=20-100 or to use y+ near 1 in order to resolve the boundary layer down to the surface). In
any case a guess of the absolute cell height at the wall has to be made using the propeller tip
speed (R) and a typical dimension of the propeller (blade chord). The expansion ratio of the
cell sizes should not be moderate (below 1.1), if the analysis of cavitation is to be included
one should use a very small rate like 1.02.
7.3.4.4. Critical mesh parts

Critical zones for a structured hexahedral grid are blade hub and tip. There are certain areas
for a propeller where the ’gradient’ of cell sizes (in terms of volume and shape) should be
small. The grid around the trailing edge and the leading edge of a propeller should be smooth
in this respect. Usually the flow is separating at the trailing edge and there is nothing like a
stagnation point close to the trailing edge. To capture this behaviour, near the trailing edge
one should give the same attention to the grid quality and the grid resolution as near the
leading edge. It seems also advisable to keep the grid around trailing edge and leading edge
locally structured. This demand does not naturally lead to an overall block-structured grid but
it is one powerful argument to rely on such a grid structure. Figure 7-4 shows a typical blade
surface resolution while Figure 7-5 and Figure 7-6 show pressure results achieved on such a
grid. Especially Figure 7-6 indicates that the leading edge resolution (though it appears quite
fine in Figure 7-4) is still on the coarser side and should be improved at least if cavitation is to
be simulated. A cut of the blade leading edge located very close to the tip is presented in

Figure 7-7. Such a detailed resolution is suitable to capture vortex structures that are shed at
the tip.
7.3.4.5. Cyclic matching mesh

If a block structured grid topology is intended, it is natural to generate it for 1/Z of the whole
flow domain if Z is the number of blades. It is to be organised such that a duplication of the
grid with an angular offset of 2/Z leads to a suitable grid for the entire propeller. This means
that the two lateral sides of the basic grid should be cyclic repeatable. Perfect, but hard to
achieve would be a cyclic matching of every cell face from one lateral side to the other. If cell
faces are non-matching one should assure that there are only moderate jumps (1:2 say) in
cell face area from one side to the other.
7.3.4.6. Local refinements

Some solvers allow for local refinements. This can be a helpful and fast tool to improve the
grid resolution locally (usually not the grid quality and not necessarily the quality of the
surface representation). However for less robust calculations like an analysis including
cavitation or a free surface it may be risky to use local refinements due to the more complex
matching of the cell faces which may not be sufficiently supported by the interpolation
algorithms.
Figure 7-4: Emphasised blade surface resolution (leading edge to the right) and indication of
the cylinder cut referenced in Figure 7-5.
Figure 7-5: Typical result for a fully wetted propeller flow: field pressure on cylinder surface
indicated in Figure 7-4; leading edge on the right.

Figure 7-6: Leading edge details including near surface mesh of cut from Figure 7-4.
TIP SECTION
Figure 7-7: Resolution of the blade tip suitable to investigate vortex structures
7.3.5. Boundary conditions and initial conditions
7.3.5.1. Outlet
For propellers, especially if a two phase flow model is run, it is generally recommended to use
a pressure boundary at the downstream end of the domain. It has to be noted, that the
slipstream structure does not really comply with a fixed pressure as commonly prescribed in a
pressure boundary condition. The condition will locally deform the slipstream and it has to be
assured that this deformation remains at a considerable distance behind the propeller plane.
Therefore in the above the downstream termination of the computational domain is
recommended to be at least 4-5 diameters behind the propeller plane.

7.3.5.2. Lateral
Usually a slip wall condition is an adequate choice for the lateral boundaries. The propeller
may however be considered in an inclined flow situation, which is the simplest way to include
a cross flow in the simplified (w/o hull) analysis. In this case the lateral boundaries should be
supplied with a fixed flow condition, which is actually provided by the inlet boundary condition.
7.3.5.3. Inlet
The velocity at the Inlet may be constant (if the propeller is studied in open water) or a
prescribed distribution (if the propeller is considered in a wake field). The turbulence
quantities have to be set at the Inlet boundaries if a turbulence model is invoked. For propeller
flow turbulence intensity I  u / U is usually set to 1-3 % at the inlet, where u  is the root-
mean-square of the turbulent velocity fluctuations ( u '  1/ 3(ux 2  uy 2  uz 2 ) ) and
U denotes the mean velocity. This medium turbulence level corresponds to values that also
hold roughly for experiments in cavitation tunnels. If a tunnel test setup is to be re-analyzed or
pre-analysed, the turbulent length scale, describing the size of the larger eddies, may also be
related to a typical tunnel dimension, say 5% of the test section diameter. If there is no
reference to any performed or planned tunnel tests one may start with the above
recommendations (assuming a typical propeller diameter to tunnel diameter ratio of 4) and
besides investigate lower estimates.
7.3.5.4. Initial conditions

Initial conditions should not change the final results but will influence the convergence. In the
propeller case, initial settings for the velocity field can be constant values of u, v and w for
every fluid cell. The constants have to comply with the undisturbed velocity field. If cavitation
is to be modelled it is natural to initialise the vapour volume fraction with zero everywhere.
7.3.5.5. Standards
Some in house standards for the nomenclature of boundary indexing should be followed. This
will make the pre processing easier. For the grid read by the solver one could for instance use
a boundary index nomenclature like...
Boundary Condition
In Out Symm. Wall Pressure
st
1 Group 1 2 3 4 5
nd
2 Group 11 12 13 14 15
rd
3 Group 21 -- 23 24 --
th
4 Group[ -- -- -- 34 --
Table 7-1: Proposed key to relate groups of boundaries to boundary conditions

7.3.6. Controlling the calculation

We presume that the decision on the equations to be solved has been taken, which includes
the choice for the turbulence equations and, if cavitation is to be addressed, the selection of
the two- phase flow model. Then the main parameters that are submitted to the solver to
control the calculation read:
 the type of analysis (steady/ unsteady)
 (for an unsteady analysis) the time step, the number of outer iterations per time step
and the total number of time steps
 (for a steady analysis) the number of iterations
 the convergence criteria
 the fluid properties (density, laminar viscosity etc.) of background fluid, phase No. 1 ...
 the interpolation scheme used for each equation (including blending with lower order
schemes)
 the accuracy of the internal number storage (single or double precision)
7.3.6.1. Time step and total number of time steps

Usually propeller flow problems, especially if cavitation is involved, need an unsteady
approach based on a time stepping procedure. For a RANS approach the wake field
harmonics determine more or less the time step length. Due to the connection to the wake
field it is reasonable to express the time step first in terms of an angular interval. Here a
distance
d  1o  2 o is recommended. To arrive a the associated time step dt one can use
with n as shaft frequency [1/s]:

dt  d[ o ] / 360 / n . For large eddy simulations, that resolve
the turbulent flow on a macroscopic scale, time steps are typically smaller by a factor of 100.
For propellers the maximum number of time steps may be counted in terms of revolutions that
are simulated after starting from the initial conditions. For a periodic propeller flow problem it
is emphasised to include about 4 revolutions. The history of the one-blade thrust may serve
as an indicator for the periodicity of the flow.
7.3.6.2. Outer iterations

If the propeller is rotating within an environment of fixed cells (being connected to the fixed
grid via sliding interfaces or overlapping grids) it appears necessary to assure about 10 outer
iterations per time step. To assure this, the convergence limit may have to be set to much
smaller values than the default value.
7.3.6.3. Differencing schemes

We consider that the actual solver at hand is based on a finite volume approach. Then the
scheme to interpolate cell centre quantities to face values has to be specified for each set of
equations (momentum, turbulence, mass and the transport equation for the second phase).
Usually it is helpful to base the special discretisation of the momentum equations on a 2nd or
higher order scheme and use some blending with a 1st order scheme to enhance stability.

