RINA

INTERNATIONAL CONFERENCE

MARINE CFD 2005

4th INTERNATIONAL CONFERENCE
ON MARINE HYDRODYNAMICS

30 – 31 March 2005

© 2005: The Royal Institution of Naval Architects

The Institution is not, as a body, responsible for the

opinions expressed by the individual authors or
speakers

THE ROYAL INSTITUTION OF NAVAL

ARCHITECTS
10 Upper Belgrave Street
London SW1X 8BQ

Telephone: 020 7235 4622

Fax: 020 7259 5912

ISBN No: 1-905040-10-5

Marine CFD 2005, London, UK

CONTENTS
A Step Towards the Numerical Simulation of Viscous Flows around Ships at
Full Scale - Recent Achievements within the European Union Project
EFFORT
Michel Visonneau, Charge de Recherche CNRS – HDR, France

Blade Shaping for Off-Design Performance: Cavitation and Efficiency in

Two-Dimensional Cascades
James J Dreyer, Penn State University, USA

Numerical Simulation of Free Surface Wave Induced Separation

S H Sadathosseini, S M Mousaviraad, and M H Sadr, Petropars Ltd., Iran

BASIN – Development of a Practical Boundary Element Code for

Hydrodynamic Analysis
Neil Southall and Brian Corlett, Burness Corlett – Three Quays (IOM) Ltd., UK

VOF-Dynamic Mesh Simulations of Unsteady Ship Hydrodynamics

M Visone, Blue Group, Italy
C Falletta P.L. Ausonio, Ship-Yacht Designers & Consultants, Italy
P Bertetti and R Gandolfi, AZIMUT, Italy
D Paterna and R Savino DISIS Univ. of Naples “Federico II”, Italy

Second-Order Wave Forces and Free-Surface Elevation Around a Moored

Ship in Steep Uni-Directional and Spread Waves
Jun Zang, K Wang, R Eatock Taylor and Paul Taylor, University of Oxford, UK

RANS Application on Ship Manoeuvring Motion

Q Gao, V Shigunov and D Vassalos, Strathclyde University, UK

Comparison Between RANSE Calculations and Panel Method Results for the
Hydrodynamic Analysis of Marine Propellers
Chiara Pittulaga and Paolo Becchi, CETENA, Italy

Consideration on Deviations in Torque Prediction for Propellers and

Waterjets with RANS Codes
Norbert Bulten and Iulia Oprea, Wärtsilä, the Netherlands

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

Sven-Brian Müller and Moustafa Abdel-Maksoud, University Duisburg-Essen,
Germany

Predictions of the Thrust and Torque Performance for Two Propeller Blades
Using Computational Fluid Dynamics
Karl Randle and Peter Bull, QinetiQ Haslar, UK

Behaviour of Ship Funnel Exhaust in the Wake of a Bluff Body

P R Kulkarni, S N Singh and V Seshadri, IIT Delhi, India

© 2005: The Royal Institution of Naval Architects

Marine CFD 2005, London, UK

Improving Carrier Operation through the Application of CFD to the CVF *

Design Process
Richard Underhill and Elizabeth Morrison, Frazer-Nash Consultancy, UK

The Effect of Ship Shape and Anemometer Location on Wind Speed

Measurements Obtained from Ships.
Ben I Moat and Margaret J Yelland and Robin W Pascal, Southampton
Oceanography Centre, UK
Anthony F Molland, University of Southampton, UK

Authors’ Contact Details

* Paper withdrawn at request of MOD

© 2005: The Royal Institution of Naval Architects

Marine CFD 2005, Southampton, UK

A STEP TOWARDS THE NUMERICAL SIMULATION OF VISCOUS FLOWS AROUND SHIPS

AT FULL SCALE - RECENT ACHIEVEMENTS WITHIN THE EUROPEAN UNION PROJECT
EFFORT

M Visonneau, Ecole Centrale de Nantes, France

SUMMARY
The EFFORT (European Full-scale FlOw Research and Technology) project is funded by the European Framework 5 program
and aims at the refinement and validation of CFD prediction methods for the viscous flow around a ship hull at full scale, and
their introduction into practical ship and propeller design.
The emphasis is on predicting the full-scale viscous flow field around a ship including the evaluation of the free surface, the
wake field, the hull/propeller interaction, the resistance and the power. RANS computations do offer that possibility, and such
full-scale viscous-flow computations have started to be used in practical ship design; but how accurate these predictions are is
not really known. Validation of full-scale ship viscous flow predictions has generally been insufficient so far, mainly due to
the virtual absence or difficult accessibility of suitable full-scale experimental flow field data such as wake-field data. This is
the fundamental point that EFFORT aims to address.
The paper presents studies carried out in one of the major work-packages which had three main objectives: (i) to develop
and implement the appropriate physical modelling for full scale flows, (ii) to perform numerical studies of full scale flows
around a real ship to check the robustness and the accuracy of the simulation tools in full scale flow conditions, (iii) to issue
recommendations to prepare the simulation tools to be used for CFD validation at ship and model scale, (iv) to develop and
implement the appropriate physical modelling for full scale flows.

1. INTRODUCTION the project manager, HSVA and CTO), five universities

Hydrodynamic aspects play an important role in the quality
of a ship. Dominant criteria for the hull form design of many
ships are the resistance and powering performance. In addi-
tion, the occurence of noise and vibrations, important for the
comfort level for crew and passengers, often have an hydro-
dynamic cause stemming from the operation of the propeller
in the flow field behind the ship hull. Critical information to
evaluate an industrial configuration is the knowledge of the
flow in the aft-part of a ship and in its near-wake in situations
as close as possible to the reality. Theoretically, Computa-
tional Fluid Dynamics (CFD) tools can provide a solution to
these questions and the use of CFD tools for simulating the
flow around a scaled ship model with or without propeller is
more and more frequent. However, such model scale com-
putations are subject to the same drawbacks as those intro-
duced by the scaling procedures. To make a real step for-
ward in prediction accuracy, it is mandatory to use accurate
and reliable viscous flow solvers for the actual full-scale ship
in the design process.
Viscous flow computations for the full-scale ship is inhibited
by:
• Uncertainty about the proper physical modelling for
full-scale ship flows,
• difficult accessibility of full-scale experimental data to
validate the computations,
(ECN/CNRS, NTUA, HUT, CTH and the Maritime Uni-
versity of Szczecin), four industrial partners (IHC Holland
N.V., Rolls-Royce Kamewa, Kvaerner Masa Yards, Van Vo-
orden Gieterij B.V. and Bassin d’Essai des Carènes) and one
classification society (Lloyd’s Register), are:
• to extend in a collaborative way, the existing CFD
codes to predict the flow around ship hulls at full scale,
• to carry out a full-scale measurement campaign in
which extensive LDV flow measurements will be done
at sea on board of two vessels,
• to collect the most complete possible set of full-scale
flows fo CFD validation and to establish the level of
accuracy and improve the modelling,
• to apply the CFD tools to design studies proposed by
the industrial participants which support the project.
More up-to-date information about the progress achieved in
this european project may be found in the EFFORT web
site hosted by MARIN (http://www.marin.nl). This paper
is more specifically devoted to the CFD development work
performed in Workpackage 3 (WP3) which was focused on
some particular aspects of RANS solvers that play an impor-
tant role in the validation for full scale, or that require further
study before they can be applied reliably for full scale.

• numerical difficulties associated with the use of very
large aspect ratio grids required to capture the viscous
effect at full-scale.
Therefore, the aims of the european project EFFORT
(European Full-scale FlOw Research and Technology)
which groups together three technical centers (MARIN
• Free-surface effects on the viscous flow will play a role,
albeit perhaps limited; a specific task has been included
to study the possibility to incorporate this effect for full
scale since this had not yet been demonstrated.
• Turbulence modelling plays a decisive role in the qual-
ity of the wake field prediction at model scale, and may

2005:
c Royal Institution of Naval Architects
 Marine CFD 2005, Southampton, UK

be expected to do so at full scale as well. Since the pos- following references: [1], [2] for the code developped by
sibility to apply all relevant turbulence models at full NTUA, [3] for COMET and [4] for the studies carried out
scale Reynolds numbers had not been established, this in HSVA, [5], [6] for the code used by HUT, [7], [8] for the
is addressed in another task. approach followed by MARIN, [9] for CTH and [10] for the
code developped by CNRS.
• A representation of the propeller effect will not be par-
ticularly different at full scale than at model scale, but
is needed for making a comparison of predictions with 3. FREE-SURFACE FULL-SCALE COMPUTATIONS
the total-wake measurements to be carried out. There- 3.1 DESCRIPTION OF THE TEST CASE
fore, a small development task addresses this subject
but, for the sake of brevity, the results related to this The research vessel selected for this study is the Nawigator
task are not described in this paper. XXI operated by the Maritime University of Szczecin. The
vessel is designed for restricted navigation on the Baltic and
North Seas within 200NM from the nearest shelter and the
2. COMPUTATIONAL APPROACHES
main destination of the ship is to train students for future
2.1 THE DIFFERENT FLOW SOLVERS work as deck officers, engineers and ship electricians. Fig-
ure 1 shows the real vessel.
The coordinator of the workpackage was CNRS and the par-
ticipants were HUT, CTH, NTUA, MARIN and HSVA. The
list of these CFD groups selected according to relevant CFD
capabilities and earlier work with CFD tools is provided
with Tab. 1 and Tab. 2. Notations NS stands for Navier-
Stokes, FS for free-surface and NLP for Non-Linear Poten-
tial.

Table 1: Partners in WP3

Partner name Abbreviation Country

Chalmers University CHALMERS S
of Technology
Centre National CNRS F
Recherche Scientifique Figure 1: The training and research ship Nawigator XXI
Hamburgische Shiffbau HSVA D
This vessel is characterised with the following parameters:
Versuchs Anstalt
Helsinki University HUT FIN • Length L = 55.155m, Draft D = 3.15m
of Technology • Wetted Surface: SDW L = 665.5 m2
Maritime Research MARIN NL
• Speed 12 knots: Re = 340 106 F r = 0.265
Institute Netherlands
National Technical NTUA G Numerical grids are generated from available CAO database
(see Fig. 2 for a global view of the hull).
University of Athens

Z
Y

Table 2: Contributors and CFD tools X

Organization Code Notes

CHALMERS CHAPMAN NS code (no FS)
CNRS ISIS NS code (FS capturing)
HSVA COMETa NS code (FS capturing)
HUT FINFLO NS code (FS fitting)
MARIN RAPID NLP code
MARIN PARNASSOS NS code (fixed FS) Figure 2: Nawigator XXI: 3D view (Free-surface capturing)
NTUA PARALOS NS code (FS fitting)
a Commercial code
3.2 THE VARIOUS COMPUTATIONAL STRATEGIES
The CFD partners participating in this collaborative work
More detailed information about the numerical strategies represent the whole spectrum of the computational strate-
implemented in the respective solvers may be found in the gies for computing viscous free-surface flows. MARIN uses

2005:
c Royal Institution of Naval Architects
Marine CFD 2005, Southampton, UK

a composite approach which is based on the computation of

the free-surface by a non-linear potential theory and, then,
uses this prescribed free-surface as a fixed boundary for the
viscous flow computations. HUT and NTUA compute the
viscous free-surface flow with a free-surface fitting algo-
rithm, which means that only the flow of water is computed
with adequate physical boundary conditions on an unknown
free-surface which results from the computations. An au-
tomatic regridding strategy is then needed to update the
shape of the free-surface boundary during the computations.
HSVA and CNRS have chosen to use a free-surface captur-
ing methodology which computes the flow in air and wa-
ter. No regridding is therefore necessary and wave-breaking
may be simulated. The accuracy of the computations is
strongly dependent on the specific compressive discretisa-
tion schemes which are used to discretize the concentration
transport equation which characterizes the presence of air or
Figure 4: Free-surface and wall streamlines at full scale
water.
Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

3.3 GLOBAL VIEWS OF THE FREE-SURFACE 40

Figures 3 and 4 show a global view of the free-surface and 30

wall streamlines computed by CNRS with a free-surface

capturing methodology for model and full scales, respec- Y
20

tively. One can notice slight differences on the positionning

10
of the main convergence line which refer to the different in-
tensity of the longitudinal vortex. The stern wave height is 0
-60 -40 -20 0 20
X
also larger at full scale than at model scale, a fact on which HSVA : FS : K-w SST : 12 knots

we will come back in the remainder of this article.

Figure 5: Wave elevations: HSVA:12 knots (FS capturing)

Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40

30
Y

20

10

0
-60 -40 -20 0 20
CNRS : FS : K-w SST : 12 knots X

Figure 6: Wave elevations: CNRS:12 knots (FS capturing)

Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40

30
Figure 3: Free-surface and wall streamlines at model scale
Y

20

Figures 5 and 6 show a global view of the free-surface ele-

10
vations around the Nawigator at 12 knots obtained by HSVA
and CNRS, both teams using the free-surface capturing strat- 0
-60 -40 -20 0 20
egy. These computations are performed on identical grids HUT : FS : K-w SST : 12 knots X

comprising roughly 3 million points, and, despite different

discretisation schemes and codes, the results appear quite
similar.
The second set of results (fig.7 and 8), obtained by NTUA
and HUT, is representative of the free-surface fitting strat-
egy. One may notice the attenuation of the waves in the far
field, which is likely due to the lack of grid points in this
region and is not related to the free-surface fitting methodol-
ogy. Lastly, figure 9 shows the free-surface elevation com-
Figure 7: Wave elevations: HUT:12 knots (FS fitting)
puted by the non-linear potential code RAPID developed by
MARIN. One may notice the capability of this methodology
to capture accurately waves in the far-field although these
specific computations are affected by some spurious short
waves occurring in the inner sector of the wave pattern.

2005:
c Royal Institution of Naval Architects
 Marine CFD 2005, Southampton, UK

0.03
Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
0.025

40 Model Scale
0.02 Full Scale

0.015
30

Z/Lpp
0.01
Y

20 0.005

0
10
-0.005

-0.01
0 -1 -0.5 0 0.5
-60 -40 -20 0 20 X/Lpp
NTUA : FS : K-ε : 12 knots X 10 knots : Wave profile

Figure 8: Wave elevations: NTUA:12 knots (FS fitting) Figure 11: Wave elevation on the waterline and in the plane
of symmetry: model scale vs full scale
Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40

in many computations that the longitudinal vortices which

30
strongly influence the stern flow are only accurately simu-
lated by the most sophisticated turbulence closures which do
Y

20
not rely on an isotropic eddy-viscosity closure (like EASM
10
or Reynolds-Stress Transport Models). Details on the tur-
bulence closures may be found for instance in [11], [12]
0
-60 -40 -20
X
0 20 or [13]. In that case, the modelling error is clearly domi-
MARIN : FS : K-w SST : 12 knots
nating the numerical uncertainty. However, during the last
Gothenburg 2000 workshop [14], several contributors per-
Figure 9: Wave elevations: MARIN:12 knots (fixed FS)
formed full-scale computations around the KVLCC2 tanker
hull and found that the full-scale flow was less vortical than
the model scale, and consequently less dependent on the tur-
3.4 LONGITUDINAL WAVE PROFILES CUTS
bulence modelling. However, it was not possible to confirm
Figure 10 shows the various wave elevations obtained by these computational findings without any detailed full-scale
each CFD partner at Y = 6m. Surprisingly, one can notice stern flow measurements. Clearly, the selection of the most
appropriate level of turbulence closures is still an open ques-
tion as long as full-scale ship flows are considered. This is
0.6 CNRS why it has been decided to compare several different tur-
HSVA

0.4
MARIN bulence closures for the full-scale flow around the Nawiga-
HUT
NTUA tor XXI. Although full-scale flow measurements are not yet
0.2
available for this particular ship, it is interesting to check if
the conclusions relative to the lesser influence of turbulence
Z

modelling at full scale are universal or highly dependent on

-0.2
the geometry. In addition to these results, a first compari-
-0.4 son between computations and full-scale flow measurements
-0.6
around the St. Michaelis kindly provided by HSVA, will
-60 -40 -20 0 conclude this brief overview.
12 knots : Wave cut Y=6m X

Figure 10: Wave cuts: Y = 6m: 12 knots 4.2 COMPARISON OF DIFFERENT TURBULENCE
CLOSURES
that there is a relatively large difference between non-linear
Several turbulence models have been employed by different
potential and viscous solutions near the bow which might
partners during this project. All except CNRS use linear
be due to the fact that this bow wave is probably strongly
eddy viscosity models. The K-ω SST model is implemented
breaking. Elsewhere, all the solutions agree reasonably well.
in each code used by every partner except NTUA which
Lastly, figure 11 shows that scale effect on the wave eleva-
uses a k- model. MARIN has also implemented a one
tion is limited to the stern region, where the waves are higher
equation model by Menter. All but HSVA who uses exclu-
at full scale than at model scale. This conclusion was con-
sively wall function approach employ low Reynolds number
firmed by all the contributors.
formulation. CNRS also has wall function approach imple-
mented for two-equation models. Two turbulence models
4. TURBULENCE MODELLING AT FULL SCALE not belonging to the class of linear eddy viscosity model
have been assessed by CNRS. The first one is a quadratic
4.1 THE CONTEXT
explicit algebraic stress model based on the linearized SSG
Although one can expect that the flow around a ship at full pressure-strain rate model called here EASM. The second
scale differs from that at model scale, one can not get any one is a Reynolds stress transport model Rij − ω using the
clear information about the influence of the Reynolds num- IP pressure-strain rate model.
ber on the structure of the flow in the stern region. For
the model-scale ship flows, it has been jointly established One of the challenging task in stern flow computation is the

2005:
c Royal Institution of Naval Architects
Marine CFD 2005, Southampton, UK

correct prediction of the bilge vortex. Detailed experimental

studies and extensive numerical validations performed at
model scale suggest that linear eddy viscosity model under
estimates the intensity of the bilge vortex. The failure of the
linear eddy viscosity model is attributed mainly to the fact
that effect of curvature and anisotropy of the normal stress
which plays a crucial role in the formation of a longitudinal
vortex can not be taken into account by the linear eddy
viscosity model. On the other hand, Reynolds stress model
is capable to give an improved prediction. Improvement Figure 13: Full-scale flow around the Nawigator: X =
over linear eddy viscosity model can also be obtained when 1.6m MARIN vs. CNRS
non-linear model such as explicit algebraic stress model is
used. One of the main objectives of the EFFORT project is
to assess if the same conclusion holds at full scale.

The intensity of the bilge vortex can be easily evaluated by

examining the streamwise velocity contours at transversal
planes in the region near the propeller. The distortion of
the velocity contours due to the bilge vortex results in a
so-called hook-shape contours. It is especially visible in the
propeller plane when the propeller is removed. It also can
be identified at upstream stations when the bilge vortex is
correctly captured. The shape of the wall limiting stream- Figure 14: Full-scale flow around the Nawigator: X =
lines, especially the location of the line of convergence near 1.6m NTUA vs. CNRS
the stern also gives a good indication of the intensity of the
bilge vortex. The line of convergence is pushed downward
when the bilge vortex is strong. formed by CNRS prove that, even at full scale, the influence
of turbulence modelling on the level of longitudinal vortic-
Figures 12, 13 and 14 show a comparison of the isowakes ity is still crucial and comparable to what was observed at
at the propeller disk obtained by HUT, MARIN and NTUA, model scale. The distortion of the streamwise velocity con-
respectively, with those obtained by CNRS. HUT, MARIN tours is clearly visible in the propeller disk section with the
and CNRS have employed the same isotropic eddy-viscosity EASM or the RSTM (not shown here) turbulence closures.
based closure (k − ω SST) and their results are quite compa-
rable although the codes and grids were not identical. NTUA
uses a k − ε turbulence model which results in a less vorti-
cal flow in the stern region. Compared with other turbulence
models mentioned above, k- model reduces the intensity of
the bilge vortex even for the modified k- model by HUT,
which is in agreement with the observation generally ac-
cepted by the community for model scale flows. However, it
is interesting to see that this is also valid at full scale for the
Nawigator XXI.