7.3.7.1. Standard results

As mentioned earlier, model propellers are rarely subject to detailed flow measurements,
especially for ‘in behind’ cases. With this lack of standard results from experiments one is
quite free to consider one’s own set of characteristic results and to extract it from the RANS
computations. In any case the characteristic results should include plots or graphs that allow
judging the quality of the analysis. For propellers, standard results are:
 surface pressure
 pressure iso surfaces
 field pressures (on cylinder surfaces cutting the blade, on surfaces cutting the tip
vortex etc.)
 blade surface streamlines
 streamlines within the fluid (e.g to visualise tip vortex structures)
 Iso surfaces of vapour volume fraction (see Figure 7-8 below)
 forces and moments (single blade and overall propeller)
If an existing design is to be improved the shortcomings may be on the performance side or
concern the local flow at hub or tip. In the latter case the focus is clearer defined but it is still
useful to produce a sequence of standard results for comparison and confirmation.
Figure 7-8: Propeller (and rudder-) cavitation calculated for ‘in-behind’ conditions. Here the
cavities are visualised by a vapour volume fraction iso surface  v =0.3. The blade angular
o
position varies in angular steps of 20 . The fluctuating cavity volume gives rise to pressure
fluctuations, which can be realised at any point in the flow.
7.3.7.2. Normalisation
We propose to normalise pressure, velocities and forces/moments as follows:
 pressure: pressure in the flow and on boundaries p normalised as
C p  ( p  p ) /(1/ 2    (nD) 2 )
(122)
 vapour pressure pv normalised as

 n  ( p  pv ) /(1 / 2  (nD) 2 ) (123)

 velocity
either by v '  v /(n  D) (124)
v' '  v / V
or by S (125)
V
with S denoting some reference speed of the undisturbed flow
 forces normalised as KT
K  F /(   n2 D 4 )
F (126)
 moments normalised as KQ
K  M /(   n2 D5 )
M (127)
7.3.7.3. User coding

It may be necessary to link user subroutines to influence the computational set up or support
the post processing. The intention to apply user programming can even influence the grid
generation process as groups of boundary elements or fluid cells are to be indexed such that
they can be easily addressed in a user subroutine. For instance one might want to monitor the
contribution of the pressure side of the blade to the total blade force. Then one has to sort out
blade surface elements belonging to the pressure side and those belonging to the suction
side. This is preferably done during the grid generation process already, giving groups of
boundary elements indices that are related to the imposed boundary condition according to a
common key.
7.4. Application Cases

At the first Virtue-WP4 Workshop in Wageningen, held October, 2007, two test cases were
defined:
 The first test case selected aimed to address the underlying flow mechanisms. To this
end, a twisted hydrofoil for which adequate experimental data and observations were
available was selected, designated “Delft twisted foil 11” (see Foeth, 2008, for an
extensive description).
 The other test case aimed to address a so called industrial challenge. To this end, the
INSEAN propeller E779A was selected, also because high quality experimental data
were available for this propeller in several conditions.
During the discussions on the above two test cases, it was noted that there were many
aspects of the simulation results that were unsatisfactory or that needed clarification and
studies in more detail. Numerical uncertainty was one of the main suspects held responsible
for the observed differences. It was therefore decided that for the second workshop, a 2D
case would have to be included as a verification exercise that could also allow for a further
study of the numerical uncertainty.

7.4.1. Case 4-a: 2D Foil
7.4.1.1. Introduction
The main objective of this exercise for the second workshop in Oct. 2008, is to compare the
numerical results of all participants for the same configuration under both non-cavitating and
cavitating conditions. The case selected for this study is a NACA0015 at an angle of attack of
6 deg. It is a simple geometry and a simple flow, thereby excluding effects of complex flow
phenomena that could blur an analysis on numerical uncertainty.
Additional arguments to select this foil were that experiments on this foil were planned at
HSVA for June, 2008, and that this case has also been used for verification and testing in
literature.
The reasons for not validating with experiments are twofold:
 First, it is of interest to discuss and clarify the differences between different numerical
solutions regarding implementation, solution techniques, turbulence modelling,
numerical schemes, liquid-vapour interface handling, mass transfer modelling, and so
forth.
 Second, 2D cavitation can not be validated since there is no such phenomenon as 2D
cavitation in the real world; the closure region will always show 3D effects.
7.4.1.2. Geometry definition

The NACA0015 foil is described as

 y  0.75 0.2969 x  0.126 x  0.3516 x 2  0.2843x 3  0.1015 x 4



 x   0,1 (128)

 y  0, 2 tmax 
1
with the chord length c = 200mm and a sharp trailing edge. It is here rotated over 6 deg
around the centre of gravity at x/c = 0.3086. The size of the computational domain for these
computations should be set to 1400 x 570 mm, thus extending 2 chord lengths ahead of the
leading edge, ending 4 chord lengths behind the trailing edge (in relation to 0 deg angle of
attack) and with a vertical extent reflecting the size of the cavitation tunnel where the foil will
be tested by HSVA, see Figure 7-9. With the origin in the centre of rotation, the extent of the
domain is x∈ [-461.72, 938.28] and y∈ [-285,285].
It is noted that this polynomial description leads to an open end of the foil at the Trailing Edge.
This undefined gap appeared to cause some scatter in the way the Trailing Edge was
modelled. Some participants tapered the aft part of the foil section, so as to obtain a fully
closed foil, where others followed exactly the polynomial definition given above, and closed
the foil with a blunt cap. Computations with the FRESCO code showed that the difference
between the two approaches lead to a difference in lift coefficient of approx. 3%.

Figure 7-9: Definition of the geometry
7.4.1.3. Flow and boundary conditions

Computations could be performed on three meshes (designated G1 through G3) to allow for
grid refinement studies. Two baseline meshes G1 were provided to all partners in CGNS
format; this should be seen as a minimum mesh density for all computations. The wall normal
resolution is for one mesh adjusted for use with wall functions and for the other for wall
resolved simulations. Uniformly refined meshes were also made available.
For those not able to read the ICEM CFD-generated meshes, also the mesh points on the
wing were provided in a separate text file for the baseline mesh, together with wall normal
distribution definition.
 Inlet: Velocity v0=6 m/s and a specified turbulence level of 1 %.
 Top and bottom: Use free slip conditions.
 Outlet: Use an outlet pressure of 0 kPa.
7.4.1.4. Modelling and numerical approach

The participant was free to choose modelling, numerical and solution approach. It was
however requested that all simulations from one partner were performed with a common
configuration, i.e. the same turbulence model, numerical schemes etc. should be used for
both non-cavitating and cavitating conditions. Additional computations for sensitivity studies of
the settings were welcomed but not required.
Flow conditions
Three conditions should be simulated:
 Non-cavitating condition; this condition could be simulated as steady.
 Cavitating (expected to be steady) at   1.6 ; this condition could be simulated both

as steady and unsteady.
 Cavitating (with expected shedding) at   1.00 ; this condition should be simulated as

unsteady.
The cavitation number  is defined here as

 pv
 (129)
1
2 l v02
where pv = vapour pressure
l = specific mass water

v0 = uniform inflow velocity
Physical quantities:
 Water temperature: 24°C
 Water density: l  998 kg/m3 (corresponding to 24°C)
 Vapour density:
v  0.023 kg/m3 (corresponding to 24°C)
7.4.1.5. Requested information

General information on the computational configuration, including
 Numerical method used, including
o Discretisation schemes
o Solver procedure
 Turbulence modelling approach
 Cavitation modeling approach, including
o Interface capturing methodology
o Mass transfer model
o Calculation procedure
o Water/vapour density ratio (i.e. actual used vapour density)
 Analysis and discussion on the effect of mesh refinement on e.g.
o Convergence behavior of global parameters like forces and shedding
frequency and flow details like minimum pressure and its location.
o Possible qualitative differences in the cavitation behaviour.
For the Non-cavitating conditions, the results computations were performed on 3

consecutively refined meshes, starting from the base line mesh.
 Lift and drag on the foil, with friction and pressure components reported separately.
 Cp distribution along the foil.
 Minimum value of Cp and the location where this pressure minimum is attained.
 Velocity and pressure fields (in Fieldview readable format, see below)

For the Cavitating conditions (expected to be steady), both average and time series (in ASCII)
were reported. In order to verify the results, computations were performed on two
consecutively refined meshes, starting from the base line mesh.
 Cp distribution along the foil.
 Minimum value of Cp and the location where this pressure minimum is attained.
 Vapour volume and cavity extent (measured for the contour α=0.5, where α denotes
the vapour fraction).
 Field data (in Fieldview readable format, see below) for velocity, pressure, density,
volume vapour fraction and mass transfer (“cavitation model”) source terms.
For the Cavitating conditions (with expected shedding), both time series (in ASCII) as well as
average (computed from data after initial transients have disappeared) were reported for:
 Cp distribution along the foil (only average).
 Minimum value of Cp and the location where this pressure minimum is attained (only
average).
 Vapour volume.
 Shedding frequency.
 Pressure and vapour fraction in the points defined in Table 7-2.
 Field data (in Fieldview readable format, see below) for velocity, pressure, density,
volume vapor fraction and mass transfer (“cavitation model”) source terms for 6
different time steps within the last computed shedding cycle.
 h. Animations of the flow being in line with the styles used in Figure 7-10.
X Y
P1 -0.025 0.0167
P2 0.07 0.025
P3 0.07 0.0035
P4 0.138 0.025
P5 0.138 0.0
P6 0.2 0.025
P7 0.2 0.0
Table 7-2: Requested positions for pressure logging Note: The origin of the reference frame
(0,0) is positioned in the rotation point (centre of gravity) with the x-coordinate in the flow
direction and the y-coordinate in the vertical direction (see also Figure 7-9)

Figure 7-10: Style for the animations

a- Vapor fraction and velocity vectors
b- Pressure and vapor fraction (in white) for  =0.25, 0.5 and 0.75
7.4.1.6. Impression of Results

All results presented during the workshop have been issued on a DVD and can also be found
on the VIRTUE website. A summary of salient results is given below. These results were
collected using a diversity of modelling and solver techniques including compressible and
incompressible viscous approaches as well as Euler flow modelling.
A first comparison between the lift and drag results of the different codes for the non-
cavitating condition is given in Figure 7-11 and Figure 7-12.
Figure 7-11: Comparison of computed lift coefficients with the mean coefficient.
Notes: CL mean 0.65  5% - 8 of 10 results were within 5% of the mean
There appears to be a significant scatter when all results are observed, although 8 out of 10
results are within a bandwith around the mean of 5%.

Figure 7-12: Comparison of computed drag coefficients with the mean coefficient. Notes:
CD mean 0.014  14% - 5 of 9 results were within 14% of the mean
The results between different codes show significantly more scatter for the drag coefficient, as
can be seen in Figure 7-12.
A sensitivity study of the non-cavitating flow characteristics is reported by Hoekstra and Vaz
(2009). Although these authors did not use the standard grids provided for practical reasons,
the grid densities and topologies are similar. The grids can be considered as multi-block
structured, but are dealt with as unstructured by FRESCO. An impression of the grid topology
is given in Figure 7-14. This consists of an O-grid embedded in an H-grid. Stretching of the
grids towards the foil surface has been applied to obtain full resolution of the boundary layer
+
flow (y = (1); no wall functions used). Only hexahedral cells appear (quadrilaterals in 2D).
Table 7-3: Grid characteristics for grid sensitivity study (Hoekstra et al., 2009)
In all computations by Hoekstra et al. (2009), the SST version of the    turbulence model
has been used (Menter, 1994). An inflow turbulence level was set at t  0.01 , where  is
the dynamic viscosity of water (   1.02 105 kg/ms). No special models were used for
laminar-turbulent transition modelling, implying that the turbulence model governs the
transition process.

The sensitivity of lift and drag for the three grids is presented in Table 7-4, from which it can
be concluded that there is little effect of the grid densities used on these integral quantities. A
similar conclusion can be drawn when looking at the extremes in the pressure distribution, as
presented in Table 7-5.
Table 7-4: Sensitivity of lift and drag for three different grids (Hoekstra et al., 2009)
Table 7-5: Sensitivity of extremes in pressure distribution for three different grids (Hoekstra et
al., 2009)
An impression of the differences in predicted sheet cavity shape and extent are presented in
Figure 7-13. It can be observed that there are significant differences again in cavity extent,
but moreover, also in vapour distribution. Without further analysis, these differences could be
caused by numerical uncertainty, differences in turbulence modelling as well as differences in
cavitation modelling.
The differences observed in the non-cavitating integral quantities as lift and drag indicate that
further studies on the improvement of numerical uncertainty and adequacy of turbulence
modelling are needed, before a more discriminating analysis can be made of the cavitation
models.
Figure 7-13: Distribution of vapour fraction  and cavity extent for a cavitation number
 =1.6 for six different codes

The effect of grid density on the viscous dissipation in the closure region of the sheet cavity
was also studied by Hoekstra and Vaz (2009).
Figure 7-14: Effect of grid density on eddy viscosity distribution near the sheet cavity interface
for a steady cavity at a cavitation number  =1.6 (from Hoekstra et al., 2009)
Figure 7-14 shows that the grid density, although not appearing to have a major impact on the
lift and drag coefficients, does have a significant effect on the eddy viscosity distribution. The
thickness of the layer in which the eddy viscosity plays an important role diminishes with
refined grid, and also the distribution of the eddy viscosity in the re-entrant jet becomes
different. This sensitivity is likely to affect the dynamic behaviour of the sheet cavity for
unsteady cavities. Reboud (1998) suggests using a reduced eddy viscosity near the cavity
interface, which affects the eddy viscosity distribution in a different way.
7.4.1.7. Conclusions and Recommendations

The 2D NACA0015 test case demonstrates that there are still significant differences in the
results of the non-cavitating characteristics, such as lift and drag. A grid sensitivity study has
demonstrated that this is not likely attributable to grid density issues, but the causes for these
differences should rather be sought in numerical uncertainties and turbulence modelling.
The importance of a sufficient grid density near the cavity-liquid interface has been identified.
This NACA0015 test case appears suitable as a standard test case to investigate the
adequacy of numerical modelling of cavitating flows.

7.4.2. Case 4-b: 3D twisted foil
A recent set of experimental data on 3D sheet cavitation has been gathered by Foeth (2008).
By using a stationary but twisted foil, interference of the cavity with the side walls of the tunnel
was avoided. The cavitation behaviour was studied under steady and unsteady inflow, the
unsteady inflow being generated by two synchronously oscillating foils just ahead of the test
object. In two workshops of the VIRTUE project, the numerical simulation of some parts of
these experiments has been set as an application challenge.
7.4.2.2. Geometry definition of the Delft Twist 11 hydrofoil

The Delft Twist 11 Hydrofoil is a wing of rectangular plan form, the section shape being
uniform over the whole span, but the orientation with respect to the incoming flow varying in
spanwise direction. If the angle of attack is 0 degrees at both ends of the wing, it is 11
degrees at mid-span. Details of the wing’s geometry are given below.
7.4.2.2.1 Coordinate system
The reference coordinate system is a right-hand system with the x-axis in flow direction, the y-
axis in spanwise direction and the z-axis directed upwards. The origin of the coordinate
system is at mid-chord and mid-height of the wing section on the tunnel wall. At the same
time, the origin is located on the side wall of the tunnel at mid-height of the test section. The
cross-section of the tunnel measures 0.30 × 0.30 m.
7.4.2.2.2 The section profile shape
The section shape is a modified symmetrical NACA 4-digit profile. Its basic shape is given by
Abbott and von Doenhof by the thickness distribution:
 z ( x) 
t
0.20

0.29690 x  0.12600 x  0.35160 x 2  0.28430 x 3  0.10150 x 4 
(130)
in which x, z and t have been non-dimensionalised with the chord length c, which is 0.15 m in
absolute size, while the thickness-length ratio is t =0.09. Although this section shape has a
finite thickness at the trailing edge, viz. 2z|x=1 = 0.00185, it was just too small to satisfy the
requirements of the milling process by which the Delft Twist 11 foil was manufactured. A
correction on the thickness distribution was therefore applied which is defined by
2
 x  x sp 
 z ( x)    t  z | x 1 H ( x sp ) (131)
 1  x  min
 sp 
with xsp = 0.35, tmin = 0.0002/c while H is the Heaviside function. So the thickness is
gradually increased from 35% of the chord-length towards the trailing edge to get an absolute
trailing edge thickness of 0.4 mm.

7.4.2.2.3 The twist-11 foil
The twisted foil has been constructed based on a spanwise angle of attack variation of the
section profile given by
Error! Objects cannot be created from editing field codes.
(132)
where y is again non-dimensionalised with the chord length c and varies over the spanwidth,
being twice the chord length ( 0  y  2 ). As the name of the foil indicates,  max = 11
degrees. The angle of attack variation is achieved by rotating the section shape about the
mid-height point at 75% of the chord length from the leading edge.
Figure 7-15: Top side and front views of the

Figure 7-16: Distribution of the geometric angle
hydrofoil. The black outline indicates the
of attack of the hydrofoil. The angle at the sides
viewing area of Figure 7-18
is taken as the reference angle for the whole
geometry.
7.4.2.2.4 Positioning of foil in the tunnel
The foil was positioned in the tunnel with the mid-chord point on the symmetry line of the
section profiles at the tunnel walls at the mid-height of the test section. Subsequently the
whole foil was rotated about the axis connecting these two points by -1 degree, i.e. leading
edge downward.

Figure 7-17: A close up of the test section, showing the hydrofoil, the camera location and
effective viewing area
7.4.2.2.5 IGES file
An IGES file with the complete geometric description of the Delft Twisted Foil as a B-spline
surface is available. It contains the geometry in real size, in mm. The coordinate system for
the IGES definition is as described above. To put the wing at a different angle of attack (which
is -1 degree on the IGES file) a rotation about the y-axis suffices.

The tunnel test section measures LxBxH=600 x 300 x 300 mm. The span of the wing is 300
mm, so it reaches from one side wall to the other.
The following settings were recommended:
 Choose the length of the computation domain as 7 chord lengths, starting 2 chord
lengths ahead of the leading edge, ending 4 chord lengths behind the trailing edge.
 Apply free-slip boundary conditions on the tunnel walls.
 Calculate only half of the wing, i.e. use a symmetry boundary condition at the centre
of the tunnel.
 Perform steady computations for the wetted flow conditions and unsteady
computations for the cavitating flow condition.
 Simulate unsteady inflow conditions by time-dependent external forces at the tail of
the oscillator foils operating at a frequency of 32 Hz.
Recommended simulations:
1. Wetted flow simulations for angle-of-attack at the tunnel wall of -1 and -2 degrees and
an onset flow speed of 6.97 m/s. Purpose: to check lift and drag prediction and to
verify the response of the numerical code to a slight change in the angle-of-attack.
Grid refinement study is mandatory to estimate the numerical uncertainty.
2. Wetted and cavitating flow for the foil at -2 degrees angle-of-attack. Inflow speed:
6.97 m/s; outlet pressure: 29.0 kPa; water temperature: 24 °C; vapour pressure: 2.97
kPa; water density: 998.0 kg/m3 (corresponding to 24 °C); vapour density: 0.023
kg/m3 (corresponding to 24 °C). Purpose: to predict the shape and dynamic
behaviour of the cavity and to study the change of the flow due to the presence of
cavitation. Check numerical accuracy by grid refinement.
3. Cavitating flow in steady and unsteady inflow. Same conditions as under 2. Purpose:
to verify the lock-in of the cavity shedding frequency by the forced oscillation of the
inflow.