Figure 15: Isowakes at the propeller disk (X=1.6m) - EASM

model vs. k − ω SST

On the other hand, all computations using linear eddy vis-

Figure 12: Full-scale flow around the Nawigator: X =
1.6m - HUT vs. CNRS
Figure 15 shows the influence of turbulence closures on the
isowakes distribution in the propeller disk and, more specifi-
cally, the influence of the isotropic linear eddy-viscosity hy-
pothesis which is the main hypothesis on which most clas-
sical turbulence closures are built. These comparisons per-
cosity model show no sign of the existence of a longitudi-
nal vortex at the same position. At the propeller plane X4,
none of the computations using linear eddy viscosity model
can give a distortion of the streamwise velocity contours as
strong as that predicted by the RSM and the EASM model.
This observation suggests that both the RSM and the EASM
model predict a stronger bilge vortex compared with linear
eddy viscosity model. However, without measurement data,

2005:
c Royal Institution of Naval Architects

assessment can not be made for the moment concerning the
performance of different models.
4.3 VALIDATION ON THE FULL-SCALE FLOW
AROUND THE ST. MICHAELIS
To conclude this study on the role played by turbulence in
full-scale ship flows, one gives the results of a validation
study performed by CNRS within the EFFORT project on
the St. Michaelis, a ship for which full-scale measurements
were performed by HSVA, several years ago. Figures 16,
17 and 18 show the flow just in front of the propeller disk
(X − XP = 0.228D where D is the diameter of the pro-
peller) computed with the k −ω SST, EASM and RSTM tur-
bulence closures, respectively. Although the experimental
results are not so easy to interprete, one has the feeling that
the real flow is characterized by a strong vortical structure
which causes the well-known hook-shape of the isowakes,
a behaviour which is only accurately simulated by the most
sophisticated turbulence closures EASM and RSTM.
Figure 16: Full-scale flow around the St. Michaelis: X −
XP = 0.228D - k − ω SST model vs experiments
Figure 17: Full-scale flow around the St. Michaelis: X −
XP = 0.228D - EASM model vs experiments
2005:
Marine CFD 2005, Southampton, UK
Figure 18: Full-scale flow around the St. Michaelis: X −
XP = 0.228D - Experiments vs Rij − ω model
5. FORCE COEFFICIENTS
The resistance coefficients are obtained from the following
normalization:
RF RP
CF = 1 2
, CP = 1 2
,
2 ρSDW L U 2 ρSDW L U
RT
CT = 1 2
2 ρSDW L U
Where SDW L is the wetted hull surface at rest equals to
665.5m2 , U is the reference velocity speed. RF and RP
are the resistance forces from friction and pressure part, re-
spectively. RT is the total resistance RF + RP .
Figure 19: Forces coefficients (x1000) - Double model
Figure 19 shows a comparison of the force coefficients ob-
tained by all the participants on the double-body problem at
12 knots with similar turbulence closures. A relatively good
agreement is observed between CNRS, HUT and MARIN,
although one can point out the relatively small value for CP
obtained by HUT. The results obtained by CTH are clearly
far from the mean values since one can observe a deviation
of 8.6% (171%) for CF (CP ), respectively from the mean
values obtained by the three other contributors.
Figure 20 shows the distribution of the forces obtained
for the computations including the free-surface. For some
c Royal Institution of Naval Architects
Marine CFD 2005, Southampton, UK
contributors (MARIN and CTH), the free-surface is fixed
and computed by a non-linear potential approach. Here
MARIN1 stands for computations based on the Menter’s
one equation turbulence model and MARIN2 for the k − ω
SST model, respectively. For HUT and NTUA, a viscous
free-surface fitting procedure is employed while CNRS and
HSVA use a similar free-surface capturing strategy, differ-
ing only by the discretisation schemes These different ap-
proaches are observed in the results. CNRS and HSVA ob-
tained relatively similar results when MARIN and CTH pro-
vide similar CF coefficients but higher values for CP force
coefficient (by 50%). This may be attributed to the higher
amplitudes of the non-linear potential waves. Once again,
one can observe the very large value of CP obtained by
CTH, although they have used a prescribed free-surface pro-
vided by MARIN.
Figure 20: Forces coefficients (x1000): Free-surface
6. CONCLUSIONS
This paper has described the studies carried out in the work-
package 3 ’CFD developments’ of the european project EF-
FORT. During this workpackage, a very impressive work has
been produced by all the partners to implement in their re-
spective CFD codes all the ingredients needed to perform a
realistic full-scale flow. Several interesting and sometimes
unexpected conclusions should be drawn.
• It has been generally observed that full-scale flows are
not dramatically more complicated to compute than
model-scale flows, the influence of the near-wall grid
size on the numerical conditioning remaining accept-
able for most flow solvers.
• Surprisingly, the role of the turbulence modeling has
been reinforced by these computations, which is in
contradiction with the results obtained during the last
Gothenburg 2000 workshop for a totally different hull
geometry. This shows that it is risky to draw very gen-
eral conclusions concerning the influence of the turbu-
lence modelling on the near-wake flow since this role
seems clearly highly dependent on the geometry of
the considered hull. The complex turbulence closures
(Rij − ω) yield a more intense vortical flow than the
classical isotropic eddy-viscosity based closures, as it
was observed at model scale. One must notice that,
if the implementation of a Reynolds-Stress transport
2005:
c Royal Institution of Naval Architects
model is relatively complex, the EASM model may be
an interesting alternative choice.
• A large comparison of different free-surface modelling
strategies has also been completed. Here again, one
must notice a relatively good agreement between the
viscous approaches (free-surface capturing and fitting
strategies) and large differences concerning the ampli-
tudes of the waves when the viscous computations are
compared with the potential approaches (the potential
waves being considerably higher than the viscous ones,
at the bow and the stern of the hull). Without any ex-
perimental information, it is difficult to conclude on the
reliability of the respective approaches but the differ-
ences near the bow are surprising since a strong vis-
cous/inviscid coupling was not expected here, contrary
to the stern region. However, while the influence of
the computational strategies is large on the free-surface
elevation, one may notice that the influence on the near-
wake flow in front of the propeller is relatively small.
• In several occasions, it has been possible to carry out
the computations on the same or very similar grid. This
is an ideal situation to verify the codes and this should
be pointed out since it is quite unusual in such a joint
project. However, an exchange of grids appear difficult
to generalize since some CFD codes are still limited in
terms of topology.
7. ACKNOWLEDGEMENTS
This work is a part of the project EFFORT supported by
the 5th Framework ’GROWTH’ program of the European
Union and by the industry. This financial support is grate-
fully acknowledged. The author would like to thank all the
WP3 partners for their patience and the remarkable work
which was performed by everybody during this collabora-
tive work.
8. AUTHOR’S BIOGRAPHY
Dr. Michel Visonneau is a Research Associate at CNRS and
holds the position of head of the CFD team of the Fluid Me-
chanics Laboratory at Ecole Centrale de Nantes. During the
EFFORT project, he was in charge of the workpackage 3
“CFD developments”.
REFERENCES
[1] G. Tzabiras, A. Dimas, and T. A. Loukakis, “A nu-
merical method for the calculation of incompress-
ible, steady, separated flows around aerofoil,” Interna-
tional Journal for Numerical Methods in Fluids, vol. 6,
pp. 789–809, 1986.
[2] G. Tzabiras, “On the calculation of the viscous flow
around bulbous or u-shaped bows at zero froude num-
ber,” Ship Technology Research, vol. 42, pp. 31–44,
1995.
[3] S. Muzaferija and M. Peric, “Computation of free
surface flows using interface-tracking and interface-
capturing methods.” Non linear wave interaction,
1998. Computational mechanics computations.
 Marine CFD 2005, Southampton, UK

[4] K. Chao, “Numerical propulsion tests,” Ship Technol-

ogy Research, vol. 48-2, 2001.

[5] P. Kaurinkoski and A. Hellsten, “Finflo: the parallel

multi-block flow solver,” Tech. Rep. A-17, Laboratory
of Aerodynamics, Helsinki University of Technology,
1998.

[6] J. Schweighofer, Investigation of Two-Dimensional

Transom Waves Using Inviscid and Viscous Free-
Surface Boundary Conditions at Model and Full-Scale
Ship Reynolds Numbers. PhD thesis, Ship Laboratory,
Helsinki University of Technology, 2003.
[7] H. Raven and B. Starke, “Efficient methods to compute
steady ship viscous flow with free surface,” in Proc.
24th Symp. Naval Hydrodynamics, Fukuoka, Japan,
2002.
[8] H. Raven, “A practical nonlinear method for calculat-
ing ship wavemaking and wave resistance,” in Proc.
19th Symp. Naval Hydrodynamics, Seoul, Korea, 1992.

[9] J. Vierendeels, K. Reimslaugh, and E. Dick, “A multi-

grid semi-implicit line-method for viscous incom-
pressible flow on high aspect ratio grids,” Journal of
Computational Physics, vol. 154, pp. 310–341, 1999.
[10] P. Queutey and M. Visonneau, “Free-surface capturing
ranse simulations for a ship at steady drift,” Ship Tech-
nology Research, vol. 51-3, 2004.

[11] R. Duvigneau and M. Visonneau, “On the role played

by turbulence closures in hull shape optimization at
model and full scale,” J Mar Sci Technol, vol. 8,
pp. 11–25, 2003.
[12] G. Deng and M. Visonneau, “Comparison of explicit
algebraic stress models and second-order turbulence
closures for steady flows around ships,” in Proc. 7th
Int. Conf. on Numerical Ship Hydrodynamics, 1999.

[13] G. Deng, R. Duvigneau, P. Queutey, and M. Vison-

neau, “Assessment of turbulence models for ship flow
at full scale,” in 6th World Congress on Computational
Mechanics, Beijing, China, 2004.

[14] L. Larsson, F. Stern, and V. Bertram, “Summary, con-

clusions and recommandations of the Gothenburg 2000
Workshop,” in A Workshop on Numerical Ship Hy-
drodynamics (L. Larsson, F. Stern, and V. Bertram,
eds.), (Göteborg), Chalmers University of Technology,
September 2000.

2005:
c Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

BLADE SHAPING FOR OFF-DESIGN PERFORMANCE: CAVITATION AND

EFFICIENCY IN TWO-DIMENSIONAL CASCADES
J J Dreyer, The Pennsylvania State University, USA

SUMMARY

In this work, shape optimization of a linear cascade of hydrofoils is demonstrated for the improvement of surface
cavitation inception and section efficiency over a range of operating conditions. Cost functions for incidence-driven
surface cavitation and section efficiency are presented. A continuous adjoint approach is adopted for the efficient
computation of cost function gradients for multiple operating point consideration. The approach is demonstrated on
several cases using a linear cascade of NACA 65410 hydrofoils. Results for single- and multi-point cavitation, efficiency,
and combined cavitation/efficiency are presented. The results establish this approach as a straightforward, robust, and
efficient means of considering off-design performance in the design of blade sections and a promising avenue to pursue
for the development of practical tools for the design of marine thrusters.

NOMENCLATURE L Selection matrix

Mξ Modal matrix, ξ -direction
A Inviscid flux Jacobian matrix, ξ -direction
ND Number of design variables
B Inviscid flux Jacobian matrix, η -direction
NP Number of terms in the composite cost
Bc Blade surface boundary
function
B 1,2
e Domain exit boundaries n Blade surface normal direction
r
Bi Domain inflow boundary n Blade surface unit normal vector
bi ith design variable nx , y Components of the blade surface unit normal
bk Vector of design variables P Periodic boundary; preconditioning matrix
CD Blade section axial force coefficient p Static pressure
C ij Scalar deflator for the ijth mesh point pa Realized surface static pressure
CL Blade section transverse force coefficient pd Target static pressure
Cp Static pressure coefficient p∞ Static pressure in the far-field
pv Vapor pressure
cn Scalar weights for the composite cost function
r R Flow field residual operator
dA Elemental blade surface directed-area vector Re Reference Reynolds number
E Flow field inviscid flux vector, ξ -direction
S x, y Components of the blade surface area
Ev Flow field viscous flux vector, ξ -direction
s Arc length
Ê v Adjoint viscous flux vector, ξ -direction t Pseudo time
F Vector of geometric parameters U ξ -contravariant velocity component
F Flow field inviscid flux vector, η -direction u Velocity component, x-direction
Fv Flow field viscous flux vector, η -direction ui ith component of the velocity vector
F̂v Adjoint viscous flux vector, η -direction V Reference velocity magnitude;
η -contravariant velocity component
Fx x-component of net section force r
Fy y-component of net section force Vw Blade surface slip velocity
~ v Velocity component, y-direction
Gk Smoothed and attenuated cost function W Domain wake-cut boundaries
gradient vector at the kth design cycle w Vector of flow field dependent variables
G kn Cost function gradient vector for the nth cost we Specified exit plane flow variables
function at the kth design cycle, i.e., ∂I n ∂b
k
wi Extrapolated interior flow variables
Gi ith component of the cost function gradient x Cartesian coordinate, independent variable
I Cost function y Cartesian coordinate, independent variable
In nth term of the composite cost function αk Scalar step size for the steepest descent
iˆ Unit vector, x-direction approach at the kth design cycle
J Metric Jacobian δ Perturbation
ĵ Unit vector, y-direction δˆ Variation due to a flow variable perturbation
L Reference length (blade chord) δ Variation due to a geometric perturbation

© 2005: Royal Institution of Naval Architects


δ ij Kronecker delta
ε Constant smoothing coefficient
εˆ Variable smoothing coefficient
γ Pseudo acoustic speed
η Generalized coordinate, section efficiency
Λξ Diagonal matrix of eigenvalues of A
µ Reference dynamic viscosity
νe Eddy viscosity
νT Total viscosity
ρ Reference density
σi Incipient cavitation index
σ ij Fluid shear stress tensor
σ̂ ij Adjoint shear stress tensor
τ ij Fluid total stress tensor
ξ Generalized coordinate
Ψ Vector of adjoint dependent variables
r
ψv Adjoint velocity vector
Ω Ratio of blade section force sensitivities
1. INTRODUCTION
By the nature of their usage, podded thrusters spend
much of their operational life in off-design conditions.
For this reason, there is a strong impetus to consider off-
design performance as early as possible in the design
process.
Broadly speaking, propulsor blade design is generally
carried out in two phases. The first (preliminary) design
phase is characterized by trade-off studies using low-
fidelity flow and geometry models. The purpose of this
phase is to establish the broad characteristics of
candidate designs (e.g., sizing, blade counts, layout, etc.)
that may meet the mission requirements. The second
(detailed) design phase is primarily concerned with the
analysis of three-dimensional candidate geometries. The
purposes of this phase are to weed-out weak designs,
iterate on promising designs, and ultimately produce a
three-dimensional design-of-record. This phase involves
both computational and experimental efforts and
accounts for perhaps 90% (or more) of the overall design
cost.
Bridging these two design phases is a well-defined series
of steps leading from a desired circumferential mean
performance to a three-dimensional geometry. Briefly,
these steps are as follows: Given the desired meridional
flow path (i.e., the end wall locations) and blade loadings,
a Streamline Curvature (SCM) solution is obtained. The
outputs of this solution are the detailed mean meridional
stream surfaces and the leading and trailing edge mean
velocity triangles for each blade row. At typically 6-12
spanwise locations, these velocity triangles along with a
desired thickness distribution are then run through a
Mean Streamline (MSM) analysis. This code provides the
blade section camber line definition appropriate for the
Marine CFD 2005: Southampton, UK
input LE/TE velocity triangles and the stream sheet
geometry at a given spanwise location. The resulting
blade section shapes, along with rake and skew
distributions are then input to a blade section stacking
code (STK). Finally, the product of the section stacking
code is a three-dimensional candidate geometry. This 3D
geometry is then available for RANS analysis or small-
scale physical model construction and testing; the
purpose of either is simply performance evaluation.
The steps just described form a design iteration. The use
of high-fidelity CFD simulations can reduce dramatically
the calendar time and manpower needed to complete
such an iteration, though with the familiar caveats
associated with RANS simulations. Whether CFD
simulations or small scale model testing is used for
performance evaluation, any optimization that is
currently attempted is done as a cut-and-try process. To
date, a formal optimization step is generally not included
as part of the marine propulsor design process.
2. MULTIPLE OPERATING POINT SHAPE
OPTIMIZATION
The long term goal of the path of research to which this
work belongs is routine shape optimization of complex
three-dimensional propulsor blades. To this end, an
incremental approach was adopted. The immediate goal
of this work is the demonstration of the technology on
relevant problems in two-dimensional inviscid and
viscous flows. Follow on work will focus on
demonstrating the approach on quasi-three-dimensional
stream sheets and the construction of a viable design tool
to be inserted into the environment described in the last
section. Work on truly general three-dimensional shape
optimization will naturally follow that effort (though
some initial efforts are described in references [1,2]). The
work described in this paper focuses on shape
optimization of two-dimensional blade sections operating
in a cascade configuration. The section shapes are
modified to improve surface-type cavitation performance
and, if possible, efficiency over a range of incidence
angles.
2.1 CONTINUOUS ADJOINT APPROACH
It is ultimately intended to insert a formal shape
optimization step into an existing and well-established
design environment. It is assumed, therefore, that the
starting point for any re-design is an existing (and
probably quite good) blade section design. For this
reason it is appropriate to adopt a gradient-based
optimization procedure. Because the number of cost
functions and constraints is low (as will be apparent) and
the number of design variables is reasonable large (as
will be shown) an adjoint equation-based approach to the
gradient calculation is appropriate. Specifically, in this
work, the continuous adjoint approach of Jameson [3] is
adopted.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
This work also considers multipoint design, i.e., shape
optimization over a series of N p flow conditions. The
approach followed in this work closely follows that of
Reuther, et al [4]. The model unconstrained multipoint
optimization problem may be stated in the following
way:
Determine the N D design
variables, bi , that minimize (1)
the cost function, I = ∑ c n I n (w n , F)
Np
where N p is the number of operating points, c n are
scalar weights, F is a vector of parameters describing the
geometry, and w n is the vector of flow variables for the
nth operating point. For the solution of this problem a
simple steepest descent approach is utilized. Whereas the
convergence rate of steepest descent is well-known to be
quite slow for most problems, the convergence properties
of the continuous adjoint approach and the small expense
of the gradient calculation will likely more than offset the
cost of the larger number of steps required to reach an
optimum in comparison to a conjugate gradient or quasi-
Newton approach (Reuther [5]). The steepest descent
update of the vector of design variables for problem (1)
is given simply by
b k +1 = b k − α k ∑c GNp
n
k
n (2)
where α k is a scalar step size, c n is the scalar weight
from the multipoint cost function, and G kn is the cost
function gradient vector for the nth cost function at the kth
design cycle.
The cost function gradient is computed following the
method of Jameson and Reuther [6] and is reviewed
briefly here. A perturbation of each of the N D design
variables results in a variation of the cost function that
can be expressed as
∂I ∂I
T T
δI = δw + δ F = δ I + δˆI (3)
∂w ∂F
where the overbar indicates a variation due to the flow
field perturbation and the carat denotes a variation due
solely to the geometric perturbation. Assembling the
gradient using N D evaluations of equation (3) (i.e., the
finite-difference approach) is clearly cost-prohibitive as
it requires at least N D + 1 evaluations of the flow field.
The costly flow field dependence, however, can be
eliminated from equation (3) by introducing the flow
field governing equations and a judiciously chosen new
field variable. The flow field governing equations (i.e.,
the Euler or Navier-Stokes equations) may be written
symbolically as
R (w, F) = 0 . (4)
© 2005: Royal Institution of Naval Architects
In analogy with equation (3), the variation of the flow
equations due to a design variable perturbation may be
expressed as
∂R ∂R
δR = δw + δ F = δ R + δˆR = 0. (5)
∂w ∂F
If equation (3) is augmented with the inner product of an
arbitrary function, Ψ , with equation (5), the result is
[ ]
δI = δ I + Ψ T δ R + δˆI + Ψ T δˆR . (6)
It is then easily shown that the expensive bracketed term
in equation (6) may be analytically eliminated if the new
field variable (i.e., the adjoint variable) satisfies the
following equation:
∂R ∂I
T
ΨT =− . (7)
∂w ∂w
The final result is that the cost function variation is given
simply by
δI = δˆI + Ψ T δˆR. (8)
The implications of equation (8) are profound. This
relation states that given a single flow solution [i.e., to
equation (4)] and a single adjoint solution [i.e., to
equation (7)], one can construct the entire cost function
gradient with N D evaluations of equation (8). Evaluation
of equation (8), moreover, requires only the geometric
sensitivity of the cost function and the residual operator
to a perturbation of a design variable. This may be
computed by finite differences very inexpensively
(especially for local design variables) with the ith term of
the gradient expressed simply as
δI
Gi = .
δbi
The following sections describe the specific
implementation of this approach first for improving
multipoint cavitation performance using the Euler
equations, then for single-point efficiency improvement
using the Reynolds-averaged Navier-Stokes equations,
and finally a combination of the two cost functions.
2.2 CAVITATION COST FUNCTION
For a well-designed rim-driven thruster, the most likely
form of incipient cavitation, especially at off-design
operating conditions, is leading edge surface cavitation.
As a given blade operates at very high or very low
incidence angles, a leading-edge suction peak will
develop on the suction or pressure surface respectively.
To first-order the corresponding incipient cavitation
number for this condition is approximated by
 Marine CFD 2005: Southampton, UK

σ i = − min(C p ) (10) axial force. The appropriate cost function for efficiency
improvement is, therefore,
where the minimum is taken over the blade surface and
I= ∫ Bc
τ xx S x + τ xy S y dξ (14)
p −p p − p∞
σ i = 1∞ 2 v and C p = 1 1  ∂u i ∂u j 
2 ρV 2 ρV τ ij = − pδ ij + + 
2
where
Re  ∂x j ∂x i 