7.4.2.4. Underlying flow mechanisms
Figure 7-18: Visualization at 4.96 m/s±6.4%,  =1 deg, =0.66±7.94%, recorded at 2 kHz but
showing every seventh frame. Flow from top to bottom

7.4.3. Case 4-c: The INSEAN E779A propeller
Two VIRTUE WP4 Workshops on propeller cavitation modelling were held in Wageningen in
2007 and in Rome in 2008. The two workshops provided a valuable contribution to the
assessment of state-of-the-art computational models. In addition to other test cases on 2D
and 3D foils addressed above, the two workshops focused on cavitating propellers in uniform
flow (2007 edition) and in a non-homogeneous wakefield (2008 edition).
As for 2D and 3D foil cases, a common propeller-flow configuration was proposed addressing
the INSEAN E779A model. This propeller, used throughout WP4 as the reference case for
validation work, was selected in view of the extensive and accurate set of experimental data
available.
Combining the two workshop editions, a total of eleven institutions have participated and
submitted results for the proposed test case. Nine institutions presented results obtained by
RANS codes, one by LES, and one by an inviscid-flow BEM code. Participation was open to a
limited number of selected institutions not involved in the VIRTUE project to have a
comprehensive picture of current capabilities of cavitating propeller flow modelling by CFD.
Details of results from the two VIRTUE WP4 Workshops may be found in Streckwall and
Salvatore (2008) and in Salvatore et al. (2009).
7.4.3.2. Geometry definition

The INSEAN E779A is a four-bladed, fixed pitch, right-handed propeller, originally designed in
the late 1950’. Blade skew and rake are small, pitch is almost constant along radius resulting
into a mean pitch ratio of P/D=1.1. The E779A propeller model and a table with basic
geometrical parameters is given in Figure 7-19.
Digital mapping of this model propeller has been performed to provide an IGES file to give a
very accurate mathematical description of blade shape. When the computational grid around
the propeller is built from this IGES description of the propeller surface, the definition of the
coordinate system is implicit to assumptions in the IGES file.
In addition to the mathematical description in IGES, the propeller geometry is also given
through classical tables describing radial distributions of major parameters (pitch, chord, rake
and skew) and through definition of sectional offsets.
Number of blades 4
Model diameter 227.27 mm
Mean pitch ratio, P/D 1.1
Expanded area ratio 0.689
Skew angle at tip 4° 48’ (positive)
Rake 4° 35’ (forward)
Hub diameter ratio 0.20
Figure 7-19: E779A propeller model and basic geometrical parameters.


To have comparable results from all the workshop participants, common geometrical layout,
flow and boundary conditions were proposed.
Propeller flow calculations address a propeller operating inside a cavitation tunnel.
Specifically, the test case refers to the set-up realized at the Italian Navy Cavitation Tunnel
(CEIMM, Rome) where most of the experimental results collected into the INSEAN E779A
dataset were obtained.
The E779A propeller model inside the cavitation tunnel is depicted in left Figure 7-20. In
particular, the set-up for analysing the propeller in non-homogeneous flow is shown. A five-
plate array flush mounted on tunnel top wall in front of the propeller is used as the wake
generator. The velocity defect distribution induced at the propeller plane and measured by
Laser-Doppler Velocimetry (LDV) is shown in the right Figure 7-20. The resulting velocity field
is roughly representative of a single-screw hull wake.
Figure 7-20: Experimental set-up of propeller model downstream a wake generator (left) and
non-homogeneous axial velocity field measured by LDV.
An idealized propeller-in-tunnel configuration is defined. To simplify CFD modelling, the actual

squared tunnel cross section is replaced by a circular section with identical sectional area.
This allows taking advantage of axial symmetry of flow field in case of propeller in uniform
onset flow. Numerical tunnel inlet section is placed at distance not lower than 1.25 D from
propeller plane, whereas tunnel outlet section is recommended at distance Lout = 4.0 D
downstream the propeller plane, as sketched in Figure 7-21. These dimensions are deemed
adequate to avoid numerical solution disturbances propagating from boundaries to the whole
computational domain in case of inlet/outlet boundaries too close to the propeller plane.

Figure 7-21: Sketch of the numerical tunnel test section idealizing the set-up in Figure 7-20.
At the inlet section, axial velocity is prescribed as uniform or by imposing te velocity

distribution from LDV measurements depicted in right Figure 7-20. Zero transversal velocities
are also enforced. At the outlet section, zero pressure is imposed. Although the modelling of
tunnel wall boundary layer would be desirable, an adequate modelling of propeller flow can be
obtained by imposing simple slip conditions at walls.
Particular care requires the case of a propeller operating in a non-homogeneous wakefield.
Modelling the whole configuration represented by wake generator and shafted propeller within
the tunnel test section is possible, as demonstrated by a study performed by VTT using the
RANS code FINFLO. Nevertheless, this represents a really challenging application of CFD
codes. In fact, the wake generator geometry implies that a very complex computational grid is
built around each plate. Grid details in this flow region are then added to grid blocks
necessary to describe the fluid region around the propeller and a huge computational burden
follows.
To overcome problems related to modelling the fluid domain around both wake generator and
propeller, a different approach was considered for workshop calculations. Specifically, the
wake generator is removed from the computational domain and the generated wakefield is
taken into account by imposing a non-uniform velocity distribution at the inlet section of the
computational domain.
Flow conditions addressed in the workshops can be summarized as follows:
 Propeller in uniform flow, J = 0.71. Non cavitating, and cavitating at σn = (p-pv)/ ½
2
(nD) = 1.763
 Propeller in non-homogeneous flow, J = 0.90. Non cavitating, and cavitating at σn =
4.455
Non-homogeneous inflow is specified through LDV measurements of the velocity field
downstream the wake generator in Figure 7-20 at the propeller plane. The propeller is
removed and hence the measured velocity distribution is akin to the ‘nominal’ wake in a
propeller-hull configuration.

7.4.3.4. Propeller flow modelling results

Combining participation to the VIRTUE-WP4 2007 and 2008 Workshops, numerical results
from nine RANS, one LES and one BEM models are compared.
A detailed description of different solvers is not given here and may be found in Streckwall
and Salvatore (2008) and in Salvatore et al (2009).
In spite of a simple geometry characterized by an almost constant pitch distribution and low
skew, the E779A propeller represents a very interesting test case for validation of CFD codes.
Even if tested at design conditions, a strong negative pressure peak is observed at blade
suction side leading edge and an intense tip-vortex is generated. No face cavitation is
observed for a wide range of operating conditions including those considered in the
workshops.
7.4.3.4.1 Uniform inflow
The analysis of non-cavitating, uniform flow calculations highlight a sensible agreement

among results from different computational models. In particular, evaluated pressure
distributions on blade suction and pressure side are in fair qualitative agreement. Quantitative
differences are observed only in the blade tip region where some solvers detect tip-vortex
separation at radial section r/R = 0.95 in contrast to separation at tip predicted by most codes.
Pressure coefficient predictions at different radial stations are compared in Figure 7-22.
Figure 7-22: E779A propeller in uniform non cavitating flow.

Pressure coefficient distributions on blade at radial stations r/R = 0.7 (left) and r/R = 0.95
(right).
The agreement in predicted blade pressure distributions is confirmed by comparing open

water performance results, as shown in Figure 7-23. With the only exception of two codes,
most results fall within a very narrow uncertainty band. For reference, Figure 7-23 also shows
the comparison with experimental data referring to propeller tests in open water. Considering
the range of variation of advance coefficient J in Figure 7-23, the uncertainty in predicting
measured thrust and torque is lower than 5%. At J = 0.71, the scatter among five out of
seven numerical results is about 1% for both KT and KQ.

Figure 7-23: E779A propeller in uniform non cavitating flow. Predicted thrust coefficient (left)
and torque coefficient (right). Experimental data from open water tests are given for
reference.
A rough estimate of the cavity extent on propeller blades without invoking a cavitation model
can be derived by analysing the flow domain where the pressure in wetted flow conditions
drops to values close to the vapor pressure. Aiming to address flow conditions at σn = 1.76,
isopressure contours at CP = -1.0 in a flow region surrounding the blade suction side provide
interesting information, as described in Figure 7-24. Common trends are observed from
different calculations. A region extending from mid-span leading edge to blade tip is clearly
observed in all solutions. It is expected that vaporization mostly occurs in this area.
Differences between the results from different codes become especially apparent in the tip
region and the blade root region. Although an accurate analysis of grid refinement effects was
beyond the scope of the workshops, it is worth noting the impact of grid resolution on
predicted isopressure contours, as shown in Figure 7-25.
Figure 7-24: E779A propeller in uniform non cavitating flow. Isopressure contours at CP=-1.0
in a flow region surrounding the blade suction side.

Figure 7-25: E779A propeller in uniform non cavitating flow. Examples of grid refinement
effects on isopressure contour predictions by two RANS codes. Coarse to fine grid results are
shown from left to right (top figures) and from right to left (bottom figures).
Considering cavitating flow conditions, combined 2007 and 2008 workshops results provide a
unique comparison among ten computational models by RANS, LES and BEM. Results are
shown in Figure 7-24 where tunnel observations are also given for comparison. For viscous-
flow calculations by RANS and LES, cavity extension is assumed to be delimited by vapour
fraction contours with  v  0.5 . The result by a BEM code represents the extension of the
cavity surface determined as the region where pressure equals vapour pressure.
The comparison in Figure 7-26 highlights that state-of-the-art CFD models are able to predict
cavity patterns that are in qualitative agreement with experimental observations. In particular,
the shape of the cavity is captured in most cases, whereas a common trend to overestimate
the extent of vapour regions on the blade surface is observed.
About CFD codes results, it is worth noting that identifying the extension of the cavitating
region with vapour fraction contours at a fixed value is open to discussion. In fact, if a value of
the v threshold larger than 0.5 is used to mark the cavitating fluid region, the resulting cavity
extent is smaller and, referring to results in Figure 7-26, the agreement between predictions
and experimental data would improve. A value v = 0.5 is deemed appropriate for
comparisons with cavity pattern observations from experiments.

Figure 7-26: E779A propeller in uniform cavitating flow. Summary of numerical predictions
presented at the VIRTUE WP4 workshops in 2007 and 2008 and comparison with
experimental data.
7.4.3.4.2 Artificial wakefield
Next, CFD calculations of the unsteady flow conditions when the propeller operates in a non-
homogeneous wakefield are considered. The differences in modelling the inflow to the
propeller are recognized here as a major source for these different results. In some
calculations, too strong a numerical dissipation weakens the wakefield in the propeller region.
As a result, the propeller blade does not reach the maximum loading and the corresponding
calculated cavity patterns tend to be underestimated compared to the experimental
observations. Examples of predicted axial velocity distributions in front of the propeller are
given in Figure 7-27, where experimental data by LDV are given for reference.

Figure 7-27: E779A propeller in non-homogeneous flow, J = 0.90. Axial velocity distributions
at transversal plane upstream the propeller disc. Numerical results and experimental data by
LDV.
Examples of unsteady blade flow predictions are given in Figure 7-28 and Figure 7-29.
Specifically, Figure 7-28 shows isopressure contours at CP=-3.0 in a flow region surrounding
the blade suction side for three different blade angular positions (   -30, 0, +30 deg, with
  0 corresponding to the reference blade in the twelve o'clock position). Figure 11 shows
predicted cavity patterns at σn = 4.455 for blade angular position   0, which roughly
corresponds to the angular position where the vaporised region attached on the blade
reaches its maximim extension.
As anticipated, large discrepancies among predicted cavity patterns are observed and this is
partly due to the different propeller inflow resulting by numerical calculations. Comparing with
experimental results, see Figure 7-30, it may be noted that both overestimated and
underestimated cavity patterns are predicted. The large scatter in estimated cavity patterns is
confirmed by numerical predictions of vaporised region volumes that differ in some cases by
order of magnitudes. These findings raise a question on the reliability of current CFD models
to predict transient cavitation effects like pressure fluctuations, noise and erosion.

Figure 7-28: E779A propeller in non-homogeneous non cavitating flow, J = 0.90. Examples of
isopressure contours at CP=-3.0 in a flow region surrounding the blade suction side for three
different blade angular positions (   -30, 0, +30 deg).
Figure 7-29: E779A propeller in non-homogeneous cavitating flow, J = 0.90. Examples of

cavity patterns prediction by RANS, LES and by BEM (vapour fraction contours with  v  0.5 ).
Blade angular position set to   0 deg.

Figure 7-30: E779A propeller in non-homogeneous cavitating flow, J = 0.90. Tunnel flow
visualisation of transient cavitation at blade angles   -35 to +20 deg.
The comparative analysis of computational results submitted to the two VIRTUE-WP4

workshops brought WP4 partners to draw some general conclusions on CFD modelling of
propeller cavitation that are summarized here below.
Based on the workshop results, it is hypothesized that differences between cavitating flow
predictions are to a large extent caused by grid density and numerical dissipation and to a
lesser extent by the different turbulence and cavitation models. The correct modelling of
cavitation requires that flow regions where vapour generation, transport and destruction occur
are discretised with a much higher grid density than typically needed for the non-cavitating
flow outside the inner boundary layer. LES simulations resolve (not surprisingly) more
structures in time and space than RANS calculations. Transient cavitation results show that
two-phase flow details like the process of leading-edge detachment are in this case
accurately described by LES. Moreover, results comply with the non-periodic nature of
cavitation usually observed on model scale propellers. From the workshop results it appears
that LES-based models are promising tools to investigate the risk of cavitation erosion.
It may be concluded that the grid should not only be refined in the wetted boundary layer flow,
but also at the cavity interface. Too much numerical dissipation seems to result from too
coarse a grid. It is believed that the grid density is likely to be the most important source for
differences between properly verified codes.
In the case of a propeller operating in a wakefield, numerical results are furthermore affected
by the different meshing techniques used to impose the prescribed axial wake. Models based
on a sliding mesh approach where the measured axial wake is imposed as inlet condition are
compared to models where the whole computational domain is rotating with the propeller, and
an idealized analytical definition of the wakefield is used.
An example of LES-predicted flowfield in the tip area of a propeller in non-homogeneous
wakefield is given in Figure 7-31; combined visualization of velocity vectors and vapour
fraction contours help to identify the strong link between cavity pattern and vorticity dynamics.

Figure 7-31: Example of two-phase flow prediction by LES: blade tip flow at two distinct time
steps during propeller rotation in non-homogeneous inflow.
In general, the predicted cavity extents for both steady and unsteady inflow do qualitatively
agree with experimental observations, whereas important quantitative differences are
observed. These differences in cavity extent and volume render computations not sufficiently
suitable for a prediction of radiated pressure fluctuations nor predictions of cavitation erosion.
It is concluded that predictions of pressure fluctuations from a potential flow BEM code give,
so far, the most reliable results.
The relatively low effect of different turbulence and cavitation models used is supported by the
results submitted to the Workshop 2008 by Schmidt et al. (2008) using an inviscid Euler
solver CATUM. In this paper, the authors show that the cavitating vortices in the wake of a
triangular prism are predicted qualitatively well and that the location of the impact pressures
caused by the breaking up of the cavitating vortex corresponds to the experimentally
determined locus of erosive damage. These authors conclude that the mechanism governing
the cavity dynamics is “strongly inertia controlled”. Similar conclusions largely hold for
transient cavitation occurring on propellers operating in the non-homogeneous hull wake.

8. Conclusion
This Best Practise document synthesises, after an introduction on modelling and
computational theory, the best practise usage of Navier-Stokes computations for naval
hydrodynamics across four main applications:
 Towing Tank
 Sea keeping Tank
 Manoeuvring Tank
 Cavitation Tank
The application cases discussed in each section have been chosen to reflect different stages
of maturity for applying Computational Fluid Dynamics: Underlying Flow regime and
Application challenge.
These Best Practise Guidelines will therefore support the CFD specialist and/or the naval
architect by providing him specific guidelines for each application cases, by providing
validation against experiments when available, by comparing with Potential Flow methods
and by underlining the challenges faced by the current state of the art modelling.