( )
r
where p ∞ is the ambient pressure, p v is the vapor and dA = S x iˆ + S y ˆj dξ
pressure, ρ is the reference density, and V is the
reference speed. Unfortunately, equation (10) does not are the total stress tensor and the elemental directed
offer a well-posed cost function in the context of the surface area respectively with the reference Reynolds is
continuous adjoint approach. An alternative approach, given as
however, is to recognize that the problematic minimum ρVL
pressure always occurs at a suction peak. The problem Re = .
µ
can thus be re-cast as an inverse design problem where
the target pressure distribution is the actual (or realized)
2.4 SECTION LIFT CONSTRAINT
pressure distribution with the undesirable suction peak
attenuated. In other words, the following cost function
For both the cavitation and efficiency optimization
can be used:
problems, meaningful results can only be obtained if one
constrains the transverse or “lift” force at one of the
I= ∫ Bc
1
2 ( p − p d ) 2 ds (11) operating points. For the efficiency problem, maintaining
lift is part of the problem statement. For the cavitation
problem, one could always improve the apparent
where s is the arc length around the section and p d is the
cavitation performance by simply loading the blade
target pressure which is a solution to the following: section less; though this is not a practical solution. For
external aerodynamic optimization problems, lift is often
∂ ∂ constrained by altering the angle of attack of the vehicle
pd − εˆ pd = pa . (12)
∂ξ ∂ξ throughout the design iterations. In this work, the lift is
constrained by changing the stagger angle, β , of the
In equation (12), ξ is a computational coordinate along cascade. The size of the adjustment is determined by a
the section, p a is the actual pressure at the current design continuously updated sensitivity maintained throughout
the design iterations. At each design iteration, after the
cycle, and εˆ is determined by section shape alteration, a stagger angle change, ∆β , is
computed to satisfy the lift constraint, the section is
ε ⋅ ∂ 2 pa ∂ 2 pa
for ξ − ξ min C ≤ δ
 ∂s 2 ∂s 2 rotated, and the domain is re-meshed. In this way a
εˆ = 
p
min C p stag pt.
(13) constant lift can be maintained throughout the design
 0 for ξ − ξ min C > δ
p
iterations.

where ε is an O(1) constant and δ is some specified 2.5 GOVERNING FLOW EQUATIONS
number of mesh intervals. Note that equation (12) is
nothing more than an implicit smoothing operation on In this work, the assumption of incompressible flow is
the blade surface pressure distribution, and specifying the appropriate. Also, because the interest is in the shape
smoothing coefficient according to (13) restricts its effect optimization of blade sections operating in time-mean
to “hammering” down the suction peak. flow, the assumption of steady flow is reasonable. A
pseudo-time coordinate can then be used to facilitate
2.3 EFFICIENCY COST FUNCTION convergence to steady state. To this end, the appropriate
form of the governing flow equations is the pseudo-
For a thruster, maximizing efficiency is equivalent to compressible Reynolds-averaged Navier-Stokes
maximizing the thrust force produced at fixed power. In equations given (in generalized body-fitted coordinates)
the context of two-dimensional, linear cascades, this is by
analogous to maximizing the axial force while fixing the
∂w ∂
transverse force. Unlike transonic or supersonic flow, in
JP + (E − E v ) + ∂ (F − Fv ) = 0 (15)
incompressible flow, this problem only has meaning ∂t ∂ξ ∂η
when viscous effects are present. Note also that if the x- where
direction is positive downstream (as is traditional), the
“thrust” force points in the negative direction. As such,
maximizing thrust is equivalent to minimizing the net

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

 p  U   V  where
      1
w =  u , E = J uU + ξ x p , F = J uV + η x p , J= = x ξ yη − xη y ξ .
v  vU + ξ p  vV + η p  ξ xη y − ξ yη x
   y   y 
and
The domain in which equation (15) is solved is shown
schematically in Figure (1) with the bulk flow running
 0   0  from left to right. Topologically, computational mesh is a
    C-type mesh with the ξ computational coordinate
E v = J ξ x σ xx + ξ y σ xy , Fv = J η x σ xx + η y σ xy .
ξ σ + ξ σ  η σ + η σ  extending from the lower outlet around the leading to the
 x xy yy   x xy yy 
upper outlet and the η coordinate extending from the
y y

blade surface outward. The boundary conditions for

In equation (15) the contravariant velocities are defined
viscous flow consistent with the labelling of Figure 1 are
by
given by
U = ξ x u + ξ y v , V = η x u +η y v . r
B c : V w = 0, ∂p ∂n = 0
The viscous stresses are given by B i : u ( y ), v( y ) specified , ∂p ∂n = 0
(16)
∂u ∂v  ∂u ∂v  B , B e2 :
1
w = M ξ (LM ξ−1 w e + (I − L)M ξ−1 w i )
σ xx = 2ν T , σ yy = 2ν T , σ xy = σ yx = ν T  + 
e

∂x ∂y  ∂y ∂x  P, W : 2 cell overlap.

(1 +ν ) In the above, the subscript w denotes the wall value, n is

where νT = e
the blade surface normal direction, M ξ is the ξ -
Re
direction modal matrix, and L is a diagonal selection
and ν e is the eddy viscosity. matrix to choose the incoming waves, i.e.,
L = diag [0 0 1] . In words: no-slip velocity and zero
The preconditioning matrix is the standard diagonal form pressure gradient conditions are applied the blade
of Chorin [7], i.e., surface; specified velocity and extrapolated static
pressure are enforced at the inflow; characteristics-based
P = diag [ 1 γ 2 1 1] conditions are applied at the exit planes; and the periodic
boundaries and wake cut are handled with
straightforward two-cell overlap conditions.
where γ is the pseudo acoustic speed. Finally, for
completeness, the metric terms are given by the
following:

Jξ x = yη , Jξ y = − xη , Jη x = − y ξ , Jη y = xξ

BC W

Bi
Be2

Be1

Figure 1: Schematic of the cascade computational domain with the various boundaries labeled.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

r
2.6 ADJOINT EQUATIONS ψ v = ψ 2 iˆ + ψ 3 ˆj.
In the interest of brevity, the step-by-step derivation of
Like the flow equations, the adjoint equations (17) are
the appropriate adjoint equations will not be covered in
solved on the domain depicted in Figure 1. The necessary
detail. The process, described by equations (3) through
boundary conditions for the adjoint variable on this
(7), is applied to the cost functions (11) and (14) with the
domain, that are common to both cost functions, are
flow equations given by equations (15) and (16). Good
given by
examples of the steps involved for various cost functions
and flow equations may be found in Jameson and r
Bi : ψ v = 0
Reuther [6], Reuther [5], and Dreyer [1,2]. Briefly, the
process begins by taking the first variation of the flow B 1e , B e2 : Ψ T PM ξ Λ ξ = 0 for outgoing waves (18a)
equations (15). An inner product of the resulting P, W : 2 cell overlap.
expression with an arbitrary adjoint variable is then
integrated over the flow domain. The boundary
conditions (16) are incorporated and the result is added to On the blade surface, however, the boundary conditions
a variation of the cost function [(11) or (14)]. differ for the two cost functions and are given
Appropriate forms for the adjoint are then easily respectively by
determined that eliminate the flow variation r r
dependencies from both the field and domain boundaries. B c : Cavitation : ψ v ⋅ n = p − p d
The final results are a field equation and boundary Efficiency : ψ 2 = −1; ψ 3 = Ω, (18b)
conditions for the adjoint variable. An adjoint solution ∂Fx ∂F y
satisfying these relations may then be used in equation Ω= .
(8) to inexpensively build the cost function gradient. ∂β ∂β

Because both cost functions (11) and (14) are blade A good discussion of the remaining (arbitrary) boundary
surface-based integral metrics, the adjoint derivations for condition choices may be found in Reuther [5]. In the
the two are very similar. Not surprisingly, the only above boundary conditions, Λ ξ is the diagonal matrix of
difference in the respective adjoint equations is the blade eigenvalues of A, Fx and F y are the axial and transverse
surface boundary condition. The adjoint field equation
for both cost functions is given by blade section forces, and β is the blade stagger angle.
Note that the blade surface boundary condition is a
∂Ψ ∂Ψ ∂Ψ ∂Eˆ v ∂Fˆ v transpiration type condition for the inverse design
JP T − AT − BT − − = 0. (17) (cavitation) cost function, and a straightforward Dirichlet
∂t ∂ξ ∂η ∂ξ ∂η condition on the adjoint velocity for the efficiency cost
function.
In equation (17) A and B are the flux Jacobian matrices
from the flow equations, i.e., 2.7 DESIGN VARIABLES AND DOMAIN
RE-MESHING
∂E ∂F
A= and B =
∂w ∂w In this work a design variable is associated with each
blade surface-coincident point in the computational mesh.
and the viscous terms are given by Each blade surface-coincident point can move in a
direction normal to the local surface element and the
    design variable is the scalar distance in this direction. In
0 0
  ˆ   this work, the design variables are updated at the kth
E v = J ξ x σˆ xx + ξ y σˆ xy , Fv = J η x σˆ xx + η y σˆ xy 
ˆ
design cycle using the steepest descent approach, i.e.
ξ σˆ + ξ σˆ  η σˆ + η σˆ  ~
 x xy y yy   x xy y yy  δb k = −α G k
where
~
where α is a scalar step size and G is a smoothed and
∂ψ 2 ∂ψ 3 attenuated cost function gradient. The smoothing of the
σˆ xx = 2ν T , σˆ yy = 2ν T ,
∂x ∂y gradient is required to ensure smooth geometry updates
given the inherently non-smooth, discrete nature of the
 ∂ψ 2 ∂ψ 3 
σˆ xy = σˆ yx = ν T  + . design variables. The smoothing operator is a standard
 ∂y ∂x  implicit type. The gradient is also attenuated at the
geometric leading and trailing edges to preserve the
In other words, the viscous operator in the adjoint chord length throughout the design cycles. It is
equations is nothing more than the physical viscous straightforward to show that neither the smoothing nor
operator acting on an adjoint velocity defined by attenuation operations alter the descent property of the
gradient [2]. The blade geometry is then updated at the ith
surface point according to

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

~ ~
δx ik = −α G ik n x and δy ik = −α G ik n y 2.8 FLOW AND ADJOINT NUMERICAL
SOLUTION
where n x and n y are the components of the local surface It is important to note that the adjoint field equations (17)
unit normal vector. are mathematically analogous the linearized form of the
flow equations (15). Both sets of equations are
Once the blade section shape is altered at the end of a advection-diffusion equations; both systems have the
design cycle, the surrounding field mesh must be same wave speeds (though pseudo-acoustic waves travel
regenerated for the flow and adjoint solutions of the next in opposite directions in pseudo-time); and both have
design cycle. This is accomplished using a very simple identical viscous operators. This similarity is exploited
automatic domain remeshing procedure. The procedure by using the same numerical scheme to solve both
algebraically relates field point movement to blade systems.
surface point movement for all field points that share an
η -direction mesh line with a surface point. (All other Briefly, the flow and adjoint solvers are based on well-
points remain stationary.) The only requirement on mesh established computational techniques. Spatial
movement is that it ceases at the η max boundary. discretization is cell-centered, finite volume; central
differences are used for advection and diffusion terms;
Discretely, the ijth field point is updated according to:
and an eigenvalue-scaled fourth-difference artificial
δx ij = C ij δx ik and δy ij = C ij δy ik dissipation is added for stability [8]. The discrete
equations are integrated in pseudo-time to a steady state
where C ij is a scalar deflator defined as condition using a five-stage Runge-Kutta-like scheme [9].
For convergence acceleration, local time-steps, implicit
C ij = 1 − (3 − 2s ij ) s ij2
residual smoothing and a W-cycle multigrid scheme are
used [10]. For turbulent flow simulations, the eddy
where s ij is the normalized arc length along the η - viscosity is constructed using either the algebraic model
direction mesh line measured from the blade surface to of Baldwin and Lomax [11] or the one-equation model of
the outer boundary. The purpose of the deflator is to Spalart and Allmaras [12].
smoothly attenuate grid point movement as the outer
boundary is approached. This simple remeshing
procedure turns out to be very robust in practice.

1.8

BASELINE
1.6

1.4

1.2
(d)
1.0
σ

0.8

0.6

0.4

0.2

0.0

α
28 30 32 34 36 38 40 42 44 46 48

SYN103I SYN103I SYN103I

0.5 0.5 0.5

0.0 0.0 0.0

Cp
Cp

Cp

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

(a) (b) (c)
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
x x x

Figure 2: Cavitation performance of the baseline NACA 65410 cascade: (a) -5° incidence, (b) 0° incidence, (c) +5°
incidence, (d) cavitation bucket.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

(a)
1.25 0.05 (b) SECTION DESIGN CYC: 0

C p DESIGN CYC: 0
-0.13
SECTION DESIGN CYC: 50
0.00
1.20
0.5 C Tp DESIGN CYC: 50
-0.05 C p DESIGN CYC: 50
1.15 -0.14

-0.10
1.10 CL
CD -0.15 -0.15
1.05 σ 0.0

CL, CD

Cp, CTp
∆β
dβ -0.20
σ

1.00
-0.16 -0.25

0.95
-0.30
-0.5
0.90 -0.17
-0.35

0.85
-0.40
-0.18
0.80 -0.45 -1.0

0.75
0 10 20 30 40
-0.50
SYN103I
NDES 0.0 0.2 0.4 0.6 0.8 1.0
x

Figure 3: Single point optimization of the NACA 65410 cascade for improvement of cavitation performance at the +5°
incidence: (a) evolution of the key parameters over the design cycles, (b) initial and final blade shapes and surface
pressure distributions.

3. RESULTS optimizing for this condition. Figure 3(a) shows the

All of the results in this paper are obtained for a NACA
65410 cascade with 30° of stagger and a pitch to chord
ratio of 1.0. The design condition for this cascade is +5°
or 35° absolute flow angle, α , as measured from the +x-
axis clockwise to the chord line. Note that the blade
sections appear inverted for the analyses in this work.
3.1 CAVITATION
Leading edge surface cavitation at off-design operating
conditions is an incidence-driven phenomenon. As such,
it is predominantly inviscid in nature and it is reasonable
to investigate it using the Euler equations. Figure 2(a-c)
shows surface pressure distributions computed for the
NACA 65410 section operating at design and +/- 5°
incidence (i.e., 35°, 40°, and 30° flow angle respectively).
Figure 2(d) summarizes the corresponding cavitation
numbers for these three conditions. These cavitation
numbers are simply the negative of the minimum surface
pressure from each solution. Not surprisingly, the best,
i.e., the lowest, cavitation number of the three occurs at
the design condition and cavitation performance
degrades at both the high and low incidence conditions.
Also, it is clear that each off-design condition exhibits a
sharp suction peak near the leading edge: on the pressure
side for the low incidence condition and suction side for
the high incidence condition. The ultimate goal is to
flatten or broaden the cavitation bucket shown
(somewhat crudely) in Figure 2(d). Note that for all
inviscid shape optimization cases, there are 113 blade-
surface-coincident mesh points, therefore, the number of
design variables, N D , is 113.
Before carrying out a multipoint optimization, it is useful
to show the results of single point optimization around
each condition. For illustrative purposes, we focus on the
+5° incidence case; Figure 3 summarizes the results of
evolution of the cavitation number, section forces, and
stagger angle over 50 design cycles. For this case, the
cavitation number was reduced from 1.225 for the
NACA 65410 to 0.837 for the optimized section while
maintaining section lift by adjustment of the blade
stagger angle. Figure 3(b) shows the initial and final
blade section shapes as well as the initial and final
surface pressure distributions. The initial and final
section shapes are very similar to one-another though
closer examination reveals the optimized blade to be
slightly thicker than the baseline NACA 65410 and
oriented at a slightly larger stagger angle. A comparison
of the initial and final pressure distributions shows that
the undesirable suction peak was effectively removed
without significantly impacting the pressure distribution
over most of the remainder of the section. However, as is
often the case with single point optimization, off-design
performance usually degrades with improvements at the
design condition. Figure 4 illustrates this phenomenon
for this example.
Figure 4(a-c) show the surface pressure distributions at -
5°, 0°, and +5° incidence for the blade that was
optimized for performance at the +5° operating point (i.e.,
the blade shown in Figure 3). Figure 4(c) shows the same
pressure distribution as the final distribution shown in
Figure 3(b) with its attenuated suction peak. Figure 4(b)
shows that this blade also avoids the formation of a
suction peak at the 0° condition. However, Figure 4(a)
shows the appearance of a significant suction peak when
this blade operates at -5° incidence. In fact, this suction
peak is deeper than that shown by the baseline NACA
65410 at this same condition [Figure 2(a)]. Figure 4(d)
summarizes this behaviour for single point designs at all
three conditions. In fact, as Figure 4(d) illustrates, all
three single point designs exhibit worse off-design
cavitation performance than the baseline NACA 65410.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

1.8

1-PT (35)
1.6 1-PT (30)
1-PT (40)
1.4 BASELINE

1.2
(d)
1.0

σ
0.8

0.6

0.4

0.2

0.0

α
28 30 32 34 36 38 40 42 44 46 48

SYN103I SYN103I SYN103I

0.5 0.5 0.5

0.0 0.0 0.0

Cp
Cp

Cp
-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

(a) (b) (c)
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
x x x

Figure 4: Performance of the single point optimized NACA 65410: (a) -5° incidence, (b) 0° incidence,
(c) +5° incidence, (d) cavitation bucket.
35 CL
35 σ
30 CL
1.40
30 σ 0.00 0.30
40 CL
40 σ
-0.02
1.30 ∆β
-0.04 0.25
1.20
-0.06

1.10
-0.08
0.20

1.00 -0.10

-0.12
∆β
CL
σ

0.90 0.15

-0.14
0.80
-0.16
0.10
0.70
-0.18

0.60 -0.20
0.05
-0.22
0.50

-0.24
0.40 0.00
0 10 20 30 40 50

NDES
Figure 5: Evolution of the cavitation number, section lift coefficient, and stagger angle for the multipoint optimized
NACA 65410 cascade. Note that the legend in the figure labels according to flow angle (i.e., 35 = 0°. incidence,
30 = -5° incidence, 40= +5° incidence)

To broaden the cavitation bucket of the NACA 65410 unweighted and the two off-design conditions (+/-5°) are
cascade requires the simultaneous consideration of equally weighted, i.e., choosing
multiple operating points. In the present context, this
amounts to choosing the scalar weights in equation (1) c1 = 0.0 , c 2 = 0.5 , and c3 = 0.5
for the three operating points under consideration. There where the subscript 1 refers to the design condition and
are an infinite number of possible choices. It turns out in subscripts 2 and 3 are the off-design conditions. In this
this case that very good off-design performance is mode, the design condition is used only to fix the blade
achieved when the design condition (0°) is completely section lift coefficient and the off-design conditions are
used exclusively to build the cost function gradient.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

1.8
1-PT (35)
1.6 1-PT (30)
1-PT (40)
3-PT (30,35,40)
1.4 BASELINE

1.2

1.0 (d)

σ
0.8

0.6

0.4

0.2

0.0

α
28 30 32 34 36 38 40 42 44 46 48

SYN103I SYN103I SYN103I

0.5 0.5 0.5

0.0 0.0 0.0

Cp

Cp

Cp
-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

(a) (b) (c)
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
x x x

Figure 6: Performance of the three-point optimized NACA 65410 cascade: (a) -5° incidence, (b) 0° incidence,
(c) +5° incidence, (d) comparison of the cavitation buckets.
2.4

2.2
NACA 65410
2.0 3-POINT OPTIMIZED

1.8

1.6 Design Points

1.4
σ

1.2

1.0

0.8

0.6

0.4

0.2

α
26 28 30 32 34 36 38 40 42 44

Figure 7: Comparison of the computed cavitation buckets for the baseline NACA 65410 cascade and the three-point
optimized cascade (design points indicated).

Figure 5 shows the evolution of the cavitation number, 6(a-c), when compared directly to Figures 4(a-c), clearly
section lift coefficient and stagger angle over the design show the improvement in section pressure distributions
iterations for this three point example. Using this across this 10° incidence range. Figure 6(d), in a crude
multipoint cost function approach, the cavitation sense, also shows the effective broadening of the
performance at both off-design operating points is cavitation bucket relative to the baseline as well as all of
simultaneously improved with only a small penalty in the single point designs. To verify the actual cavitation
cavitation performance at the design condition. Blade performance of the baseline and optimized blades, both
section lift at the design condition is maintained blade sections were simulated over a 16° angle of
throughout the iterations by adjustment of the section incidence range in 1° increments. Figure 7 compares the
stagger angle with the optimized section having about more resolved cavitation buckets of the original and
0.27° more stagger than the baseline NACA 65410. optimized blade sections. Also indicated in this figure are
the points about which the baseline blade was optimized.
Figure 6 summarizes the cavitation performance of the 3- Although the depth of the bucket for the optimized
point optimized blade relative to the baseline NACA section is slightly shallower, the optimized blade clearly
65410 and the single point optimized sections. Figures shows the desired broadening of the bucket indicating

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

less sensitivity to surface cavitation in off-design Figure 8(a) shows the evolution of the section efficiency,
conditions, especially in the low incidence end of lift, and drag over 100 design cycles. It also shows the
operating range. It should also be noted that both blades variation of the stagger angle needed to maintain lift. In
generate identical lift at the design condition ( α = 35° ). the present context, the following relations are relevant:
In essence, some of the cavitation bucket depth has been
traded off for width. The asymmetry of the improvement C L = transverse force coefficient
also suggests that it may be desirable to asymmetrically C D = axial force coefficient
weigh the cost function to further manipulate the off-
design performance. η ≡ CD CL .

3.2 EFFICIENCY Clearly evident in Figure 8(a) is a monotonic increase in

section efficiency over the design iterations. Also evident
Improving the robustness of a blade design for leading is a large increase in stagger angle. Over the course of
edge incidence-driven cavitation performance at off- the 100 design iterations, section efficiency is increased
design conditions tends to thicken the blade. On the by 0.5% and it continues upward even as the iterations
downside, thicker blades will increase the weight of a were terminated. Figure 8(b) shows the mechanism for
propulsor but, on the upside, will likely improve strength this improvement. Absent any constraints on blade
and acoustic performance. Hydrodynamically, thicker thickness, in order to improve efficiency, the blade
blades will likely degrade the propulsive efficiency section has thinned considerably over the course of the
relative to the baseline propulsor. In incompressible flow, design iterations. In the effort to maintain lift, the section
one can only consider efficiency in the context of viscous incidence has been increased by adjustment the stagger
flow. For this reason, all results hereafter are generated angle. The net result is a leading edge-biased loading and
from solutions of the RANS flow equations and the the formation of a suction peak that was absent in the
viscous adjoint equations. The baseline geometry is once NACA 65410 at this condition. Intuitively this is what
again the NACA 65410 cascade with a pitch to chord one would expect to occur, i.e., that the section would
ratio of 1.0 and the flow conditions are such that the evolve toward a cambered, very thin shape. Aggressively
chord Reynolds number is 1x106. Note that for all thinning the section has negative ramifications for both
viscous shape optimization cases in the following, there structural and acoustic performance; from Figure 8(b) it
are 225 blade surface-coincident mesh points, therefore, is also apparent that it will degrade cavitation
N D = 225 . performance. Clearly, unconstrained section efficiency
maximization, in this context, is of limited practical
As a sanity check, we briefly consider single point value. One possible remedy is to enforce permissible
efficiency maximization of the baseline NACA 65410 thickness constraints throughout the design cycles. A
cascade at the design condition of 0° incidence. more interesting approach, however, is to use a
Specifically, we examine the case where N P = 1 and the compound, multipoint cost function.
cost function is given by equation (14).

(a) 1.0
0.610
1.70
(b) SECTION DESIGN CYC: 0

C p DESIGN CYC: 0
0.609 1.60 SECTION DESIGN CYC:100
-0.07
1.50 C p DESIGN CYC:100
0.608
1.40 0.5
1.30
0.607
-0.08 1.20

0.606 1.10
C L, C D

∆βCp

1.00 0.0
η

0.605 0.90
CL -0.09
0.80
0.604 CD
∆β 0.70

0.603
η 0.60
-0.5
-0.10 0.50

0.602 0.40

0.30
0.601
0.20
-0.11 -1.0
0.600
0.10 SYN103I
0 10 20 30 40 50 60 70 80 90 100

NDES 0.0 0.2 0.4 0.6 0.8 1.0

x

Figure 8: Summary of the single point efficiency maximization case: (a) evolution of the efficiency, lift and drag
coefficients, and stagger angle, (b) comparison of the initial and final blade section shapes and surface pressure
distributions.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

(a) (b)
0.6046 -0.060 1.00 1.25 0.6046 -0.060 1.00 1.25
0.90 1.20 0.90 1.20
0.6044 0.6044
0.80 0.80
-0.080 1.15 -0.080 1.15
0.70 0.70
0.6042 0.6042
0.60 1.10 0.60 1.10

0.6040 -0.100 0.50 1.05 0.6040 -0.100 0.50 1.05

0.40 1.00 0.40 1.00
0.6038 0.6038
0.0/1.0 0.30 0.30
-0.120 0.95 -0.120 0.95
0.20 0.20
0.6036 35 CL 0.6036
0.10 0.90 0.10 0.90
35 CD 5.0/1.0

CL, CD

CL, CD
35 σ

∆β

∆β
η

η
σ

σ
0.6034 0.00 0.85 0.6034 0.00 0.85
-0.140 -0.140
40 σ -0.10 35 CL -0.10
0.6032 35 η 0.80
0.6032 35 CD 0.80
∆β -0.20
0.75 35 σ
-0.20
0.75
0.6030
-0.160 -0.30
0.6030 40 σ -0.160 -0.30
-0.40 0.70 35 η -0.40 0.70
∆β
0.6028 -0.50 0.65 0.6028 -0.50 0.65
-0.180 -0.180
-0.60 0.60 -0.60 0.60
0.6026 0.6026
-0.70 -0.70
0.55 0.55
-0.200 -0.80 -0.200 -0.80
0.6024 0.6024
-0.90 0.50 -0.90 0.50

0.6022 -1.00 0.45 0.6022 -1.00 0.45

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100

NDES NDES

Figure 9: Evolution of efficiency, lift and drag coefficients, stagger angle, and cavitation number over 100 design cycles
for the two-point, combined efficiency/cavitation shape optimization of the NACA 65410 cascade:
(a) relative weights = 0.0/1.0, (b) relative weights = 5.0/1.0.

3.3 COMBINED CAVITATION AND

EFFICIENCY c1 = 0.833 and c 2 = 0.167

For illustrative purposes, consider a two-point for the design point efficiency and off-design cavitation
optimization problem. The problem is to decrease the 5° cost functions respectively. Once again, the design point
incidence (40° flow angle) incipient cavitation number lift was constrained by pitch angle adjustment. [Note that
while constraining the lift at the design point (35° flow Figure 9(b) is plotted on the same scales as Figure 9(a) to
angle). Figure 9(a) summarizes the results for 100 design facilitate comparisons between the two cases.] The final
cycles (though from the figure it is apparent that the incipient cavitation number spread for the optimized
design stabilizes after only 25 design cycles). The figure blade in this case is 0.68/0.88 – worse for both conditions
shows that the initial NACA 65410 cascade has an than the previous case [Figure 9(a)] but this blade
incipient cavitation number spread of 0.53/1.15 for the achieves this spread with only a 0.07% reduction in
design/off-design conditions respectively (at design cycle efficiency; which is considerably better than the previous
0). At the end of 100 design cycles, the cavitation case.
number spread for the optimized blade has improved to
0.59/0.71. As has been evident in other cases, some
NACA 65410 Cascade
design point performance has been sacrificed for off-
Cavitation (40) Optimized
design improvement. The weights for this shape
Efficiency (35) & Cavitation (40) Optimized
optimization case were set at

c1 = 0.0 and c 2 = 1.0

for the design and off-design cost functions respectively;

implying, once again, that the design point was used only
to constrain the blade lift. Also plotted in Figure 9(a) is
the section efficiency at the design flow condition. From
this it is apparent that the efficiency of the optimized
cascade is about 0.33% lower than the initial NACA
65410 cascade. This is, of course, due to the inevitable SYN103I
thickening of the blade section to increase robustness to 0.0 0.2 0.4 0.6 0.8
x
incidence angle. Depending on the particular design
problem, this loss in efficiency may not be tolerable. Figure 10: Comparison of the baseline NACA 65410
cascade with off-design cavitation-only and multipoint
To mitigate this undesirable phenomenon, a compound, efficiency/cavitation optimized blade sections.
multipoint cost function is examined. Specifically, the
efficiency-based cost function (14) is implemented at the Figure 10 directly compares the optimized blade section
design point in conjunction with the cavitation mitigation shapes for the baseline NACA 65410 cascade and the
cost function (11) at the off-design condition. Figure 9(b) optimized sections described in Figures 9(a) and (b). All
shows the results of 100 design cycles using a relative three sections are very similar in shape. The baseline
weighting of NACA 65410 cascade provides the largest axial force of

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK
the three for the specified transverse load; though it has
the poorest off-design cavitation performance. The off-
design, cavitation-only section yields the best cavitation
characteristics of the three but provides the least axial
force. The combined efficiency and cavitation optimized
section is a compromise; it demonstrates better cavitation
performance than the baseline cascade and provides
higher thrust than the cavitation-optimized cascade.
Interestingly, the final stagger angle of the compromise
cascade does not lie between the other two sections – this
is an outcome that doesn’t seem to be intuitively obvious.
The point to be stressed here is that there is a lot of
flexibility inherent in this approach to trade-off cavitation
and efficiency over a range of operating conditions while
maintaining constant powering performance at the design
condition.
4. CONCLUSIONS
In this work, efficient multipoint shape optimization was
demonstrated. The context was marine thruster design
and the performance metrics used were leading edge
incipient surface cavitation and section efficiency. The
optimization procedure used was a simple gradient-based,
steepest descent approach. The required gradients were
constructed very efficiently from the solution of a
continuous adjoint problem. The derivation of the
appropriate adjoint problem for both performance
metrics were outlined in the context of two-dimensional,
turbulent cascade flow.
Several demonstrative shape optimization procedures
were executed, they included: single and multiple
operating point cavitation performance improvement,
single-point section efficiency improvement, and
multipoint combined cavitation and efficiency
performance improvement. In each case, performance
improvement in the desired metric or metrics was
achieved while maintaining the specified design point
performance; this latter feature is essential for developing
a useful design tool. The multipoint combined
cavitation/efficiency cost function case in particular also
vividly demonstrated the flexibility inherent in the
proposed approach to these types of shape optimization
problems.
5. ACKNOWLEDGEMENTS
This work was supported by the Office of Naval
Research under contract number N00014-00-G-0058
monitored by Tom Calvert and Lynn Petersen. Also
essential to this work was the advice and assistance of
Professor Luigi Martinelli of the Department of
Mechanical and Aerospace Engineering at Princeton
University.
6. REFERENCES
1. Dreyer, J. J., Martinelli, L., ‘Hydrodynamic Shape
Optimization of Propulsor Configurations Using a
© 2005: Royal Institution of Naval Architects
Continuous Adjoint Approach,’ AIAA Paper 2001-2580,
June 2001.
2. Dreyer, J.J., ‘Hydrodynamic Shape Optimization of
Propulsor Configurations Using a Continuous Adjoint
Approach,’ Ph.D. Dissertation, Dept. of Mechanical
Engineering, The Pennsylvania State University, 2002.
3. Jameson, A., ‘Aerodynamic Design via Control
Theory,’ J. Sci. Comp., 3(3):233-260, 1988.
4. Reuther, J.J., Jameson, A., Alonso, J.J., Rimlinger,
M.J., and Saunders, D., ‘Constrained Multipoint
Aerodynamic Shape Optimization Using an Adjoint
Formulation and Parallel Computers, Part I,’ AIAA
Journal of Aircraft, Vol. 36, No. 1, 51:60, 1999.
5. Reuther, J.J., ‘Aerodynamic Shape Optimization Using
Control Theory,’ Ph.D. Dissertation, University of
California Davis, 1996.
6. Jameson, A. and Reuther, J., ‘Control Theory Based
Airfoil Design Using the Euler Equations,’ AIAA Paper
94-4272, 1994.
7. Chorin, A. J., ‘A Numerical Method for Solving
Incompressible Viscous Flow Problems,’ Journal of
Computational Physics, 2, 1967.
8. Jameson, A., Schmidt, W., and Turkel, E., ‘Numerical
Solutions of the Euler Equations by Finite Volume
Methods Using Runge-Kutta Time-Marching Schemes,’
AIAA Paper 81-1259, 1981.
9. Martinelli, L., ‘Calculations of Viscous Flows with a
Multigrid Method,’ Ph.D. Dissertation, Department of
Mechanical and Aerospace Engineering, Princeton
University, 1987.
10. Jameson, A., Baker, T.J., ‘Multigrid Solution of the
Euler Equations for Aircraft Configurations,’ AIAA
Paper 84-0093, 1984.
11. Baldwin, B.S., Lomax, H., ‘Thin Layer
Approximation and Algebraic Model for Separated
Turbulent Flows,’ AIAA Paper 78-257, 1978.
12. Spalart, P.R., Allmaras, S.R., ‘A One-Equation
Turbulence Model for Aerodynamic Flows,’ AIAA Paper
92-0439, Jan. 1992.
7. AUTHOR BIOGRAPHY
James Dreyer holds the current position of Associate
Research Engineer and Head of the CFD Analysis
Department at the Applied Research Laboratory at the
Pennsylvania State University. His primary responsibility
is to provide high-fidelity RANS simulations in support
of the design of marine propulsors and pumps for
manned and unmanned underwater vehicles.
Marine CFD 2005: Southampton, UK

NUMERICAL SIMULATION OF FREE SURFACE WAVE INDUCED SEPARATION

S H Sadathosseini, S M Mousaviraad, and M H Sadr, Petropars Ltd., Iran

SUMMARY

The present simulation deals with numerical calculation of free surface wave induced separation, which is an important
phenomenon in naval architecture and offshore engineering problems. Numerical modeling is found to be very effective
for studying such complicated flows, because of its lower cost and higher level of producible data. The first simulation is
performed for NACA0024 surface piercing hydrofoil over a range of Froude number (0.19, 0.37, 0.55), along with wave
breaking at Fr=1. The NACA0024 foil was exploited for the reason that it almost has no separation at large depths,
where there is no free surface wave, thus isolating wave induced separation. Wave induced separation results are
presented and compared with both available experimental data and previous numerical computations. The wave breaking
flow is also simulated successfully and results are presented. The second modeling is carried out for a circular cylinder,
in order to investigate the shape effects on the wave induced separation. Flow features are studied and discussed with
regard to separation, free surface elevations, and drag coefficients.

NOMENCLATURE
The wave induced separation was first identified by
g Acceleration of gravity Chow (1967) using vertical (surface piercing) and
L Foil chord length & Cylinder diameter horizontal (submerged) foils, designed for insignificant
t Time separation at large depths. Chow observed regions of
separated flow originating just beyond the wave trough
U∞ Free-stream velocity
and in some cases beyond the trailing edge. It was also
X, Y, Z Cartesian coordinates studied by Stern et al. (1989) using a surface piercing flat
′ ′ plate with attached wave generating upstream horizontal
ui u j Reynolds stresses submerged foil (foil-plate model). As with Chow,
r
v Velocity vector separation initiated just beyond the wave trough and
λ Wave length extended to the following wave crest. These studies
υ Kinematic viscosity showed the dependence of the streamwise and depthwise
extent of the separation region on Froude number and
αw Volume fraction of water wave steepness, and that the transverse extent is wedge
αa Volume fraction of air shaped with significant free surface vorticity and
turbulence. Choi and Stern (1993) performed laminar and
ρ Density turbulent CFD calculations for a surface piercing flat
µ Viscosity plate with an external Stokes wave, which simulates the
U∞ Stern et al. experimental geometry. In comparison to the
Fr Froude number, experimental data, the extent was grossly over/under
gL predicted for the laminar/turbulent solutions. Zhang and
U∞L Stern (1996) studied the problem through RANS
Re Reynolds number,
υ (Reynolds Averaged Navier Stokes) simulation with
1. INTRODUCTION exact nonlinear kinematic and approximate dynamic free
surface boundary conditions. The nature of the flow in
Free surface wave induced separation, i.e. separation the separation region was qualitatively similar as Choi
solely due to wave induced effects, is very important in and Stern, but described in detail using a topological rule
ocean and marine engineering. It involves the derived for free surface wave induced separation in
complexities of free surface deformations, vorticity, and which free surface streamlines are treated similarly as
turbulence, along with the already formidable subject of skin friction lines. Zhang and Stern also remarked that
three dimensional boundary layer separation. The the flow is naturally unsteady for high Froude numbers,
boundary layer becomes complicated when there is a free and both further numerical and experimental study are
surface due to the effects of gravity waves and free needed for accurate analysis of flow characteristics and
surface boundary condition. Such effects are a unique wave breaking. Pogozelski et al. (1997) performed
and poorly understood problem of ship and platform experimental study of free surface wave induced
hydrodynamics due to wave making, wave breaking, separation, but with different foil geometry. Metcalf et al.
and/or incident waves, and are important with regard to (2001) provided detailed experimental data
ship performance, wake signatures, and platform documentation of the wave elevations and surface
stability. The NACA0024 foil is a simplified geometry pressures for surface piercing NACA0024 hydrofoil.
that has insignificant separation at large depths, thus Kandasamy et al. (2001) used CFDSHIP-IOWA (a
making an ideal geometry by isolating the wave induced general purpose research code for ship hydrodynamics)
separation. for RANS simulation of wave induced separation. They

© 2005: Royal Institution of Naval Architects


also studied the effects of blockage considering four
different solution domains.
The present computation presents an improvement in
numerical modeling and shows better agreement with
experimental data than all previous numerical
calculations. It is due to using a robust free surface
modeling technique, VOF (Volume Of Fluid), which
takes the effects of outer air into consideration, and
solves RANS equations simultaneously for both water
and air.
Wave breaking, which occurs at very high Froude
numbers, is also an intricate problem, and is simulated in
this study. The results of wave breaking at Fr=1 are
presented, but there are no experimental data or previous
numerical results for comparison.
Moreover, the present study includes shape effects
investigation for free surface wave induced separation. A
circular cylinder, which has significant separation at
large depths, is modeled to examine the interaction of
wave induced and shape induced separations.
2. COMPUTATIONAL METHOD
The CFD (Computational Fluid Dynamics) results are
obtained solving RANS equations by finite volume
method. The treatment for the free surface flow uses an
interface capturing method, volume of fluid (VOF). In
this method, an additional transport equation is solved for
the volume fraction of water in each cell. If the volume
fraction of water and air in each cell is denoted as α w
and α a , the tracking of the interface between the phases
is accomplished by the solution of a continuity equation
for the volume fraction of water. This equation has the
following form:
∂α w r
+ v .∇α w = 0 (1)
∂t
The volume fraction equation will not be solved for air;
the volume fraction of air will be computed based on the
following constraint:
αw + αa = 1 (2)
The properties appearing in the transport equations are
determined by the presence of the component phases in
each control volume. For example, the density in each
cell is given by:
ρ = α w ρw + α a ρa (3)
All other properties (e.g., viscosity) are also computed in
this manner.
A single momentum equation is solved throughout the
domain, and the resulting velocity field is shared among
the phases. The momentum equation is dependent on the
volume fractions of all phases through the properties ρ
and µ .
Marine CFD 2005: Southampton, UK
Turbulence is modeled using the Reynolds Stress Model
(RSM). This model involves calculation of the individual
′ ′
Reynolds stresses, ui u j , using differential transport
equations. A single set of transport equations is solved,
and the Reynolds stresses are shared by the phases
throughout the field. The individual Reynolds stresses
are then used to obtain closure of the Reynolds averaged
momentum equation.
3. NUMERICAL MODELLING
A NACA0024 foil with a chord length of 1.2 m, a span
of 2 m (75% in water), and a thickness of 0.29 cm is
considered the first test case.
Three conditions are simulated, with reference to the
experimental data, i.e., Fr=(0.19, 0.37, 0.55) and the
6
corresponding Re=(0.822, 1.52, 2.26)× 10 . In addition,
the flow at Fr=1 is calculated to study the wave breaking
flow of the hydrofoil.
Since the geometry is symmetrical, only half domain,
which consists of 215000 hexahedral structured cells, is
solved. The cells near free surface in both air and water
fields are designed to be very small (2 mm height) to
catch more accurate water deformation results.
The second test case is a circular cylinder, which is
vastly used in offshore platforms. Having a diameter of
1.2 m (equal to the foil chord length), it is intended to
evaluate the shape effects on the wave induced
separation.
Fig. 1 shows the grids generated for the NACA0024 foil
and the circular cylinder.
4. RESULTS AND DISCUSSIONS
Fig. 2 presents the wave profile along the foil, and
compares it with Zhang & Stern experimental and
numerical results for Fr=0.19. The wave profile is similar
to that of a typical ship, and the wave length is slightly
greater than that given by Kelvin wave theory
( λ = 2πFr ). The results are nearly as accurate as
2
Zhang & Stern numerical results, indicating that the
effects of air on numerical results of this test case are not
significant at low Froude numbers. The bow wave peak
is about 1.6 percent of L.
Fig. 3 shows the wave profile along the circular cylinder
for Fr=0.19. The free surface waves are more dominated
by the strong pressure distribution of bluff body, than by
Kelvin waves generated by the high pressure stagnation
point. The water piles up in front of the body, and then
the free surface elevations decrease with an almost
constant steepness before the cylinder shoulders. The
bow wave peak is almost 6.6 percent of L.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

0.15
Experiment [Zhang &Stern] Numerical [Zhang &Stern]
0.1
Present Simulation
0.05

Z(m)
0

-0.05

-0.1

-0.15
0 0.2 0.4 0.6 0.8 1

X/L

Figure 2: Wave Profile along the NACA0024 Foil for

Fr=0.19

0.1

0.05

z(m)
0

-0.05

-0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L

Figure 3: Wave Profile along the Circular Cylinder for

Fr=0.19
0.15

0.1 Experiment [Zhang &Stern] Numerical [Zhang &Stern]

Present Simulation
0.05
Z(m)