9. References
1. AIAA (1988). “AIAA guide for the verification and validation of computational fluid
dynamics simulations”, AIAA G-077-1998.
2. Alessandrini, B., Delhommeau, G. (1999). “A fully coupled Navier-Stokes solver for
calculations of turbulent incompressible free surface flow past a ship hull”,
International Journal for Numerical Methods in Fluid,
vol. 29, pp. 125-142.
3. Alessandrini, B., Delhommeau, G. (1994). “Simulation of three-dimensional
unsteady viscous free surface flow around a ship model”, International Journal for
Numerical Methods in Fluids, vol. 19, pp. 321-342.
4. Apsley, D., Chen, W-L., Leschziner, M. & Lien, F-S. (1997). “Non-linear eddy-
viscosity modelling of separated flows”, IAHR J. Hydraulic Research, vol. 35, pp.
723-748.
5. Bäck (2004). “Evolution Strategies Module R1.1 for modeFRONTIER”, April 9, 2004
Technical Report 2003-003.
6. Baldwin, W.S. & Lomax, H., (1978). “Thin-layer approximation and algebraic model
for separated turbulent flows”, AIAA Paper 87-257.
7. Bensow, R.E., Berchiche, N. and Lu, N.-X. (2008). “NACA0015 – 2D Verification
Case description”, VIRTUE project.
8. Bradshaw, P. (1994). “Turbulence: the chief outstanding difficulty of our subject”,
Experiments in Fluids, vol. 16, pp. 203-216.
9. Bui-Thanh, Damodarany and Willcoxz (2002). “Proper Orthogonal Decomposition
Extensions for Parametric Applications in Transonic Aerodynamics”. AIAA.
10. Campana E. F., Peri D., Tahara Y., Stern F. (2004). “Comparison and Validation of
CFD Based Local Optimization Methods for Surface Combatant Bow”, 25th
Symposium on Naval Hydrodynamics, St John’s, Newfoundland and Labrador,
Canada.
11. Cebeci, T. & Smith, A.M.O., (1974). “Analysis of turbulent boundary layers”, Series
in Appl. Math. & Mech., Vol. XV, Academic Press.
12. CFD Workshop, proceedings (2005). Editor T. Hino, National Maritime Research
Institute, Tokyo.
13. Chatterjee (2000). “An introduction to the proper orthogonal decomposition”, Current
o
Science, vol. 78, n 7, 10 April 2000.
14. Clarich, Rigoni and Poloni (2004). “A new Algorithm based on Game Theory for
Robust and Fast Multi-Objective Optimisation”.
15. Clarke D., Gedling P. and Hine G. (1982). "The application of Manoeuvring Criteria
in Hull Design Using Linear Theory". Transactions RINA, Vol. 124.
16. Cressie (1993). “Statistics for Spatial Data”, Revised Edition. Wiley, New York.
17. Cura-Hochbaum A. (2006). “Virtual PMM Tests for Manoeuvring Prediction”, 26th
Symposium on Naval Hydrodynamics, Roma, Italy.
18. Dacles-Mariani J, Zilliac GG, Chow JS, Bradshaw P (1995).
“Numerical/experimental study of a wing tip vortex in the near field”. AIAA J
33:1561–1568.

19. Das I. and Dennis J.E. (1996). “Normal-boundary-intersection: A new method for
generating the pareto surface in nonlinear multicriteria optimization problems”, July
1996, SIAM J. Optimization 8(1998), pp 631-657.
20. Dawson, C.W. (1977). “A practical computer method for solving ship-wave
problems”, Proc. Of the 2nd Int. Conf. On Numerical Ship Hydrodynamics, Berkley,
California.
21. Delannoy, Y. and Kueny, J.L. (1990). “Two phase flow approach in unsteady
cavitation modeling”, Cavitation and Multiphase Flow Forum, pages 153-158.
22. Désidéri J.A. and Bélahcène F. (2003). “Paramétrisation de Bézier adaptative pour
l'optimisation de forme en aérodynamique”, rapport de recherche n°4943 INRIA
Sophia Antipolis.
23. Di Mascio A., Broglia R., Muscari R. (2004). “Unsteady RANS Simulation of a
Manoeuvring Ship Hull”, 25th Symposium on Naval Hydrodynamics, St John’s,
Newfoundland and Labrador, Canada.
24. Dommermuth D., Gharib M., Huang H., Innis G., Maheo P., Novikov E., Talcott J. &
Wyatt D. (1997). “Turbulent Free-Surface Flows: A Comparison between Numerical
Simulations and Experimental Measurements”, 21st Symposium on Naval
Hydrodynamics.
25. Eça and Hoekstra (2004). “On the Grid Sensitivity of the Wall Boundary Condition of
the k- Turbulence Model”, J. Fluids Eng. Volume 126, Issue 6, 900.
26. ERCOFTAC Best Practice Guidelines.
27. Ferrant, P., Le Touzé, D. (2002). “Fully-non-linear spectral/BEM solution for
irregular wave interactions with a 3D body”, OMAE Conference, Oslo.
28. Ferziger, J.H. & Peric, M. (1997). “Computational methods for fluid dynamics”,
Springer, Berlin.
29. Fisher, E.M. & Rhodes, N. (1996). “Uncertainty in computational fluid dynamics”,
Proc. Mech. Eng., vol. 210, Part C: Journal of Mech. Eng. Sci., pp. 91-94.
30. Fletcher, C.A.J. (1991). “Computational techniques for fluid dynamics, vol. I & II”,
Springer, Berlin.
31. Fletcher R. (1989). “Practical Methods of Optimization”, Second Edition, John Wiley
& Sons, 1989.
32. Foeth, E.J. (2008). “The Structure of Three-Dimensional Sheet Cavitation“, PhD
thesis, Delft University of Technology.
33. Foeth, E.J., Van Terwisga, T., Van Doorne, C. (2008). “On the collapse structure of
an attached cavity on a three- dimensional hydrofoil”, Journal of Fluids Engineering,
Vol. 130.
34. Gatski, T. B., and Speziale, C. G. (1993). “On Explicit Algebraic Stress Models for
Complex Turbulent Flows,” Journal of Fluid Mechanics, Vol. 254, pp. 59-78.
35. Giles, M.B., (1990). “Non-reflecting boundary conditions for Euler equation
calculations”, AIAA Journal, vol. 28, no. 12, pp. 2050-2058.
36. Goldstein, S. (1929). “On the vortex theory of screw propellers”, Proc. Roy. Soc.,
London, Series A 123.
37. Grilli S., Gilbert G., Lubin P., Vincent S., Astruc D., Legendre D., Duval M.,
Kimmoun O., Branger H., Devrard D., Fraunnié P. & Abadie S (2004). “Numerical
Modeling and Experiments for Solitary Wave Shoaling and Breaking over a Sloping
Beach”, 14th International Offshore and Polar Engineering Conference, Toulon.

38. Grotjans H., Menter F.R. (1998). ''Wall functions for industrial applications''. In
Proceedings of Computational Fluid Dynamics’98, ECCOMAS, 1(2), Papailiou KD
(ed.). Wiley: Chichester, U.K. pp. 1112–1117.
39. Gui, L., Longo, J., Metcalf, B., Shao, J., Stern, F. (2001). “Forces, Moment and
Wave Pattern for Naval Combatant in Regular Head Waves”, 23rd Symposium on
Naval Hydrodynamics.
40. Guillerm P.E., Alessandrini B (2003). “3D free surface flow using a RANSE/Fourier-
Kochin coupling”, International Journal for Numerical Methods in Fluid, Vol. 43,
Issue 3, pp. 301-308.
41. Harries S., Birk L., Abt C. (2003). “Parametric Hull Design - the FRIENDSHIP-
Modeler”, NAV 2003, Palermo, June 2003.
42. Hellsten (1998). “On the solid-wall boundary condition for ω in the k-ω-type
turbulence models”, Helsinki University of Technology, Report No B-50, Series B.
43. Hess, J.L. & Smith, A.M.O. (1964), “Calculation of non-lifting potential flow around
arbitrary three-dimensional bodies”, Journal of Ship Research, vol. 8, no. 2.
44. Hirsch, C. (1991), “Numerical computation of internal and external flows, vol. I & II”,
Wiley, New York.
45. Hoekstra, M., Vaz, G. (2009). “Cavitation simulations for NACA0015 foil and Delft
twisted wing”, VIRTUE Report D4.2.4, April 2009.
46. Hoekstra M. and Raven H.C. (2003). “A practical approach to constrained
hydrodynamic optimization of ships”. NAV 2003, Palermo, June 2003.
47. Hooft JP (1994). “The cross flow drag on a manoeuvring ship”, Ocean Eng.
21(3):329–342.
48. Huang, T., Liu, H.-L., Groves, N., Forlini, T., Blanton, J. and Gowing, S. (1992).
“Measurements of flows over an axisymmetric body with various appendages in a
wind tunnel: the DARPA SUBOFF experimental program”, 19th Symposium on
Naval Hydrodynamics, Seoul, South Korea, pp. 312–346.
49. Inoue S, Hirano M, Kijima K (1981). "Hydrodynamic Derivatives on Ship
Manoeuvring", International Shipbuilding Progress, Vol. 28, No. 321.
50. Isay, W.-H. (1964). “Propellertheorie”, Springer-Verlag, Berlin / Göttingen /
Heidelberg.
51. ITTC (2008). “Recommended Procedures and Guidelines, Testing and
Extrapolation Methods, Manoeuvrability, Captive Model Test Procedures” (7.5-02-
06-02).
52. ITTC (2008). “Recommended Procedures and Guidelines, Testing and
Extrapolation Methods, Manoeuvrability, Free Running Test Procedures” (7.5-02-
06-03).
53. ITTC (1999). Reports to the 22nd ITTC.
54. Jacquin E., Guillerm P.-E., Alessandrini B. (2007). “From drag resistance to ship
power optimization using CFD”, 9th International Conference in Numerical Ship
Hydrodynamics, Ann Arbor, Michigan.
55. Jacquin E., Guillerm P.-E., Drouet A., Perdon P., Alessandrini B. (2006). “Simulation
of unsteady ship manoeuvring using free-surface RANS solver”, 26th Symposium
on Naval Hydrodynamics, Roma, Italy.
56. Jacquin E., Derbanne Q., Bellevre D., Cordier S., Alessandrini B., Roux Y. (2004).
“Hull Form Optimization Using a Free Surface RANSE Solver”, 25th Symposium on
Naval Hydrodynamics, St John’s, Newfoundland and Labrador, Canada.