-0.05

-0.1

-0.15
0 0.2 0.4 0.6 0.8 1

X/L

Figure 4: Wave Profile along the NACA0024 Foil for

Fr=0.37
0.15
Figure 1: Computational Grids for the NACA0024 Foil Experiment [Zhang &Stern] Numerical [Zhang &Stern]
0.1
and the Circular Cylinder 0.05
Present Simulation
Z(m)

-0.05
Figs. 4 and 5 show the wave profiles along the -0.1
NACA0024 for Fr=0.37 and Fr=0.55, respectively. The -0.15
present modeling agrees better with the experimental 0 0.2 0.4 0.6 0.8 1

X/L
data, which suggests the significance of the air effects at
higher Froude numbers. Figure 5: Wave Profile along the NACA0024 Foil for
Fr=0.55
At these Froude numbers, the wave profiles are different
0.1
from those of ships. The wave profiles are relatively flat
0.05
in the separation regions; and this flat region is smaller
z(m)

for Fr=0.55 than for Fr=0.37. The bow wave peak for 0

Fr=0.37 and Fr=0.55 are 6 and 12 percent of L, -0.05

respectively. -0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L
Figs. 6 and 7 show the wave profiles along the circular
cylinder for Fr=0.37 and Fr=0.55, respectively. The bow Figure 6: Wave Profile along the Circular Cylinder for
wave peak for Fr=0.37 and Fr=0.55 are 9 and 18 percent Fr=0.37
of L, respectively, which are much greater than those of
the NACA0024 foil. Unlike the foil, the wave profiles 0.3

0.2
are not flat in the separation regions. 0.1
z(m)

0
-0.1
-0.2

-0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L

Figure 7: Wave Profile along the Circular Cylinder for

Fr=0.55

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Fig. 8 compares the wave profiles of the NACA0024 Fr=0.19

hydrofoil at different Froude numbers. For Fr=0.19, the 0.1

wave pattern is similar to that of ships. For Fr=0.37, the 0.05

Circular Cylinder NACA0024

bow wave becomes more significant; the wave steepness

z(m)
0
is larger; and the wave pattern is relatively flat in the
-0.05
separation region. For Fr=0.55, the free surface has an
-0.1
even more complicated wave system, with increase in 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

bow wave peak, wave steepness, and distortion of the X/L

free surface in the separation region. Fr=0.37

0.15

Circular Cylinder NACA0024

0.1

0.15 0.05

z(m)
0.1 Fr=0.19 Fr=0.37 Fr=0.55 0

0.05
-0.05
Z(m)

0
-0.1
-0.05 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

-0.1 X/L

-0.15
0 0.2 0.4 0.6 0.8 1 Fr=0.55
X/L 0.3

0.2 Circular Cylinder NACA0024

Figure 8: Comparison of the Wave Profiles of the 0.1

z(m)
0
NACA0024 Foil at Different Froude Numbers
-0.1

Fig. 9 compares the wave profiles of the circular cylinder -0.2

-0.3
at different Froude numbers. For Fr=0.19, the free 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

surface elevations are affected by the strong pressure X/L

distribution of the body, which is similar to that of

circular cylinder at large depths. For Fr=0.37 and Figure 10: Comparison between the Wave Profiles of the
Fr=0.55, the wave patterns are Froude dependent. The NACA0024 Foil and the Circular Cylinder
wave height, the wave steepness, and the distortion in the
separation region become larger with Froude. NACA0024
Frictional Pressure Total
0.3 0.015
Variations of Drag Coefficients

0.2 Fr=0.19 Fr=0.37 Fr=0.55

0.1 0.01
z(m)

-0.1 0.005
-0.2

-0.3 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x
-0.005
Figure 9: Comparison of the Wave Profiles of the 0.19 0.37
Fr
0.55

Circular Cylinder at Different Froude Numbers

Circular Cylinder
Fig. 10 is intended to investigate the shape effects on the Frictional Pressure Total
wave pattern of surface piercing bodies. For Fr=0.19, the 0.3
shape effects are dominant; the wave pattern of the foil is
Variations of Drag Coefficients

0.25
similar to ships, mainly affected by the high pressure 0.2
stagnation point, and the wave pattern of the circular 0.15
cylinder is affected by the strong pressure distribution of 0.1
the blunt shaped body. For Fr=0.37, the wave pattern, i.e. 0.05
the wave steepness and the trend of the wave elevations, 0
is Froude dependent. However, the bow wave peak and -0.05
the distortions in the separation region are larger for the 0.19 0.37 0.55
Fr
circular cylinder, because of its blunt shape. For Fr=0.55,
although the primary pattern of the wave remains Froude Fig 11: Variations of Drag Coefficients versus Fr for the
dependent, the shape effects seems to become stronger; NACA0024 Foil and the Circular Cylinder
the wave steepness, the wave height, and the distortions
in the separation region are more significantly affected
by the shape effects. Fig. 11 presents the variations of pressure, frictional and
total drag coefficients versus Froude number for the
NACA0024 and the circular cylinder. All values are

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

subtracted from the corresponding values for Fr=0.19. As

Fr increases, the pressure drag coefficient increases due
to the effects of the bow wave. The frictional drag
coefficient, on the other hand, decreases as Fr increases
from 0.19 to 0.37 and then increases a little as Fr
increases further to 0.55, which is consistent with the size
of the separation region.

Fig. 12 shows the pressure contours in the symmetry

plane for the NACA0024 foil, in an attempt to show the
free surface elevations around the foil. At low Froude
numbers, the waves are found to be insignificant far from
the body, unlike Fr=0.55 at which water deformations are
extended behind the trailing edge.

Fig. 13 shows the pressure contours for the circular

cylinder. The water deformations are considerably
extended behind the body, especially at Fr=0.55. Besides,
the wake pattern of the circular cylinder is much different
from that of the NACA0024 foil.

Figure 13: Pressure Contours for the Circular Cylinder in

Figure 12: Pressure Contours for the NACA0024 Foil in
the Symmetry Plane
The wave patterns in the wake of the NACA0024 foil
and the circular cylinder are presented in Fig. 14. For
Fr=0.19, Kelvin waves are generated around the
hydrofoil, and the wave pattern of the circular cylinder is
dominated by the pressure distribution of the bluff body.
For Fr=0.37 and Fr=0.55, the primary wave patterns of
the foil and the circular cylinder are similar, though there
are more disorders in the wake of the circular cylinder.
This suggests that at high Froude numbers, the free
surface waves are a function of Froude, and not shape.
Hence, the wave patterns of the NACA0024 foil and the
circular cylinder are similar, despite of their different
shapes. The shape effects are evident on the bow wave
peak, the extent of the free surface effects, and the
distortion in the separation and wake regions.
the Symmetry Plane
Fig. 15 presents the contours of X component of the wall
shear stress on the NACA0024 foil. The separation
regions, where the wall shear stress values are negative,
are also marked in the figure. For Fr=0.19, the wave
effects are limited to depths very close to the free
surface, and the separation region in free surface area is
very small. Therefore, the flow recovers to 2D at about
Z=-0.3 m. For Fr=0.37, the wave effects become strong,
extend to about Z=-1 m, and the separation in free
surface area starts at about X/L=0.32.
For Fr=0.55, the wave effects become even stronger and
extend to about Z=-1.15 m. The separation region is
smaller than that of Fr=0.37, and the separation in free
surface area occurs at about X/L=0.56.
Fig. 16 presents the contours of X component of the wall
shear stress on the circular cylinder. For Fr=0.19, the
location of the separation point in free surface area is at
about X/L=0.86, and the separation point at large depths
is located at about X/L=0.67. Therefore the separation
pattern is dominated by the shape effects, and the free
surface wave effects only delay the separation.

© 2005: Royal Institution of Naval Architects


Figure 14: Wave Patterns of the NACA0024 Foil and the Circular Cylinder
For Fr=0.37 and Fr=0.55, the separation points in free
surface areas are located at about X/L=0.4 and X/L=0.64,
respectively, which are very different from large depths,
and close to the NACA0024 hydrofoil. This suggests that
at these high Froude numbers, the separation pattern is
Froude dependent, and free surface wave induced
separation is dominant. The shape effects also slightly
delay the separation point.
At all Froude numbers, the wave effects for the circular
cylinder extend to very larger depths than for the foil.
This indicates that the shape effects are more significant
on the depthwise extent of the separation region, than on
its streamwise extent, especially at high Fr.
At very high Froude numbers, the flow becomes
unsteady, and the waves arise and break down
periodically. Fig. 17 shows the free surface waves for the
NACA0024 foil at Fr=1. The phenomenon is extremely
complicated due to the effects of unsteadiness,
turbulence, and air trapping. The present simulation was
able to cope with these difficulties; nevertheless further
Marine CFD 2005: Southampton, UK
developments seem to be needed to explain the details of
the problem.
5. CONCLUSIONS
The free surface flow is calculated for surface piercing
bodies of NACA0024 foil and circular cylinder, over a
range of Froude numbers. Flow results are presented and
analyzed with regard to the wave and viscous flow,
nature of the separation, and drag coefficients.
The study indicates that the drag coefficients, the free
surface waves, and the separation patterns are all Froude
dependent. The bow wave peak increases with Fr and the
separation region increases as Fr increases from small
(0.19) to medium (0.37), and then decreases as Fr
increases further to high (0.55). Associated with the
wave pattern, the pressure drag coefficient increases with
Fr. The frictional drag coefficient increases and then
decreases with Fr, in agreement with the separation
region.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

Figure 15: X-Wall Shear Stress Contours on the Figure 16: X-Wall Shear Stress Contours on the Circular
NACA0024 Foil Cylinder

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

of its streamwise extent, is highly affected by the shape

effects. Moreover, the distortions of the wave pattern in
the separation and wake region are shape dependent.

For very high Fr, the wave breaking occurs, and the flow
becomes extremely unsteady. As a result, numerical
simulation is possible only with robust free surface and
turbulence modeling techniques, in addition to fine grids.
The present modeling was able to solve the wave
breaking flow, but further numerical and experimental
investigations are recommended for identifying detailed
flow features.

The present numerical model is useful both in taking

insight into the complicated problem of free surface
wave induced separation, and in providing an implement
of design and optimization for ocean engineering
applications. The next steps are to evaluate the
performance of turbulence models, study the effects of
Reynolds number, and assess the critical Fr, at which the
wave breaking begins.

6. REFERENCES

1. CHOI, J.E. and STERN, F., “Solid-Fluid

Figure 17: Wave Breaking of NACA0024 Foil at Fr=1
This study also proves that the effects of air on the
accuracy of the numerical modeling are very significant,
especially for medium and high Fr.
At very low Fr, wave induced separation is not strong,
and shape effects dominate the flow and separation
regime.
At higher Fr, the wave effects become very significant,
and govern the primary pattern of the separation.
However, the depthwise extent of the separation, despite
Juncture Boundary Layer and Wake with
Waves”, Proc. 6th International Conference on
Numerical Ship Hydrodynamics, Iowa City, IA,
pp. 215-238, August 1993.
2. CHOW, S.K., “Free-Surface Effects on
Boundary-Layer Separation on Vertical Struts”,
PhD Thesis, University of Iowa, June 1967.
3. KANDASAMY, M., “RANS Simulation of
Free-Surface Wave-Induced Separation on a
Surface–Piercing NACA0024 Hydrofoil”,
University of Iowa Thesis, Iowa City, IA, 2001.
4. METCALF, B., “Experimental investigations of
Wave-Induced Separation on a Surface-Piercing
NACA0024 Hydrofoil”, University of Iowa
Thesis, Iowa City, IA, 2001.
5. POGOZELSKI, E., KATZ, J. and HUANG, T.,
“The Flow Structure around a Surface-Piercing
Strut”, Physics of Fluids, Vol. 9, No. 5, pp.
1387-1399, 1997.
6. STERN, F., HWANG, W.S. and JAW, S.Y.,
“Effects of Waves on the Boundary Layer of a
Surface–Piercing Flat Plate: Experiment and
Theory”, Journal of Ship Research, Vol. 33, No.
1, pp. 63-80, 1989.
7. ZHANG, Z., “Wave-Induced Separation”, M.S.
Thesis, University of Iowa, August 1995.
8. ZHANG, Z. and STERN, F., “Free-Surface
Wave-induced Separation”, Journal of Fluids
Engineering, Vol. 118, pp. 546-554, 1996.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

BASIN – DEVELOPMENT OF A PRACTICAL BOUNDARY ELEMENT CODE FOR

HYDRODYNAMIC ANALYSIS
N R Southall and B J Corlett, Burness Corlett - Three Quays (IOM) Ltd., UK

SUMMARY

This paper describes the development of a time domain potential flow code which was conceived as a tool for the
analysis of general hydrodynamics problems, with a particular emphasis on large amplitude ship motions. The code,
named BASIN (Boundary-element Analysis for Seakeeping Investigation), has been developed in house by Burness
Corlett - Three Quays. A primary aim of the project was to ensure its effectiveness in solving practical naval
architectural problems. A source/doublet boundary element method is used in conjunction with Euler-Lagrange time
stepping to solve the fully non-linear free surface equations.

Robust dynamic meshing plays a critical part in the success of the method. A NURBS-based definition of the geometry
of each body and the free-surface is used, and each body and the free surface are re-meshed up to the dynamic waterline
at every time step.

In addition to describing the background to the development of the code, the authors present some very encouraging
validations of the code against standard data, including steady state wave making, heave decay of a displaced spheroid
and large-amplitude heave and pitch motions in regular and irregular head seas for the S-175 containership.

NOMENCLATURE 1. INTRODUCTION

x,y,z Cartesian coordinates Accurate prediction of six degree-of-freedom ship

Φ Velocity potential motions in heavy seas without resorting to model testing
Φi Inner potential has always been a difficult problem for the naval
Φ∞ Free stream potential architect. Strip-theory codes are widely used in the
σ Source density industry for prediction of ship motions and loads, but
µ Doublet density whilst being very useful and computationally
n Unit normal to the surface undemanding, they have inherent limitations that restrict
r Position vector their usefulness. They are based on linearising
S Surface of the bodies, tanks etc. assumptions, although various methods have been
W Surface of the wake(s) devised to analyse non-linear problems with varying
v Perturbation velocity degrees of success. Their applicability to problems with
vs Velocity of a point on a body large amplitude motions, where slamming and green
vr Flow velocity relative to the body water effects become significant, is questionable and
vb Velocity of the body reference system problems can also occur for high Froude numbers and
ω Angular velocity of the body about an axis in ships with significant bow flare. The method is only
the body reference frame strictly valid for long, slender vessels, and roll damping
vn Normal component of panel perturbation must be linearised.
velocity
vt Tangential component of panel perturbation The time domain strip theory FREDYN [1,2] is an
velocity example of one of the more sophisticated codes; it uses a
P, Pref Static pressure, Atmospheric pressure blended method combining non-linear hydrostatics and
ρ Water density Froude-Krylov exciting forces with linear radiation-
g Acceleration due to gravity diffraction forces computed using conformal mapping. A
V Prescribed velocity for free surface vertices correction is made to the linear components to account
t Time for large amplitude motions. Other effects are modelled
η(x,y,t) Free surface elevation by the inclusion of various analytic and empirical
n Total number of panels components. The use of conformal mapping restricts the
NH Number of body panels range of sectional shapes that can be analysed.
NFS Number of free surface panels
NW Number of wake panels 3D Diffraction theory is used extensively in the offshore
λ Wavelength industry, but generally has limited applications for ship
L Vessel length between perpendiculars seakeeping studies due to the inherent restrictions on
k Wave number forward speed.
a Wave amplitude
za Heave amplitude
θa Pitch amplitude

© 2005: Royal Institution of Naval Architects


Several researchers have developed non-linear time
domain methods with encouraging results. The LAMP-4
code developed by Lin et al. [3] is a fully non-linear
method for prediction of six degree-of-freedom ship
motions and loads. It uses a split-domain method
separated by a matching surface, with a Rankine
singularity method used in the inner domain, and a linear
solution used in the outer domain. It assumes that waves
due to radiation/diffraction effects are small compared to
the incident waves. Some problems have been
encountered with the hull coming into contact with the
matching surface during analysis in oblique seas. The
UMDELTA method developed by Beck et al. [4-7] at the
University of Michigan uses a desingularised source
panel approach and solves the non-linear problem using
an Euler/Lagrange approach. This code produces
excellent results but it is difficult to apply the method to
complex geometries. Maskew [8-10] also uses an
Euler/Lagrange method but uses a source/doublet panel
method to solve the Euler boundary value problem. This
appears to be a good approach to the problem but it does
not seem to have had its true potential realised in this
code.
Reynolds-Averaged Navier-Stokes (RANS) codes have
become increasingly popular in the marine industry in
recent years. Commercially-available RANS codes
initially conceived for other industries have been
developed to be more useful for marine problems, with
treatment of free surfaces becoming much more
sophisticated. With the codes available today, excellent
results can be obtained for steady state problems such as
prediction of ship wave resistance and time-varying
problems such as sloshing in tanks, but the computational
requirements for a full seakeeping analysis in irregular
seas are enormous and huge advances in computational
power will be required before these codes can be used by
Naval Architects as a practical everyday tool for this
application.
The BASIN (Boundary-element Analysis for Seakeeping
Investigation) code currently under development by
Burness Corlett - Three Quays was conceived as a
practical tool for use in the analysis of real problems in
realistic timeframes. BASIN uses a similar boundary-
element formulation to that used by Maskew, but
includes more accurate calculation of some of the critical
components, using methods derived from work in Beck’s
UMDELTA code. The formulation is fully non-linear
and is applicable to a wide range of hydrodynamic
problems.
The code operates in the time domain by solving the
static problem at discrete time steps with unsteady
boundary conditions. The complete six degree-of-
freedom motion history of the body can be obtained by a
direct integration of the pressure distribution over the
body. Viscous models can be added to provide estimates
of viscous damping effects. As the full pressure field
over the body is calculated, the method is particularly
Marine CFD 2005: Southampton, UK
suitable for generating load cases for finite element
structural analysis.
The method is computationally demanding, but it is
possible to run useful problems in a realistic timeframe
on a fast PC. It is particularly useful for investigation of
episodic events involving large loadings and motions
experiencing rapid change in the time domain that cannot
realistically be investigated computationally by any other
means.
2. BUSINESS JUSTIFICATION
Burness Corlett - Three Quays is part of a broad based
marine consultancy group. We operate in a highly
knowledge based market where much of the knowledge
necessary for the success of the business depends on the
long term technical and commercial experience of the
staff in the marine industry. The training and background
of our staff are fundamental to this and to a large extent
this capability is developed and maintained in the course
of ongoing consultancy work carried out by the various
units. In addition to that, the business has always
cultivated a measure of higher level technical knowledge
and capability that has enabled it to sell more analytical
or research orientated consultancy to a wide range of
clients. This is worthwhile in its own right and has
important cross fertilisation benefits for the more general
business as well. In the modern world that means having
access to high level and cost effective analytical tools.
Group companies carry out various activities including
design of a wide range of vessel types, performance
studies, casualty investigations, quantitative risk analysis
and other technical studies. Many of the activities
involve hydrodynamic analysis in one form or another
including resistance prediction, hull form optimisation,
ship motion investigation, manoeuvring analysis and
optimisation, ship-ship and ship-bank interaction
investigation, etc. To carry out this type of work the
naval architects involved need to know about and
understand the principles involved and they need to have
available flexible, reliable, practical and accessible
analytical tools. The tools have to be capable of being
used by naval architects who are generally
knowledgeable but who are not narrow specialists and
will be involved in other technical activities for much of
the time.
We currently use various programs for sea keeping
prediction including strip theory and 3D diffraction
theory methods but we are very aware of the limitations
of them. Aspects of our sea keeping work involve
investigation of very large motions and the accelerations
and loads arising from them including the effects of
slamming and green water. Empirical and semi-empirical
methods are available for dealing with this type of
problem and there are various analytical codes being
developed which can start to address them. In practical
terms we have taken the view that there is nothing
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

available to us in the market place which meets the full
range of requirements. At present the RANS type codes
are too heavyweight for real sea keeping analysis and
other non-linear time domain codes are either in-house,
too expensive or too restricted for our purposes.
method renders it unsuitable for analysis of problems
where circulation is highly significant, such as analysis
of vortex shedding from tubular elements.
3.2 GOVERNING EQUATIONS

After careful consideration of the scale of the project we Assuming incompressible, inviscid and irrotational flow,
decided to develop our own in-house software package. the fluid motion can be described by a velocity potential,
Over a period of many years we have developed and Φ, which satisfies Laplace’s equation:
maintained our own software for intact and damage
stability, longitudinal strength, strip theory ship motions, ∇2 Φ = 0 (1)
manoeuvring simulation, etc. so that we are very aware
of the potential problems and pitfalls. We also know that Following Green’s theorem, the general solution to (1)
these can be offset by the knowledge gained during the can be constructed by a sum of source and doublet
software development process and the benefits that distributions placed on each boundary S
owning and intimately understanding the code provides.
In-house development gives us the flexibility to identify
1 1
the initial priorities, in this case to generate a well
validated large motion sea keeping application, to update
Φ( x, y, z ) =
4π ∫ µn ⋅ ∇ r dS +
S

and extend the software capabilities in future as required 1 1

and to integrate the code with our other software
4π ∫ σ  r dS + Φ
S
∞

packages, whether third party or in-house. We can also

set the development timetable, both to fit in with other
commitments and to spread the budget appropriately. This must be modified to allow inclusion of wake effects:

The panel code development which is described in this 1 1 1 1

paper meets our basic requirements in that the scale of Φ( x, y, z ) =
4π S ∫
µ n ⋅ ∇ dS +
r
σ  dS +
4π S  r  ∫
the project is within reasonable bounds for an
1 1
organisation of our size, the method has few basic
compromises, it should be capable of running on 4π W ∫
µ n ⋅ ∇ dW + Φ ∞
r
relatively straightforward hardware and it offers the
possibility of extension into a wide range of The first integral is the contribution over the boundaries
hydrodynamic application and problems. These include: from a surface distribution of normal doublets of strength

• Resistance optimisation studies, µ =Φ (2)

• Design for minimisation of wave making,
• Added resistance in waves, The second integral is the contribution from a surface
• Motions and loads in waves including green water distribution of sources of strength
loading,
• Probability of deck wetness, slamming, propeller σ = n.∇ Φ (3)
emergence, etc.,
• Motions when moored, e.g. at SPMs or for FPSOs, The third integral is the contribution from any wake
• Interaction between vessels, e.g. RAS or STS, or surfaces included in the problem. The wake surfaces
between vessels and banks, carry a doublet distribution only, with strengths set by
• Squat prediction, use of a Kutta condition implemented along wake-
• Manoeuvring prediction and numerical PMM shedding lines.
(Planar Motion Mechanism)
• Manoeuvring in waves Use of the Dirichlet boundary condition requires that the
perturbation potential Φ has to be specified everywhere
3. MATHEMATICAL MODEL on the body surface S. By placing the point (x, y, z)
inside the body the inner potential Φi is obtained:
3.1 ASSUMPTIONS
1 1 1 1
The main assumptions of Potential Flow methods are that Φ i ( x, y , z ) =
4π ∫ µn ⋅ ∇ r dS + 4π ∫ σ  r dS +
S S
the fluid is incompressible and that away from the (4)
boundaries, flow is considered inviscid and irrotational. 1 1
Wake effects can be included by expelling wave panels 4π ∫W
µ n ⋅ ∇ dW + Φ ∞
r
from specific separation lines and a simple viscous
model based on calculated streamlines can be included
[8]. However, the boundary-element based nature of the

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

The Dirichlet condition gives: P − Pref 1 ∂Φ

=− ∇ Φ⋅ ∇ Φ − gz − (11)
ρ 2 ∂t
Φ i = (Φ + Φ ∞ ) i = Φ ∞ (5)
where z is the free surface elevation. Taking the reference
Substituting (5) into (4) gives the basic boundary integral pressure, Pref, as the atmospheric pressure and assuming
problem: that the pressure at the free surface is uniform, we obtain:

1 1 1 1 ∂Φ
∫ µn ⋅ ∇ r dS + 4π ∫ σ  r dS +
1
= − ∇ Φ⋅ ∇ Φ− gz (12)
4π S S
∂t 2
(6)
1 1
4π ∫W
µ n ⋅ ∇ dW = 0
r
The free surface elevation is updated using the kinematic
condition by moving the free surface nodes with the local
flow:
3.3 BOUNDARY CONDITIONS ∂r
=∇Φ
On each body, the source distribution is determined by ∂t
specifying zero-flow normal to the boundary. The flow Or alternatively:
velocity relative to the surface is:
∂η ∂ Φ
= − ∇ Φ .∇η (13)
vr = v − vs ∂t ∂z

where v is the perturbation velocity in the global where z = η ( x, y, t ) is the free surface elevation.
reference frame, which is given by the gradient of the
potential: It is convenient to rewrite the free surface boundary
conditions in terms of the time derivative of a point
v =∇Φ (7) moving with a prescribed velocity V relative to the
global origin. This allows the horizontal movement of the
vs is the velocity relative to the global reference frame: free surface nodes to be restricted to prevent the piling up
of nodes. The approach is described in [4].
vs = vb + ω × r (8) The Bernoulli condition becomes:

where vb is the velocity of the body relative to the global δΦ 1

= − gη − ∇ Φ .∇ Φ + V.∇ Φ (14)
reference frame, ω is the velocity of rotation of the body δt 2
about an axis in the body reference frame and r is the
position vector relative to that axis. and the Kinematic condition becomes:

The condition of zero-flow normal to the boundary gives: δη ∂ Φ

= − (∇ Φ − V ).∇η (15)
δt ∂z
n ⋅ v r = n ⋅ (v − v s ) = 0 (9)
where:
Using Equations (3) and (7-9), the source distribution on δ ∂
the body is given by: ≡ + V.∇
δt ∂t
σ = n ⋅ v s = n ⋅ (v b + ω × r ) (10)
is the time derivative following the moving node. Setting
 ∂η 
On the free surface, the mixed Eulerian/Lagrangian V =  0,0,  results in the horizontal locations of the
approach of Longuet-Higgins and Cokelet [11] is used.  ∂t 
This involves two major tasks at each time step: the nodes remaining fixed in the global coordinate system.
problem is solved in an Eulerian frame, then the fully
non-linear boundary conditions are used to track 3.4 HYDRODYNAMIC FORCES
individual Lagrangian points on the free surface to find
their new positions and potentials. The initial conditions The forces and moments acting on each body are found
assume that the free surface is undisturbed and so the by integrating the pressure distribution over the body.
source and doublet terms are both set to zero. The free The pressure is given by Bernoulli’s equation (11).
surface velocity potential (i.e. the doublet strength) is However, this is the pressure at a stationary point. At a
updated using the Bernoulli condition: point moving with speed vs this becomes:

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

P − Pref δΦ NH NH + NFS

∑µ A ∑σ B
1
=− ∇ Φ⋅ ∇ Φ − gz − + v s ⋅ ∇ Φ (16) + =
ρ δt
i ij i ij
2 i =1 i = NH +1

δΦ
Accurate evaluation of the
δt
term is critical for
computation of the hydrodynamic forces. Typically, this
is calculated using backward differencing but for large-
amplitude free body motions this approach is not
δΦ
sufficiently accurate. The term must be calculated
δt
directly and an approach similar to that described in [4]
is used to accomplish this, where the problem is
δΦ
reformulated to solve for instead of Φ.
δt
4. IMPLEMENTATION
4.1 PROBLEM SETUP
The first task is to import one or more geometries into
the program describing each body to be analysed. This
must be in the form of a surface definition and at present
the IGES and STEP file formats are supported. Various
other run parameters must be specified at this stage,
including the extent of the free surface, mass and inertia
properties of each body, Froude numbers, panel densities
for each body and the free surface, parameters for the
wavemakers for seakeeping problems etc. Each body has
its own reference system and items such as rudders etc.
can be included and moved independently. The bodies
are defined by one or more NURBS (Non-Uniform
Rational B-Spline) surfaces and the free surface is also
described by a NURBS surface created by the software.
Intersections between the free surface and each body are
found and the initial mesh is created automatically on the
hull and free surface, using quadrilateral panels where
possible. The panels are oriented so that their normal
vectors point into the fluid.
4.2 MATRIX FORMATION
At each time step, the procedure is as follows. The
source strengths on each body panel are calculated from
the zero normal flow condition (3). The doublet strengths
on the free surface are set to zero at the first time step.
The surface integrals in (6) are performed on a panel-by-
panel basis assuming uniform source and doublet
distributions on each panel. The influence of each panel
is found at the centroid of each other panel, forming a set
of n simultaneous equations, where n is the total number
of panels used at that time step. The unknowns (doublet
strength on each body panel, source strength on each free
surface panel) are collected on one side of the equation
and the known quantities are summed to form the right
hand side, giving the main problem formulation:
 NH NH + NFS NW

 i =1
∑
−  σ i Bij +
i = NH +1
∑
µ i Aij + ∑µ
i =1
Wi
C ij 

where NH = Number of hull panels, NFS = Number of
free surface panels, NW = Number of wake panels. Aij, Bij
& Cij are the influence coefficients for the doublet,
source and wake doublet respectively. This can be
written in matrix-vector form as:
M LHS ⋅ VLHS = −VRHS
where MLHS is the n-by-n matrix of doublet and source
influence coefficients, VLHS is the size-n vector of
unknown source and doublet strengths and VRHS is the
size-n vector containing the summations of known
quantities.
4.3 MATRIX SOLUTION
BASIN has a range of solvers to calculate the unknown
source and doublet strengths, using either the Intel Math
Kernel Libraries (MKLs) or the FASTLAP Multipole
method [12] for efficient solving. For a single matrix
solution, the Multipole method is considerably faster
than the MKL solvers once the problem size exceeds
around 3000 panels, but for problems involving free
body motions, multiple calculations are required during
each time step meaning that the Multipole solver only
really provides a benefit for 5000 panels and over. The
Multipole method is always used for problems which
cannot use a plane of symmetry, as the memory
requirement using a conventional solver soon becomes
prohibitive.
If the MKL direct solver is used, the inverse of the
influence matrix MLHS can be calculated and used on
δΦ
each iteration for the term in the pressure equation,
δt
whereas when the Multipole method is used, the full
calculation must be carried out each time.
4.4 PANEL VELOCITIES
Once the source and doublet strengths are known, the
perturbation velocity must be evaluated on each panel.
This has two component parts – a normal component vn
provided by the source strength, and a tangential
component vt found from the surface gradient of the
doublet strength:
v = v n n + v t = σ n + ∇µ (17)
The doublet gradient is calculated in two directions (the
panel’s local x and y directions) over each panel using
second-order differencing.

© 2005: Royal Institution of Naval Architects


On the body, the perturbation velocity is combined with
the local velocity vs to give the resultant velocity:
vr = v − vs
4.5 FREE SURFACE MOVEMENT
The free surface panel velocities are interpolated to the
panel nodes and the nodes are moved using a simple
Euler scheme:
 δη 
r ( t + ∇t ) = r ( t ) + ∇ t  
 δt 
The doublet strengths on the free surface panels are
updated in a similar way, using the results of (14):
 δΦ 
µ ( t + ∇t ) = µ ( t ) + ∇ t  
 δt 
4.6 WAVE GENERATION AND ABSORPTION
For seakeeping problems, it is necessary to include a
wave generator upstream of the vessel. Various wave
generation methods have been evaluated for the code,
culminating in the present method. This uses pulsating
source terms on tank walls for wave generation,
combined with absorbing wavemaker conditions on the
free surface panels in the vicinity of the wave generators.
The method is attributed to Cointe et al. and is described
in Tanizawa [13]. The damping zones on the free surface
absorb the differences in potential and wave amplitude
between specified reference values and the actual values
found from the free surface boundary conditions. In the
absorbing regions in the vicinity of the wave generators,
the reference values are found from analytical solutions
[14]. In the regions used as absorbing beaches, the
reference values are set to zero. Tanizawa found that the
reflection coefficient from the damping zone for deep
water waves was less than 2%.
Because the method transforms the wave generators into
absorbing wavemakers, the method is valid for low-
speed or zero-speed problems. For problems with
forward speed, the beaches track the vessel. For simple
wave resistance calculations, open boundaries can be
used at the sides of the free surface, but it is still sensible
to have upstream and downstream boundaries, especially
for low-speed runs. The disadvantage of the method is
that the damping zones are large and so have a great
effect on the problem size. Tanizawa states that the
damping zones must be at least as long as the wavelength
for regular wave problems, although for irregular wave
analysis we have found that it is sufficient to set the
damping zone length equivalent to the mean wavelength.
Figure 1 shows a typical arrangement of damping zones
for a seakeeping analysis without a plane of symmetry –
here the damping zones extend for one vessel length
Marine CFD 2005: Southampton, UK
from each boundary of the free surface. This particular
arrangement is a good illustration of the computational
overhead required for the beaches – 75% of the free
surface panels lie in the damping zones. It is possible to
reduce the panel density in the zones acting as simple
absorbers, but not unfortunately in the wavemaking
zones as this would result in attenuation of the higher
frequency waves.
Figure 1: Typical damping zones for a six-degree-of-
freedom seakeeping analysis
For long-crested head seas problems the vessel is run in a
‘tank’ instead – the tank walls prevent transverse
dissipation of the generated waves and allow the use of a
narrower free surface. A wave absorber is still needed at
the downstream edge for low/zero speed problems.
4.7 BODY FORCES AND MOMENTS
The pressure acting on each panel is found from (16) and
integrated over the body to find the forces and moments
acting upon it. The difficulty for free body motions
δΦ
problems is that for an accurate solution, the term
δt
∂ δ Φ
must be calculated directly, and the   term
∂n  δt 
required for this calculation requires that the velocity and
acceleration of the body is known. The approach used
here is to estimate the velocity and acceleration (using
the values from the previous time step), calculate the
δΦ
term and use this in (16) to find the pressure
δt
distribution over the body. Forces and moments are
calculated and used in the equations of motion to
calculate the new velocity and acceleration terms. The
process is then repeated until the values stabilise, which
typically takes 5 iterations.
For problems where a vessel is free to roll, a viscous roll
damping factor must be included in the equations of
motion. At the present time, this simply takes the form of
an equivalent linear damping coefficient predicted using
the method described in Himeno [15] and Ikeda [16].
This will be updated in the future to a full non-linear
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

damping correction. The damping components that are The distribution of doublet strengths on the old free
inherent to the potential flow model (e.g. wave damping) surface mesh is transferred to the new mesh by
must be excluded during the calculation of the additional interpolation and the old mesh is discarded. The new
damping term. More sophisticated prediction of the mesh is then knitted into the mesh outside of the re-
various wave damping components is possible – for meshed region ready for the next time step cycle.
example, vortex shedding from bilge keels can be
modelled by shedding wake panels from the bilge keels.
Panel methods have been used to predict tightly coiled
wake patterns behind slowly heaving airfoils [17] and we
are developing a method to apply this technique to bilge
keel vortices.

Once the body velocity has been calculated, each body is

moved in all of the degrees of freedom that have been
designated as ‘free’ – full six degree-of-freedom
integration can be used for non-symmetric problems.

4.8 MESHING

At each time step, once each body has been moved to its
new position, both they and the free surface must be re-
meshed to the new waterline(s). Implementing a method
for re-meshing is probably the most critical and difficult
task during the development of a code of this nature – it
must be robust, fast, run with no user-intervention and
the resulting mesh must be of sufficiently high quality to
ensure that the numerical method is not affected. The
total number of panels is very dynamic – there are often
entire surfaces in the description of each body which are
dry when the body is at rest, but which may completely
submerge at some stage during an analysis.

For each wetted patch of each surface making up the

geometrical definition of each body, a distribution of
points is found around the boundary of the patch. These
points are transformed to the 2D parametric space of the
surface to calculate a mesh of grid points inside the patch
boundary, and the resulting grid is transposed to 3D
space to form the new mesh for that patch.

The major tasks during the re-meshing process are shown

in Figure 2. We begin with the free surface panels which
have been moved according to the velocities calculated at
that time step. In the next frame, a new NURBS surface
has been fitted through the free surface nodes in the
vicinity of the body, and the body has been moved to its
new position for the next time step. There is deck edge
immersion at this time step, and the NURBS surface
extends over part of the deck. The third frame shows the
lines of intersection which have been found between the
body surfaces and the free surface, with the old body
mesh still displayed to illustrate the change in the wetted
area. The final frame shows the new meshes on the body
and free surface. The re-meshing is a significant
computational task at present but the performance will be
greatly improved after some optimisation of the method. Figure 2: Four stages in the procedure to generate a new
mesh on the body and free surface

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

5. RESULTS

A wide range of verification and validation exercises

have been carried out using the code, with comparisons
made both with public-domain and in-house data. Some
examples of the results from these exercises are shown
below.

5.1 SERIES 60 WAVEMAKING

Although the code is primarily focussed on seakeeping

analysis, the method is applicable to problems of
prediction of steady state wavemaking. The Series 60
Cb=0.6 hullform was run at fixed draft, sinkage and trim
at a Froude number of 0.316 and compared to data from
the Iowa Institute of Hydraulic Research (IIHR) [18].
The experimental results obtained by the IIHR are
extensive and include an 8000-point wave height survey.

The wave profile along the hull is shown in Figure 3. The

agreement with the experimental results is reasonable,
although there is a discrepancy in the shape of the bow
wave. This is probably due to insufficient mesh density
in this region; the mesh used here was more typical of
that used for a seakeeping problem and is too coarse to
resolve the region of high curvature in the free surface at
the bow.

Figure 3: Wave profile for Series 60 at Fr. 0.316

The wave contour plot (viewed from above the vessel) is

shown in Figure 4. The IIHR results are shown on the
port side and the BASIN results are shown on the
starboard side. Most of the major features of the wave
pattern are predicted well, although the complex system
of small waves and troughs are smoothed by the coarse
mesh used. The predicted peak just aft of the stern is not
as cusped as the experimental result, and the under-
prediction of the peak at the bow identified in the wave
profile is also obvious.

Figure 4: Wave contours for the Series 60 at Fr. 0.316

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

The results presented in Figure 6 correspond to O’Dea’s

5.2 MOTIONS OF A FLOATING SPHERE FREE
higher speed runs at a Froude number of 0.3 – this
TO HEAVE
equates to approximately 30 knots for this vessel. Heave
A standard test case for ship motions codes is the results are non-dimensionalised by wave amplitude and
decaying heave motions of a freely floating sphere. At wavelength by L. Pitch results are non-dimensionalised
equilibrium, the sphere’s centroid lies in the plane of the by (wave number * wave amplitude).
undisturbed free surface. The sphere is given an initial
upwards displacement of 0.5r where r is the sphere
radius and is then released. Results are compared to
experimental results from Liapis, published in Maskew
[10]. The heave decay over time is shown in Figure 5.
Results compare very well with the experimental results
by Liapis for the first three oscillations. The result from
Maskew’s USAERO code experiences excessive
damping.

Figure 5: Motions of a floating sphere, radius r, free to

heave after an initial displacement Ah

5.3 SL7 CONTAINER SHIP

Results for free-body motions in regular head seas were

generated for the SL-7 containership. The principal
particulars of the vessel are shown in Table 1.

Length between perpendiculars (L) 268.4 m

Breadth 32.2 m

Draft 10.4 m

LCB (LCB/L fwd AP) 0.472

Long. radius of gyration (assumed value) 0.25 L

CB 0.526
Table 1: SL7 Principal Particulars

The vessel was run in regular waves, head seas for a

range of wavelengths. Wave slope was kept constant, and
wave height was set at (wavelength/50).
Runs were made using a moving free surface, moving
approximately with the vessel. Results for magnitude of
heave & pitch oscillation and relative motions at the FP
are compared to the experimental results published in
O’Dea [19].
Figure 6: Head seas motions for the SL7, Fr. 0.3
The heave results are slightly under predicted throughout.
The under-prediction is most significant at shorter
wavelengths, which may be due to numerical damping of
the generated waves. At the shorter wavelengths, there
are far fewer nodes over the length of a wave for a given
panel density and so the damping effect of the node

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

velocity calculations becomes significant. This can be
alleviated by use of a finer panel density. Pitch results are
in good agreement except at λ/L = 2, where the pitch is
under-predicted.
Relative motions at the forward perpendicular are also
shown in Figure 6. The results shown the correct trend
although as for the heave results, the relative motions are
under-predicted.
The heave results are shown in Figure 7, and the pitch
results in Figure 8. The heave results for λ/L = 1.0 are in
good agreement with the experimental results of O’Dea
up to ka = 0.12. At this point, the effects of green water
are becoming significant and are not correctly modelled
by the present method. The effect of a large volume of
green water on the foredeck would be to reduce the
upward heave, thus giving a smaller response.

5.4 S175 CONTAINER SHIP – REGULAR

WAVES

The S175 containership was selected by the ITTC as a

standard test case and so a large amount of test data is
available. The principal particulars of the vessel are
shown in Table 2.

Length between perpendiculars (L) 175 m

Breadth 25.4 m

Depth 15.4 m

Draft 9.5 m

Displacement 24742 tonnes

LCG aft midships 2.5 m

Longitudinal radius of gyration 0.236 L

CB 0.572
Table 2: S175 Principal Particulars

The problem was approached in a similar manner to that

used for the SL-7. This time, however, wave slope was
not held constant for each run. Heave and pitch transfer
functions were found for a range of wavelengths (1.0L –
1.4L) and wave heights. All runs were at a Froude
number of 0.275 and the maximum wave height used
was just over 9m. Experimental results are by O’Dea and
were taken from Wang [20].