57. Jacquin E., Alessandrini B., Bellèvre D., Cordier S. (2003). “New hull design method
for competitive sailing yacht”. 9èmes Journées de l'Hyrodynamique, Poitiers,
France.
58. Kalyanmoy Deb (2000). “A fast and elitist multi-objective genetic algorithm: NSGA II,
KanGAL”. Report n°2000001, 2000.
59. Kato, M. & Launder, B.E. (1993). “Three-dimensional modelling and heat-loss
effects on turbulent flow in a nominally two-dimensional cavity”, Int. J. Heat and
Fluid Flow, vol. 16, pp. 171-177.
60. Koop, A.H. (2008). “Numerical simulation of unsteady three-dimensional sheet
cavitation”, PhD thesis, University of Twente, Netherlands.
61. A. Kubota, H. Kato, and H. Yamaguchi (1992). “A New Modelling of Cavitating
Flows; A Numerical Study of Unsteady Cavitation on a Hydrofoil Section”, Journal of
Fluid Mechanics, 240:59–96.
62. Kume, K., Hasegawa, J., Tsukada, Y. et al. (2006). “Measurements of
hydrodynamic forces, surface pressure, and wake for obliquely towed tanker model
and uncertainty analysis for CFD validation”. Journal Marine Science and
Technology, 11:65–75.
63. Kunz, R.F., Boger, D.A., Stinebring, D.R, Chyczewski, S., Lindau, J.W., Gibeling,
H.J., Venkateswaran, S. and Govindan, T.R. (2000). “A Preconditioned Navier-
Stokes Method for Two-Phase Flows with Application to Cavitation Prediction”,
Computers and Fluids, 29:849–875.
64. Kraichnan R. (1970). “Diffusion by a Random Velocity Field”. Physics of Fluids,
11:21-31.
65. Launder, B.E. (1984). “Second-moment closure: methodology and practice”,
Chapter in Turbulence Models and their Applications, vol. 2, Collection de la
Direction des Etudes et Recherches d’Electricité de France, Eyrolles, Paris.
66. Launder, B.E. & Spalding, D.B. (1974). “The numerical computation of turbulent
flow”, Comp. Meth. In Appl. Mech. And Engng., vol. 3, pp. 269-289.
67. Launder, B.E. & Spalding, D.B. (1972). “Mathematical models of turbulence”,
Academic Press.
68. Lecocq Franck (2000). “Distribution spatiale et temporelle des coûts de politiques
publiques sous incertitudes: théorie et pratique dans le cas de l’effet de serre”.
Chapitre 2: Figures logiques de la décision sous incertitude [Decision under
uncertainty: a review of approaches], Thèse ENGREF, 2000.
69. Longo J.F. (1996). “Effects of Yaw on Model-Scale Ship Flows”, PhD thesis,
University of Iowa, May 1996.
70. Luecke, T. and Streckwall, H. (2009). “Cavitation Research on a Very Large Semi
Spade Rudder”. Paper to be published at 1st International Symposium on Marine
Propulsors, Trondheim, Norway, 22nd-24th June 2009.
71. Luquet R., Gentaz L., Ferrant P., Alessandrini B. (2004). “Viscous Flow simulation
past a ship in waves using the SWENSE approach, 25th Symposium on Naval
Hydrodynamics, St John’s, Newfoundland and Labrador, Canada.
72. MacKay D.J.C. (2003). “Information Theory, Inference, and Learning Algorithms”.
Copyright Cambridge University Press 2003.
73. Mathey F., Cokljat D., Bertoglio J. P., and Sergent E. (2003). “Specification of LES
Inlet Boundary Condition Using Vortex Method”. In K. Hanjali, Y. Nagano, and M.
Tummers, editors, 4th International Symposium on Turbulence, Heat and Mass
Transfer, Antalya, Turkey. Begell House, Inc.

74. Menter, F.R. (1997). “Eddy viscosity transport equations and their relation to the k–
model”. J Fluid Eng Vol. 119, pp.876–884.
75. Menter, F.R. (1996). “A comparison of some recent eddy-viscosity turbulence
models”, Transactions of the SNAME, vol. 118, pp. 514-519
76. Menter, F.R. (1994a). “Two-equation eddy-viscosity turbulence models for
engineering applications”, AIAA Journal, vol. 32, No. 8, pp. 1598-1605.
77. Menter, F.R. (1994b). “Eddy viscosity transport equations and their relation to the k-
 model”, NASA-TM-108854.
78. Menter, F.R. (1993). “Zonal two equation k- turbulence models for aerodynamic
flows”, AIAA Paper 93-2906.
79. Merkle, C.L., Feng, J.Z., and Buelow, P.E.O. (1998). “Computational Modeling of
the Dynamics of Sheet Cavitation”, In Proceedings of 3rd International Symposium
on Cavitation, Grenoble, volume 2, pages 307–311.
80. modeFRONTIER. Documentation manual for use [accessed on 19 February 2009].
81. Muzaferija S., Peric M., Sames P., and Schellin T. (1998). “A Two-Fluid Navier-
Stokes Solver to Simulate Water Entry”. In Proc 22nd Symposium on Naval
Hydrodynamics, pages 277-289, Washington, DC.
82. Musker, A.J. (1988). “A panel method for predicting ship wave resistance”, 17th
Symp. Of Naval Hydrodynamics, Den Haag.
83. Patankar, S.V. (1980). “Numerical heat transfer and fluid flow”, McGraw-Hill, New
York.
84. Patel, V.C., Rodi, W. & Scheuerer, G. (1985). “Turbulence models for near-wall and
low Reynolds number flows: a review”, AIAA Journal, vol. 23, no. 9, pp. 1308-1318.
85. Poles S. (2004). “Constraint Satisfaction Problem using Record-to-Record Travel”,
Master Thesis in Modelling and Simulation of Complex Realities. The Abdus Salam
International Centre for Theoretical Physics, September 21, 2004.
86. Poloni C. (2003). “Genetic algorithms – Basics, Multi-criteria optimization and
constraints handling, Optimistic, Optimization in Marine Design”, Berlin, 2003.
87. Reboud J.L., Stutz B. and Coutier-Delgosha O. (1998). Two-phase flow structure of
cavitation: Experiment and Modelling of Unsteady effects, 3rd Int. Symp. On
Cavitation, Grenoble, France.
88. Richardson, L. F. (1922). “Weather Prediction”, Cambridge.
89. Rienecker, M.M., Fenton, J.D., Fourier A (1981). “Approximation method for steady
water waves”, Journal of Fluid Mechanics, Vol. 104, pp. 119-137.
90. Rigoni E. (2003). “MOSA, Multi Objective Simulated Annealing”, Technical Report,
May 13, 2003.
91. Rizzi, A. & Voss, J. (1998). “Towards establishing credibility in computational fluid
dynamics simulations”, AIAA Journal, vol. 36, no. 5, pp. 668-675.
92. Roache, P.J. (1998). “Verification and validation in computational science and
engineering”, Hermosa Publishers, Alberquerque.
93. Rodi (1976). “A New Algebraic Relation for Calculating the Reynolds Stresses”,
Journal of Applied Mathematics and Mechanics ZAMM, Vol. 56, T219-T221.
94. Rodi, W. (1981). “Progress in turbulence modelling for incompressible flows”, AIAA
Paper 81-45, St Louis.