Green water effects will have a significant effect on some

of the runs at larger wave heights, and these effects are
not correctly modelled in these simulations. The motions
are fairly large – for example the pitch magnitude for the
λ = 1.4L, ka = 0.12 run was almost 17 degrees. The
foredeck was wet aft of 1/3L back from the bow during
the large wave amplitude runs, and because the effect of
green water is not modelled some over-prediction is
experienced. Figure 7: Heave results for the S175 in regular waves
Each analysis took approximately six hours to run on a The pitch results are in good agreement, although there is
Pentium IV 2.53Ghz PC with 1Gb RAM. A line of some under prediction at shorter wavelengths.
symmetry was used to reduce the problem size.
Approximately 2000 panels were used although this Both the heave and pitch results for λ/L = 1.2 are in good
number varied during the run due to the dynamic agreement with the experimental results up ka = 0.08.
meshing. 320 time steps were used corresponding to ten
wave encounters for the λ = L case.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK
The results for ka = 0.12 show an over-prediction both of
heave and of pitch.
Figure 8: Pitch results for the S175 in regular waves
At λ/L = 1.4, both results are in good agreement,
although at the larger wave height (ka = 0.12) the pitch is
over-predicted. This is to be expected as the heave result
collapses to a direct response of 1.0 for this longer
wavelength. The inaccuracy due to the modelling of
green water manifests itself in the pitch result in this case.
5.5 S175 CONTAINER SHIP – IRREGULAR
WAVES
Results for ship motions in regular waves are interesting,
but in most cases the Naval Architect wants to be able to
predict with reasonable accuracy the motions and loads
for a vessel in irregular waves. A problem arises when
© 2005: Royal Institution of Naval Architects
attempting a validation exercise of a time domain code in
irregular waves because most available validation data is
presented in the frequency domain. It is possible to use
BASIN to generate a sufficiently long sample (>100
wave encounters) so that reliable results for the response
spectra will be obtained, but making a direct comparison
for a known time history of wave amplitude and
responses is potentially more informative for validation
purposes. Correctly modelling an irregular spectrum
presents some difficulties. High frequency waves have a
tendency to ‘drop out’ as the waves travel along the
length of the free surface – because they are short, there
are insufficient panels to define them properly and there
is a tendency for them to experience some numerical
damping. If an appropriate panel density is used,
however, the time history can be recreated with sufficient
accuracy.
Fonseca [21] published results for the S175 vessel in
irregular head seas using Pierson-Moskowitz spectra of
significant wave height (Hs) 4.2m, 6.1m and 9.9m, and
we were able to obtain time histories for the 4.2m and
6.1m cases. The amplitudes and phases of the
components making up the spectrum as realised in the
tank were obtained using Fast Fourier Transforms, and
using these results it was possible to recreate the two
spectra in BASIN.
The results of the analysis are shown in Figures 9 and 10.
Fonseca presents pitch results which are positive bow
down, and we have kept to that convention here. Wave
amplitude was measured during the tests by a wave probe
located 140m (full scale) forward of the vessel centre of
gravity. The simulated waves are an excellent match to
the experimental results, with the small discrepancies at
some of the peaks almost certainly due to numerical
damping of the high frequency waves. The heave and
pitch results are generally in good agreement,
particularly with regard to phasing. For the 6.1m Hs
spectrum, the heave amplitudes are generally in good
agreement, but the response seems to take longer to settle
after a series of large waves are encountered (e.g. T =
170s – 180s, T = 285s – 295s) suggesting that the BASIN
calculation is under-damped. The pitch results show a
good agreement on the magnitudes of bow-down pitch,
but bow-up pitch is over-predicted. Some of this over-
prediction may be due to the effects of green water being
excluded from the BASIN calculations – there are
significant green water episodes in the region of the time
history from 200 to 270 seconds.
For the 4.2m Hs spectrum, the heave results exhibit a
noticeable trend of over-prediction of amplitude,
although the agreement of phasing is excellent. The pitch
results exhibit over-prediction of bow-up pitch, and there
are no green water episodes during this run. Overall,
however, the results are extremely encouraging.
 Marine CFD 2005: Southampton, UK

Figure 9: Time Histories for 6.1m Hs spectrum

Figure 10: Time Histories for 4.2m Hs spectrum

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Figure 11: Wave contours at four stages of a wave encounter during the 6.1m Hs simulation

© 2005: Royal Institution of Naval Architects


Figure 11 shows wave contour plots at four stages in a
wave encounter during the 6.1m Hs analysis. The first
image shows a stage with significant green water and the
last shows a stage with significant forefoot emergence,
with the two images in between showing intermediate
stages. The effects of radiation/diffraction from the
vessel can be clearly seen, as can the significant green
water boarding.
Each analysis took approximately 30 hours to run on a
dual Intel Xeon 3.2Ghz PC with 2Gb RAM. Almost half
of this time was taken up with re-meshing. A line of
symmetry was used to reduce the problem size.
Approximately 3500 panels were used although this
number varied during the run due to the dynamic
meshing. The simulation was run for 1000 time steps
corresponding to 338 seconds full scale.
If statistical results are required, the code can be used in
three ways. Obviously the code could simply be run for a
long enough period to generate a reliable sample – in the
case here, a run of three times this duration would be
sufficient as there would be over 100 wave encounters
during this period. A second approach would be to find
the extremes by statistical analysis of a relatively short
time history. Alternatively, the input spectrum can be
modified to hit the most likely extreme responses,
identified using linear methods [22].
6. CONCLUSIONS
Encouraging progress has been made in the development
of a method capable of predicting large amplitude ship
motions in waves. Preliminary results compare well with
experimental data, although the effects of green water
must be included for a more rigorous analysis of large-
amplitude motions. We have identified areas where we
can further improve the method, which will result in
better prediction of the results.
Although the calculations are computationally
demanding, it is possible to obtain useful results in a
realistic timeframe on a single PC. The code is largely
un-optimised at present and a large reduction in runtime
will be achievable after optimisation work is completed.
A significant part of the difficulty in developing such a
method is in areas such as re-meshing. The actual
theoretical background is relatively simple in comparison
with some other methods. The method combines
practicality with an open theoretical approach which
ensures that once the development is complete, it will be
possible to use BASIN for a huge range of applications.
The results presented here are free to heave and pitch
only, but the formulation is applicable to full six degree
of freedom problems. Work on the six degree of freedom
model is well advanced with some initial test runs carried
out. We feel that progress to date has justified the
decision to start the project, although unsurprisingly
Marine CFD 2005: Southampton, UK
development has taken longer than had been hoped. This
has been due in part to the limited resources available
and the demands of other projects and R&D work. We
hope to be running validation on the six degree of
freedom model in the next few months.
7. FUTURE WORK
Immediate work will focus on improvements to the
method so that better agreement can be obtained for the
simulations in irregular head seas. This will include the
addition of various methods to study the effects of green
water.
Development and testing of the six degree of freedom
model will be completed so that motions in oblique
irregular waves can be predicted. Amongst other things,
this will necessitate improvements to the roll damping
model and optimisation of the code to improve
performance, particularly during re-meshing.
Other planned developments include the ability to
analyse multihulls/multiple vessels, prediction of
manoeuvring in waves, ship motions in short-crested seas,
and prediction of bilge vortex shedding.
8. ACKNOWLEDGEMENTS
The authors would like to thank their colleagues at
Burness Corlett - Three Quays for their assistance with
the work in this paper, and we would especially like to
thank Nuno Fonseca at the Technical University of
Lisbon for providing the time histories used in the
irregular wave study.
9. REFERENCES
1. De KAT, J. O. and PAULLING, J. R., ‘The
Simulation of Ship Motions and Capsizing in Severe
Seas’, Transactions of SNAME, 1989
2. LIN, R-Q. and THOMAS, W., ‘Ship Stability Study in
the Coastal Region: New Coastal Wave Model Coupled
with a Dynamic Stability Model’, Proceedings of the
23rd Symposium on Naval Hydrodynamics, 2001
3. LIN, W. and YUE, D., ‘Numerical solution for large-
amplitude ship motions in the time domain’, Proceedings
of the 18th Symposium on Naval Hydrodynamics, Ann
Arbor, 1990
4. BECK, R., CAO, Y., SCORPIO, S. and SCHULTZ,
W., ‘Non-linear ship motion computations using the
desingularized method’, 20th Symposium on Naval
Hydrodynamics, Santa Barbara, California, 1994.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
5. SUBRAMANI, A.K., BECK, R.F., and SCORPIO,
S.M., ‘Fully Non-linear Free-Surface Computations for
Arbitrary and Complex Hull Forms’, Proceedings, 22nd
Symposium on Naval Hydrodynamics, Washington, D.C.,
1998.
6. BECK, R., CAO, Y. and LEE, T., ‘Fully Non-linear
Water Wave Computations using the Desingularized
Method’, Proceedings 6th International Conference on
Naval Hydrodynamics, University of Iowa, 1993.
7. SCORPIO, S., ‘Fully Non-linear Ship-Wave
Computations Using a Multipole Accelerated
Desingularized Method’, Ph.D. Thesis, Department of
Naval Architecture and Marine Engineering, University
of Michigan, Ann Arbor, 1997
8. MASKEW, B., ‘A Non-linear Numerical Method for
Transient Wave/Hull Problems on Arbitrary Vessels’,
Transactions of SNAME, 1991
9. MASKEW, B., ‘USAERO/FSP - A Time domain
Approach to Complex, Free Surface Problems’,
Symposium on High-Speed Marine Vehicles, Naples,
Italy, 1991
10. MASKEW, B., and TIDD, D., ‘Prediction of Non-
linear Wave Hull Interactions on Complex Vessels’, 6th
International Conference on Numerical Ship
Hydrodynamics, Iowa, 1993
11. LONGUET-HIGGINS, M., and COKELET, C., ‘The
Deformation of Steep Surface Waves on Water: I. A
Numerical Method of Computation’, Proceedings of the
Royal Society of London, Vol. A350, 1976.
12. KORSMEYER, T., YUE, D., NABORS, K. and
WHITE, J. 'Multipole-accelerated preconditioned
iterative methods for three-dimensional potential
problems’, Boundary Element Methods 15, Worcester,
MA, 1993
13. TANIZAWA, K., ‘The State of the Art on Numerical
Wave Tank’, Proc. of 4th Osaka Colloquium on
Seakeeping Performance of Ships, 2000.
14. DEAN, R. and DALRYMPLE, R. ‘Water Wave
Mechanics for Scientists and Engineers’, Advanced
Series on Ocean Engineering - Volume 2, World
Scientific, 1998.
15. HIMENO, Y., ‘Prediction of Ship Roll Damping –
State of the Art’, Report of NA & ME, The University of
Michigan, No.239, 1981.
16. IKEDA, Y., ‘Prediction Methods of Roll Damping of
Ships and Their Application to Determine Optimum
Stabilization Devices’, 6th International Ship Stability
Workshop, 2002
© 2005: Royal Institution of Naval Architects
17. KATZ, J., and PLOTKIN, A., ‘Low-speed
Aerodynamics - from Wing Theory to Panel Methods’,
McGraw-Hill, New York, 1991
18. TODA, Y., STERN, F., and LONGO, J., ‘Mean-Flow
Measurements in the Boundary Layer and Wake and
Wave Field of a Series 60 CB=0.6 Ship Model - Part 1:
Froude Numbers 0.16 and 0.316’, Journal of Ship
Research, Vol. 36, No. 4, pp. 360-377, 1992
19. O'DEA, J., ‘Relative Motion and Deck Wetness
Investigation of the SL-7 Containership’, David W.
Taylor Naval Ship R&D Center, Report SPD-1081-01,
1983
20. WANG, Z-H., ‘Hydroelastic Analysis of High-Speed
Ships’, PhD Thesis, Technical University of Denmark,
January 2000.
21. FONSECA, N. and GUEDES SOARES, C.,
‘Experimental Investigations of the Non-linear Effects on
the Statistics of Vertical Motions and Loads of a
Containership in Irregular Waves’, Journal of Ship
Research, Vol. 48, No. 2, 2004
22. ADEGEEST, L., BRAATHEN, A. and VADA, T.
‘Evaluation of Methods for Estimation of Extreme Non-
linear Ship Responses Based on Numerical Simulation
and Model Tests’, 22nd Symposium on Naval
Hydrodynamics, 2000
10. AUTHORS’ BIOGRAPHIES
Neil Southall currently holds a position of Consultant
Naval Architect at Burness Corlett - Three Quays. He has
an M.Eng. in Mechanical Engineering from Oxford
University and an M.Sc. in Yacht and Small Craft Design
from Southampton University. He is a Graduate Member
of RINA. At Burness Corlett - Three Quays he has been
involved in the execution of a wide range of sea keeping
investigations including the investigation of structural
response to sea loads. He is responsible for
hydrodynamic software development.
Brian Corlett currently holds the position of Managing
Director of Burness Corlett - Three Quays and is
technical director of the parent company, Oceanic
Investment Corporation Ltd. He has a B.Sc. in
Engineering Science from the University of Durham and
an M.Sc. in Marine Technology from the University of
Newcastle. He is a Fellow of RINA and a former
Member of Council. At Burness Corlett - Three Quays he
has overall responsibility for all technical work including
group responsibility for IT and software management and
development. He has been involved in many technical
studies and casualty investigations which have included
hydrodynamic analysis.
Marine CFD 2005: Southampton, UK

VOF-DYNAMIC MESH SIMULATIONS OF UNSTEADY SHIP HYDRODYNAMICS

M Visone M Eid, Blue Group Engineering & Design, Italy

P Bertetti and R Gandolfi, AZIMUT, Italy
C Falletta and P L Ausonio, Ship-Yacht Designers & Consultants, Italy
D Paterna and R Savino, University of Naples, Italy.

SUMMARY

The computation of the complex hydrodynamic and aerodynamic flows around motor boats is extremely challenging and
requires state-of-the-art numerical techniques and computer technology. In this paper, turbulent flow simulations around
the hull of a planning boat, at steady speed through calm water, have been performed to compute the hull attitude in
static and dynamic conditions. The presence of the water-air interface has been taken into account with a Volume-Of-
Fluid (VOF) technique, that allows to track the wave form at each time during the numerical simulation. Turbulence has
been modelled using the Reynolds Average Navier Stokes (RANS) equations and the k-ω model, with suitable wall
functions for near wall treatment. An “ad hoc” iterative computational procedure has been developed, based on a
dynamic mesh algorithm, to compute the time-dependent position and orientation of the hull at each time during
calculations.

Approximately one million hybrid tetrahedral/hexahedral cells have been used in the simulations. The parallel
computations have been carried out on a Linux cluster.

Validation of the computational simulation has been carried out, comparing the results of different model tests
performed at the Brodarski Institute in Zagreb. Two AZIMUT hulls have been taken into account: Hull-A, scale ratio
1:3.8, fully equipped with appendages, and a Hull-B, scale ratio 1:6.0, naked hull with spray rails. The lift, drag and
dinamic trim have been computed in the final equilibrium positions. The results have been compared with the
experimental data, showing a very promising agreement.

NOMENCLATURE most often prohibitive costs associated with extensive

L : computed total lift (N)
W : boat weight
D : computed total drag (N)
Dexp : total drag from experiments (N)
M: computed pitching moment (Nm)
Teq: thrust in equilibrium conditions (N)
θ : computed trim angle (deg)
θexp : trim value from experiments (deg)
ZG : computed dynamic vertical position of CG (m)
ZGexp: dynamic vertical position of CG from experiment
(m)
XTG : distance, in x direction, between T and CG (m)
ZTG : distance, in z direction, between T and CG (m)
α: thrust angle with respect to x axis (deg)
Hull-A : 15 mt overall length.
Hull-B : 18 mt overall length
model testing.
Very often the performance prediction of small planing
boats is performed on a semi-empirical basis, using e.g.
the Savistky method [1] with some literature or home-
made correction procedure, adapted to meet the full scale
performance of similar hulls. This is mainly due to the
fact that model tests are very expensive and time-
consuming. Moreover, for this type of hulls, the results
are sometimes unreliable, especially if performed on
small models, due to the Reynolds effects.
The numerical simulation could be of large benefit in this
field to overcome these problems and give to the Naval
Architect a guidance in the design of optimized hull
forms, provided that a reliable, sufficiently fast and
reasonably expensive approach is available.

SUBSCRIPTS Today computer power and progress in the field of

z: derivative wrt z direction
θ: derivative wrt θ direction
1. INTRODUCTION
Prediction of hull performances with the aid of
computational tools is particularly important for two
main reasons: 1) to avoid the uncertainties associated
with the use of empirical equations, which are only
applicable to similar hull shapes, and 2) to mitigate the
computational fluid dynamics (CFD) make the numerical
approach an interesting tool, which presents several
advantages over traditional techniques. We want here to
mention only two main applications in the marine field:
1) the simulation of the behaviour of propulsors (screw
propellers or water-jets) in different operating conditions
(efficiency, cavitation, bearing forces, induced
pressures); 2) the ship motion simulation in a given sea-
state (seakeeping calculations) or in manoeuvring
conditions.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

This paper presents a CFD approach for the calculation
of lift, drag, dynamic trim and sinkage of planing hulls
moving at steady speed through calm water, using a
commercial RANS CFD code.
In 2002 Thornhill and others [2] presented the results of
calculations performed using the FLUENT CFD code on
a planing hull, compared with the results obtained by the
Savitsky approach and by experiments using a relatively
small model. In the numerical calculation they didn’t
model turbulence and they used a relatively low
resolution grid (of the order of 150,000 elements),
calculating the frictional components separately. In
general they found an over prediction of the net pressure
on the hull surface, leading to higher lift values, with
respect to the experimental values. They indicated as
possible causes of that discrepancy an insufficient grid
resolution, the lack of turbulence modeling and the
treatment of spray.
The approach here presented is an attempt to overcome
these problems.
As for the Hull-A, several resistance tests were
performed on a model with full appendages, scale ratio
1:3.8, having an overall length of about 4 m; for the Hull-
B, the towing tests were performed on a naked model
with spray rails, in scale 1:6, leading to an overall length
of about 3 m. Both models were towed through the shaft
line thrust bearing position, with the same inclination of
the shaft. The same towing condition was simulated in
the numerical computation.
The next sections describe in further detail the global
calculation procedure.
2. CFD COMPUTATIONS
The Fluent code solves the complete set of Navier-Stokes
equations on structured/unstructured computational
domains, using a finite volume method.
An hybrid tetrahedral/hexahedral mesh has been set up to
calculate the flow field around the hull in a mixed air-
water environment.

Computations have been performed taking into account
both the water-air interface and the flow turbulence. The
air-water interface has been explicitly captured during
computations, using a Volume-of-Fluid algorithm [3].
This approach allows the real shape of the free surface to
be obtained as a result of the simulation. The VOF
technique is particularly advantageous for this kind of
problems, in terms of efficiency and robustness, over
techniques which explicitly track the surface during the
iterations.
The hybrid grid has been coupled with “matching
surface” techniques to join the different grid blocks. This
procedure allows smaller mesh sizes, faster modelling of
complex geometries and a faster dynamical adaption of
the mesh, with a computational time cost reduction [5].
In this paper, 6E05 computational cells have been used
for the numerical simulations.
Hexaedral
Element

Flow turbulence has been taken into account through a

Reynolds Averaged Navier Stokes (RANS) approach.
The Wilkox k-ω model has been utilized in the
computations [4].

All computations have been performed using the Fluent

6.1 CFD code [3]. Tetrahedral
Element
Results of the CFD simulations give the forces and
moments acting on the hull surface. A global iterative
procedure has been developed to utilize these results to
determine the position of dynamic equilibrium of the hull. Figure 1: Computational Grid : Hybrid Mesh

This iterative procedure is based on the possibility,

Matching Surface
offered by the Fluent code, to dynamically adapt the
mesh during the CFD algorithm iterations. Two degrees
of freedom have been taken into account: a vertical
movement, corresponding to the hull sinkage, and a
rotation around the axis normal to the symmetry plane,
and passing through the hull center of mass, that
determines the hull attitude. The results, in terms of
equilibrium position, forces and moments, are then
compared to the available experimental data obtained on
scaled models at the Brodarski Institute in Zagreb.

Figure 2: Computational Grid : Matching Surface

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

The cell spacing near wall surface has been chosen such
that the wall y+ value does not exceed 100 (figs. 1-.2).

Both hulls have been considered to be initially at rest in

calm water, and then to move with a relative velocity of
water
15 m/s for the Hull-A and 17.5 m/s for the Hull-B (fig.3). air

Interface Figure 6: Path Lines around the Hull-B

Outlflow
Inlet Air

Surface Hull V

Inlet Water

Symmetry

Figure 3: Computational Grid with Boundary Condition

The unsteady RANS and VOF equations have been

solved using the sequential algorithm available in Fluent, (Pa)
based on the SIMPLE method by Patankar [3]. A time Figure 7: Static Pressure on the Hull-A surface
step of 0.01 s has been chosen in the time implicit
algorithm, to avoid numerical instabilities associated
with the highly nonlinear equations. Each run of the code
has been performed on a four Pentium processors Linux
cluster and required about ten days to reach a steady state
solution.