95. Roddy, R.F. (1990). “Investigation of the stability and control characteristics of
several configurations of the DARPA SUBOFF model (DTRC Model 5470) from
captive-model experiments”, Report No. DTRC/SHD-1298-08, September 1990.
96. Salvatore, F., Streckwall, H., Van Terwisga, T. (2009). ‘Propeller Cavitation
Modelling by CFD - Results from the VIRTUE 2008 Rome Workshop’ Proceedings
of the SMP 2009 Symposium, Trondheim, Norway.
97. Salvatore, F., Streckwall, H. and Terwisga, T.v. (2009). “Results from the Rome
2008 Workshop on Propeller Cavitation Modelling”, Paper to be published at 1st
International Symposium on Marine Propulsors, Trondheim, Norway, 22nd-24th
June 2009.
98. Sasaki D. (2005). “ARMOGA An efficient Multi-Objective Genetic Algorithm”,
Technical Report, January 17, 2005.
99. Sauer, J (2000). “Instationär kavitierende Strömungen-Ein neues Modell basierend
auf Front Captering (VOF) und Blasendynamik”, Doctor Thesis, Univ. of Karlsruhe,
Germany.
100. Sauer, J. and Schnerr, G.H. (2000). “Unsteady Cavitating Flow - A New Cavitation
Model Based on a Modified Front Capturing Method and Bubble Dynamics”, In
Fluids Engineering Summer Conference, Proceedings of FEDSM’00.
101. Saurel, R., Cocchi, J.P. and Butler, P.B. (1999). “A Numerical Study of Cavitation in
the Wake of a Hypervelocity Underwater Profile”; Journal of Propulsion and power,
15(4):513–522.
102. Schittkowski K. (2006). “NLPQLP: A Fortran Implementation of a Sequential
Quadratic Programming Algorithm with Distributed and Non-Monotone Line
Search”, User’s Guide, Version 2.2. Department of Computer Science, University of
Bayreuth.
103. Schmidt, S.J., Sezal, I.H., Schnerr, G.H. and Thalhamer, M. (2008). “Numerical
analysis of shock dynamics for detection of erosion sensitive areas in complex 3-D
flows”, Proceedings of the WIMRC Forum 2008, Warwick University.
104. Schmidt, S.J., Sezal, I.H. and Schnerr, G.H. (2006). “Compressible Simulation of
High-Speed Hydrodynamics with Phase Change”, ECCOMAS CFD.
105. Sculptor. Available on line on: http://www.optimalsolutions.us [accessed on 19
February 2009].
106. Senocak, I. and Shyy, W. (2002). “Evaluation of Cavitation Models for Navier-
Stokes Computations”, Proceedings of ASME FEDSM 02, FEDSM2002-31011.
107. Smirnov R., Shi S., and Celik I. (2001). “Random Flow Generation Technique for
Large Eddy Simulations and Particle-Dynamics Modelling”, Journal of Fluids
Engineering, 123:359-371.
108. Spalart, P.R. & Allmaras, S.R. (1992). “A one-equation turbulence model for
aerodynamic flows”, AIAA Paper, 92-0439.
109. Speziale, C.G. (1987). “A Second-order closure model for rotating and turbulent
flows”, Q. Appl. Math., vol. 45, pp. 69-71.
110. Streckwall, H. and. Salvatore, F. (2008). “Results from the Wageningen 2007
Workshop on Propeller Open Water Calculations Including Cavitation”, RINA
MARINE CFD 2008 Conference, Southampton, UK, 26 – 27 March 2008.
111. Streckwall, H., Park, J.-J. and Jang, Y.-H. (2007). “A numerical Approach to Predict
the Cavitation Performance of Semi-balanced Rudders”, PRADS Symposium 2007,
Houston, Texas, Sept. 30 -Oct. 5. 2007.

112. Tahara Y., Paterson E., Stern F., Himeno Y., (2000). “Flow- and Wave-Field
Optimisation of Surface Combatants Using CFD-Based Optimization Methods”, 23th
Symposium on Naval Hydrodynamics, Val de Reuil, France.
113. Tennekes, H. & Lumley, J.L. (1972). “A first course in turbulence”, MIT Press,
Cambridge. Latest edition 1994.
114. Toxopeus SL, Lee SW (2008). "Comparison of manoeuvring simulation programs
for SIMMAN test cases", SIMMAN 2008 Workshop on Verification and Validation of
Ship Manoeuvring Simulation Methods, pp E56–61.
115. Toxopeus S.L. (2006). "Validation of slender-body method for prediction of linear
manoeuvring coefficients using experiments and viscous-flow calculations".
ICHD2006 7th International Conference on Hydrodynamics, pp 589–598.
116. Ubbink (1997). “Numerical Prediction of Two Fluid Systems with Sharp Interfaces”,
PhD thesis, Imperial College of Science, Technology and Medicine, London,
England.
117. Vaz, G.N.V.B. (2005). “Modelling of sheet cavitation on hydrofoils and marine
propellers using Boundary Element Methods”, PhD thesis, University of Lisbon,
Lisbon.
118. Verkuyl J.-B., Raven H. (2003). “Joint EFFORT for Validation of Full-Scale Viscous-
Flow Predictions”, the Naval Architect.
119. Vogt, M. (2007). “Preliminary results of HTC manoeuvring tests for Virtue Project”,
HSVA.
120. Wikipedia on Smoothed Particle Hydrodynamics (2009). Available on line on:
http://en.wikipedia.org/wiki/Smoothed_particle_hydrodynamics [accessed on 28
January 2009]
121. Wilcox, D. C. (1988). “Multiscale model for turbulent flows”, AIAA Journal, Vol. 26,
pp. 1211-1320.
122. Wilcox, D.C. (1998). “Turbulence modelling for CFD”, DCW Industries, Inc.
123. Wilson R., Stern F. (2002). “Unsteady RANS Simulation of a Surface Combatant
with Roll Motion”, 24th Symposium on Naval Hydrodynamics, Fukuoka, Japan.
124. Wolfstein, M.W. (1969). “The velocity and temperature distribution in a one-
dimensional flow with turbulence augmentation and pressure gradient”, Int. J. Heat
and Mass Transfer, vol. 12, pp. 301-312.
125. Yoshimura, Y., (2002). “Problems in measuring and analyzing hydrodynamic
forces”, Bulletin of Society of Naval Architects of Japan 869: 32-35.
126. Youngs (1982). “Time-Dependent Multi-Material Flow with Large Fluid Distortion”,
In K. W. Morton and M. J. Baines, editors, Numerical Methods for Fluid Dynamics.
Academic Press.
View publication stats

VIRTUE BPG Integrated Final PDF

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

VIRTUE BPG Integrated Final PDF

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author proﬁles for this publication at: https://www.researchgate.

Best Practice Guidelines for the application of Computational Fluid

Technical Report · April 2009

Richard Marcer Christian Berhault

SEE PROFILE SEE PROFILE

Luís Eça Leif Broberg

corrosion in shipbuilding View project

The user has requested enhancement of the downloaded ﬁle.

VIRTUE – The Virtual Tank Utility in Europe

Ir. Benoît Post from Atkins.

Issue Date: 28 April 2009 Page 1

Issue Date: 28 April 2009 Page 2

2. MODELLING THEORY FOR MARINE APPLICATIONS ....................... 9

Issue Date: 28 April 2009 Page 3

3.2.2. Meshing ................................................................................................................. 47

4. APPLICATION 1: NUMERICAL TOWING TANK ................................ 62

5. APPLICATION 2: NUMERICAL SEA KEEPING TANK ..................... 102

Issue Date: 28 April 2009 Page 4

6. APPLICATION 3: NUMERICAL MANOEUVRING TANK .................. 144

Issue Date: 28 April 2009 Page 5

7. APPLICATION 4: NUMERICAL CAVITATION TANK ....................... 184

8. CONCLUSION .................................................................................... 224

9. REFERENCES ................................................................................... 225

Issue Date: 28 April 2009 Page 6

Issue Date: 28 April 2009 Page 7

1.3. Notes for readers

Issue Date: 28 April 2009 Page 8

2. Modelling theory for Marine applications

2.1. Fluid equations of motion

2.1.1. General Fluid Dynamic Equations

Issue Date: 28 April 2009 Page 9

and the other, relating static enthalpy to temperature and pressure:

2.1.2. Incompressible and Isothermal assumption

The momentum equations become:

Issue Date: 28 April 2009 Page 10

2.2. Inviscid and irrotational hypotheses

2.2.1. Potential Flow Equations

The continuity equation becomes a Laplace equation:

As the Laplace equation is linear it is possible to combine elementary solutions, such as

Issue Date: 28 April 2009 Page 11

2.2.2. Steady ship flow application

Where u   is used to derive the flow velocity components

The following boundary conditions are formulated throughout the domain:

The steady dynamic condition at z =  is:

Issue Date: 28 April 2009 Page 12

     B GB .dS     F GF .dS    W GW .dS (23)

(U   )n  0 (on the hull) (24)

Issue Date: 28 April 2009 Page 13

2.2.3. Free surface modelling

Issue Date: 28 April 2009 Page 14

problems, it is therefore common to assume a single frequency harmonic incident potential,

2.2.4. Integration of viscous effects

2.2.4.1. Empirical and semi-empirical methods

2.2.4.2. Solution of Navier Stokes equations

2.3. Turbulence Modelling

Issue Date: 28 April 2009 Page 15

Figure 2-1: Turbulent Energy Cascade (Richardson, 1922)

Issue Date: 28 April 2009 Page 16

U ( x, t ) is the unsteady disturbance quantities in the flow, such that U   0

 (u 2 ) (uv) (uw)  dP   u    u 

Issue Date: 28 April 2009 Page 17

2.3.1.1. Eddy Viscosity model

2.3.1.1.1 Algebraic (or zero-equation) models