The following figures show the computational results for

Hull-A and Hull-B in equilibrium position:
a) The free surface shape (figs. 4-5);
b) Path lines around surface hull (fig. 6)
a) Pressure contours on both hull surfaces (figs. 7-8)
4. FIGURES

(Pa)

Figure 8: Static Pressure on the Hull-B surface

3. DESCRIPTION OF THE PROCEDURE TO

DETERMINE THE HULL EQUILIBRIUM
POSITION

An iterative methodology has been developed to

Z (m)
determine the hull equilibrium position, corresponding to
Figure 4: Free Surface shape around the Hull-B constant values of the hull speed and weight. The first
step consists of solving the flowfield in an initial position
(obtained using the Savitsky method) of the hull, and
then to evaluate the corresponding forces and moments.
Then two other steps follow, where either the hull
sinkage or the hull trim angle are arbitrarily modified. In
these new positions, new CFD computations are
performed, to determine the corresponding forces and
moments acting on the hull surface.

Figure 5: Free Surface shape around the Hull-B

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

It must be noted that the new computations do not The results of this computational procedure, compared
require an externally generated grid. Indeed, a user with the experimental data are shown in the tables 1 and
defined function has been developed to change the hull 2. Figures 11 and 12 show, for the Hull-A, the free
position and orientation, which, in combination with the surface shape with the corresponding experimental image.
Fluent capability to dynamically adjust meshes when
boundaries are modified, allows the new mesh to be Model (L – W)/W (D – Dexp)/Dexp
regenerated inside Fluent. Hull-A +5% -3%
Hull-B -4% - 11 %
At this point, three sets of forces data are available, that
can be used to predict the equilibrium position, assuming Model (θ – θexp)/ θexp (ZG – ZGexp)/ ZGexp
a linear dependence of forces on position. Indeed, Hull-A +7% + 13 %
performing a Taylor series expansion of lift, drag and Hull-B - 18 % +3%
pitching moment around the (unknown) equilibrium Table 2: Relative Error between computational and
values, one may write: experimental results
 ∆L   ∆L  
∆L =L − Leq =   ⋅( ϑ −ϑeq ) +   ⋅(Z −Zeq ) 
 ∆ ϑ z  ∆Z ϑ  Savitsky Method
θ (i ) , Z ( i )
 ∆D   ∆D   (1)
∆D =D − Deq =   ⋅( ϑ −ϑeq ) +   ⋅(Z −Zeq )  CFD
 ∆ ϑ z  ∆Z ϑ 
L(i ) , M (i ) , D(i ) , L(zi ), M z( i ) , Dz( i ) , L(ϑi ), Mϑ( i ) ,Dϑ( i )
 ∆M   ∆M  
∆M = M − Meq =   ⋅( ϑ −ϑeq ) +   ⋅(Z −Zeq ) 
ϑ
 ∆ z  ∆Z ϑ 
Equilibrium
Equation
where the equilibrium values are related by the static
equilibrium conditions (fig. 9). θ (i +1) , Z (i +1) ,T ( i +1 )
eq
L
XTG
CFD
z
M
G
D
ϑ L(i +1) , M (i +1) , D (i +1)
v Teq ZTG
x O

W
Equilibrium?
Figure 9: Force Equilibrium
L eq = W – Teqsina = L NO Yes
End

Deq = Teqcosa = D (2) Figure 10: Flow diagram to predict the equilibrium
position.
Meq = Teqsina XTG – Teqcosa ZTG = M
Equations (1), together with (2), constitute a linear
algebraic system of three equations in the three
unknowns: z (sinkage), θ (trim) and Teq (thrust). The
derivatives appearing in the system (1) may be computed
by finite differences, using the results of the previous
CFD computations. A flow diagram describing the
procedure is presented in Fig. 10.

Model L (N) W (N) D (N) Dexp (N)

Hull-A 106 550 101 500 16 620 17 150

Z (m)
Hull-B 121 000 126 050 17 200 19 340 Figure 11: Free Surface shape around the Hull-A
(Comp. Results)
Model θ (deg) θexp (deg) ZG (m) ZGexp (m)
The results presented in Tables 1 and 2 show a fair
Hull-A 4.5 4.2 2.44 2.16 agreement between experimental and numerical data. In
particular, the computed drag values seem to
Hull-B 3.1 3.8 2.03 1.97
underestimate the experimental ones in both the
Table 1: Sinkage, Trim angle and Thrust values in computed cases. A lower computed drag is, in turn,
equilibrium Position. Comparison between experimental associated with the lower values of the computed sinkage
and computational results. z and the different values of the trim angle θ (table 1).

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

5. REFERENCES

[1] Hydrodynamic Design of Planning Hull, Daniel

Savitsky, Marine Thechnology, October 1964
[2] Planning Hull Performance Evaluation Using a
General Purpose CFD Code, Eric Thornhill et al.
Twenty-Fourth Symposium of Naval Hydrodynamics,
2002.
[3] FLUENT 6.1 User Guide, Fluent Inc., Lebanon, New
Hampshire, 2003.
[4] Turbulence Modeling for CFD, D. C. Wilcox, DCW
Industries Inc., La Canada, California, 1998.
[5] Advances in computational fluid dynamics
Technology, The Naval Architet, April 2003.

Figure 12: Free Surface shape around the Hull-A (Exp.
Results).
In the case of the Hull-A, the effect of the higher trim
angle (compared with the experimental value) on the
computed drag is partially compensated by the effect of
the lower sinkage, so that the overall effect is low
(differences on drag are about 3%).
On the contrary, the computations on the Hull-B show
that the effects of the trim angle and the boat sinkage act
in the same direction to reduce the drag with respect to
the experimental data.
6. AUTHORS’ BIOGRAPHIES
Michele Visone, Aerospace Engineer, CFD Responsible
at BLUE Group.
Mhoammed Eid, Aerospace Engineer, Research and
Development Responsible at BLUE Group.
Calogero Falletta, Naval Architect, partner SYDAC Srl
Ship-Yacht Designers & Consultants, Italy
Pier Luigi Ausonio, Naval Architect, partner SYDAC
Srl Ship-Yacht Designers & Consultants, Italy

Further analyses are foreseen to try to explain the
observed discrepancies. From an experimental point of
view, further model and/or full scale tests with pressure
distribution measurements should be advisable to obtain
a more detailed comparison with the computed values.
Diego Paterna, PhD, Research Assistant, DISIS,
University of Naples, Italy
Raffaele Savino, Associated Professor of Aerodynamic,
DISIS, Univ. of Naples, Italy.

From a numerical point of view, a sensitivity study of the

computed solution to the mesh size and quality, and to
the turbulence model should be performed.

4. CONCLUSIONS

CFD computations have been performed to determine the

forces and moments acting on the surface of a motor
yacht, using a numerical procedure that combines the
RANS and VOF approaches. Furthermore, a global
iteration procedure has been developed to dynamically
determine the hull sinkage and attitude for a given speed
condition. The numerical results have been compared
with the available experimental data, and a fair
agreement has been found. However, it is necessary to
find a suitable trade off between mesh resolution and the
surface capturing methodology to reduce the
computational time and costs to acceptable levels.

Although further investigation seems necessary from

both the numerical and experimental point of, the
combined RANS-VOF CFD numerical simulation
appears to be very promising in the planing hull field.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

SECOND-ORDER WAVE FORCES AND FREE-SURFACE ELEVATION AROUND A

MOORED SHIP IN STEEP UNI-DIRECTIONAL AND SPREAD WAVES
J Zang, K Wang, R Eatock Taylor and P H Taylor, University of Oxford, UK

SUMMARY

Based on a quadratic boundary element method, our in-house computer program DIFFRACT has been developed to
allow the calculation of non-linear wave forces and wave run-up around an arbitrary fixed or moored ship. Regular
waves and uni-directional and directionally spread waves can be treated in a computationally efficient manner. A wide
range of benchmarking tests in unidirectional and directional waves have been performed to validate the scheme. The
non-linear interaction of steep waves with an FPSO is discussed first in this paper. The significance of wave directional
spreading on mean-drift forces is then addressed and discussed.

1. INTRODUCTION for including the effect of wave directionality on the non-

The use of ship-shaped floating oil production systems is
becoming commonplace around the world, even in areas
with severe wave environments. Recent research
undertaken in Oxford has demonstrated the importance
of non-linear effects for the surface run-up around the
bow of a floating ship in unidirectional waves [11, 12].
Fixed and floating offshore production units are very
much prone to the effects of the weather, in particular as
a result of their requirement to remain in the same
position. For ships it is often assumed that there is an
equal probability of encountering waves from any
direction and thus long-term calculations are made on the
assumption of a uniform distribution of directions.
FPSOs have a ship-like form with one axis of symmetry
and with the longitudinal dimension much larger than the
transverse one, which make them particularly sensitive to
the direction of the loading. Furthermore, as FPSOs are
very sensitive to combined loading from more than one
direction, wave directionality is much more important
than other types of platform. However, there are
difficulties using with standard tools for computing the
effects of wave directionality in the prediction of second
order mooring forces. The conventional method is very
demanding on computer and human resources – in
particular adequate resolution of wave spreading is hard.
Therefore, advanced methods for the design and planning
of operation of these vessels require more sophisticated
models of the directionality of the environment [5] as
well as models of hydrodynamic loads and structural
response that are able to cope with the additional
directional information.
Using a quadratic boundary element method, our in-
house computer program DIFFRACT has been developed
to allow the calculation of non-linear wave forces and
wave run-up around an arbitrary fixed or moored ship.
Regular waves and uni-directional and directionally
spread waves can be treated in a computationally
efficient manner. It is able to deal with uni-directional
and directional bi-chromatic input wave systems,
calculating second-order wave diffraction under regular
waves and focused wave groups. Recently, it has been
extended to tackle spread seas, using a novel approach
linear hydrodynamics of an FPSO.
A wide range of benchmarking tests in unidirectional and
directional waves has been performed to validate the
numerical scheme. Non-linear wave loads and free
surface elevations on a bottom mounted circular cylinder
have been checked against published results for bi-
directional and bi-chromatic regular waves and focussed
wave groups.
The non-linear interaction of steep waves with a simple
representation of an FPSO is discussed first in this paper.
For head-on uni-directional waves, our numerical results
show excellent agreement with experiments conducted at
Imperial College. The non-linear sum and difference
frequency free surface components at the bow of the ship
are very significant, with implications for both wave
impact and green water on the structure.
The results for the same body exposed to spread seas are
then presented and discussed. Numerical solutions of the
problem are obtained for various types of directional
spreading. The results are compared with the
corresponding results for a unidirectional wave.
Finally some results based on a design similar to the
Schiehallion FPSO are discussed, illustrating the effects
of wave directionality on low frequency drift forces.
2. WAVE DIFFRACTION THEORY AND
NUMERICAL APPROACH
An arbitrary, three-dimensional structure is considered to
be fixed or freely floating on water of depth h in
unidirectional waves and spread seas. We shall make the
following assumptions: the fluid is incompressible and
inviscid, and the flow is irrotational. Under these conditions,
the equation of continuity shows that velocity potential Φ
satisfies the Laplace’s equation.
A justification of the second order wave diffraction
theory and the methodology implemented in our
numerical analysis for unidirectional wave groups, were
given in [11]. Here, emphasis is placed on the special
numerical treatments for spread seas.

© 2005: Royal Institution of Naval Architects


By applying Stokes’ perturbation method, the wave
velocity potential in directionally spread seas to second
order can be expressed by
N N
Φ( x, y , z , t ) = ∑ φi(1) ( x, y , z )e −iωi t + ∑∑ [φij+ ( x, y , z )e
i =1
−iωij−t
+ φ ( x , y , z )e
−
ij ]+K
Here ωi is the ith wave frequency, and N is the total
number of frequency components. The velocity potential
φ is a spatial variable: its first order and second order
terms can be expressed as
M
φi(1) ( x, y, z ) = ∑ φim(1) ( x, y, z )ei ( k ( x cos(θ
i m ) + y sin(θ m )))
m=1
M M
φij± ( x, y , z ) = ∑∑ φijmn
± i ( k i ( x cos(θ m ) + y sin(θ m )) ± k j ( x cos(θ n ) + y sin(θ n )))
( x , y , z )e
m =1 n =1
where ki is wave number, θm is wave direction, and M is
the total number of components of wave direction. The
sum and difference frequency components are designated
by + and – respectively.
A linear representation of a directionally spread
NewWave group can be given by
N M
η (1) ( x, y, z ) = ∑ an ∑ bnm ( k n ( x cos(θ m ) + y sin(θ m )) − ω n t )
n =1 m =1
where
M
∑ bnm = 1.
m =1
an is the wave amplitude of the nth wave in the wave
group, and bnm is the coefficient for the mth wave
direction with the nth wave frequency.
For an FPSO, with complex body geometry, numerical
simulation is required to perform the non-linear
diffraction analysis. The computation for conventional
second-order diffraction in directional spread seas needs
quadratic transfer function (QTF) matrices for a large
number of bi-directional and bi-chromatic waves. The
normal procedure for obtaining QTFs in unidirectional
waves requires the integration of free surface integrals,
involving first order scattered wave results at pairs of
frequencies [11,12]. The most obvious way of
performing the analysis in a spread sea would be to do
these integrals for each pair of frequencies at each pair of
directions. Thus if there are N frequencies and M
directions in the discretisation of a directional spectrum,
N2× M2 pairs of second order diffraction calculations
would be required, which leads potentially to an
extremely large computation.
However, examination of the structure of the second
order wave diffraction theory and its implementation in
Marine CFD 2005: Southampton, UK
numerical analysis using a boundary element method
shows that a major reduction of computational demand
may be obtained if we perform the analysis based on
non-planar waves. The idea is to group all the waves
N from different directions but with same frequency
−iωij+ t
i =1 j =1
together as a single incoming wave, then input this non-
planar wave as a component for each pair of diffraction
calculations in bi-chromatic waves. Using this new
approach, one only needs to evaluate N2 pairs of second
order solutions, rather than N2× M2. This will reduce the
calculation for spread waves to around the same amount
of the computational effort as for unidirectional waves.
3. RESULTS AND DISCUSSIONS
In order to validate the numerical model, the first
calculation presented in this paper is to apply the
numerical algorithm to a simple representation of an
FPSO in a head-on unidirectional wave group, to
compare with experiments performed at Imperial
College. Inclined unidirectional wave groups incident on
the same body are analyzed next, to highlight the effect
of wave direction on the wave hydrodynamics.
Following this, results for spread seas are discussed and
compared with unidirectional results, to provide first-
hand information on nonlinear wave interaction with a
body in a spread sea. Finally, a moored vessel similar to
the Schiehallion FPSO is considered and mean-drift
forces are discussed.
3.1 COMPARISON OF NUMERICAL RESULTS
WITH EXPERIMENTS FOR HEAD-ON UNI-
DIRECTIONAL WAVES
For comparison with experiments at Imperial College, a
simple representation of an FPSO in a head-on uni-
directional wave group is considered in our numerical
modelling. A focused wave group is generated in the
wave channel at Imperial College, and the wave group is
investigated in the two cases of crest focussing and
trough focussing at the bow of the FPSO model, which is
headed directly into the wave direction. Only free surface
elevations are measured in the experiments, and the
model is rigidly mounted at the centre of the tank. There
are 16 gauges aligned along the centre of the wave
channel, which has a total width of 2.8m.
Wave condition
Wave periods in top-hat spectrum wave group
T=0.8s ~ 1.2s
Water depth d=1200mm
Wave crest elevation at focus A=62mm
FPSO model structure layout
Length of FPSO model L=962mm
Width of FPSO model w=325mm
Draught of FPSO b=125mm
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

Two-plane symmetry was used in the numerical amplitude of the incoming wave is increased, this
computation for calculation efficiency. The final mesh enhanced value will be increased rapidly due to the
used for the numerical analysis is shown in Figure 1. squared relationship with the input amplitude. What is
clear from the work so far is the complex local
0
−0.05
−0.1
interaction between the incoming and diffracted waves
2 and the resulting greatly increased water surface
1.8 elevations local to the body.
1.6

1.4
elevation (m)
1.2
0.2
1 1st order incident
0.8 1st order diffracted
0.6
2
0.15 full 1st+2nd order diffracted
1.5
0.4
1
0.2
0.5
0 0.1
0

Figure 1: Numerical mesh 0.05

Expt free surface time history at bow (without ship)

crest focussed (Expt.) 0

0.1 trough focussed (Expt.)

0.05
-0.05
0

−0.05 -0.1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
−6 −4 −2 0 2 4 6 time t (s)
time t (s)

Expt free surface time history at bow (without ship)

0.1
0.05
0
−0.05
−6 −4 −2
time t (s)
Figure 2: Measured free surface time histories (without
ship) and linearised incoming wave
The incoming waves used for the numerical simulation
was derived from the experiments. The bottom picture in
Figure 2 shows the derived linearised incoming wave
time series, which is half the difference of the two
measured traces (crest focused and trough focused free
surface time histories).
The comparison of numerical results with the
experiments at Imperial is very good, as shown in detail
elsewhere [11]. Both free surface elevations for sum and
difference frequency agree very well, as do the total non-
linear free surfaces.
linear input (Expt.)
0 2
Figure 3: Incoming wave, first order diffracted wave and
complete first plus second order diffracted wave time
history around bow
3.2 INCLINED UNI-DIRECTIONAL WAVES
CASE
4 6
By applying the same wave group to a few non-head on
wave directions, we can see the effect of the wave
heading on the non-linear free surface elevations and
run-up on the ship in Figure 4. The maximal and minimal
total non-linear free surface to second order at four
different wave headings (β=0o, 15o, 30o, 45o) are
presented for waves focusing at x=-0.4m. The maximal
value in the graph is obtained by taking the maximum of
the non-linear free surface elevations reached during the
diffraction process for each location within a range of
time from –4s to 4s for wave focusing at x=-0.4m.
Similarly, the minimal values are given in the same way
for each location along the center line outside the body,
and along the waterline. The upper and lower figures
correspond to the lee side and the weather side
respectively. Note the stretched vertical scale.

Figure 3 gives three different free surface time histories The bow is the crucial location for both maximal and
predicted from the numerical analysis. The trace in the minimal values for the head-on case. But for other wave
figure appearing as the smallest wave is the linear headings, the maximal and minimal wave run-up on the
incoming wave used for numerical modelling; the middle body doesn’t always occur at the bow. Particularly for a
wave is the first order diffracted free surface; and the wave heading of 45o, the maximum of the non-linear free
largest wave is the complete first plus second order surface takes place at a mid part of the ship, while the
diffracted case. Both sum and difference frequency minimal value happens between the mid-section of the
components of the second order diffraction effects are ship and the stern.
significant. First order diffraction increases the incident
crest elevation by 45%, and the second order effects
increase the crest elevation by a further 30%. If the

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Maximal and minimal non-linear free surface spatial profile for lee side QTFs are progressively reduced below the uni-
0.2
directional values at high frequency.
β=0
β=15
0.15 β=30
β=45 Surge QTF β 0=0o
0.1 0.4
uni-directional
o
0.35 Spread sea σ =15 focusing at -0.4
elevation (m)

0.05 θ o
Spread sea σ =30 focusing at -0.4
θ o
0.3 Spread sea σ =15 focusing at 0.0
θ o
0 Spread sea σ =30 focusing at 0.0
θ
0.25
-0.05
0.2

-0.1
0.15

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0.1

x (m)
Maximal and minimal non-linear free surface spatial profile for weather side 0.05
0.2
β=0 0
β=15 4 5 6 7 8 9 10 11 12 13
0.15 β=30 ω
β=45
0.1
Figure 5: Surge QTFs for various values of Gaussian
spreading function focused at different locations in head-
on case
elevation (m)

0.05

0 Figure 5 also shows the effect of focusing position on the

-0.05
-0.1
-2 -1.5 -1 -0.5
x (m)
Figure 4: Maximal and minimal values of non-linear free
surface for various wave headings.
3.3 SPREAD SEA CASE
The non-linear interaction of spread seas with the
simplified FPSO model is considered in this section. A
Gaussian directional spreading function
1 2σ θ 2
D(θ ) =
2π σ θ
with various values for σθ, is used to model waves of
various angular spreading. For uni-directional waves
(σθ=0), our numerical results show excellent agreement
with experiments conducted at Imperial College [11, 12].
The QTFs for mean drift forces (i.e. for ωij− = 0 ) with
directional spreading σθ=15o and 30o in the main wave
direction βo=0.0o (head-on), 15o, 30o, and 45o are
obtained and compared with the results from uni-
directional waves. The forces are normalised by
πρgbA1A2, where A is the wave amplitude and b is the
draft. The influence of wave directionality on the mean
drift surge force in the head-on case is shown in Figure 5.
For this case the uni-directional forces are largest at all
frequencies. As the spreading angle is increased the surge
mean surge force. As discussed in [10], the local
behaviour of a spread wave close to the focus point leads
to the idea of a ‘focus spot’ over which the spread wave
looks locally uni-directional. The relative size of this
‘focus spot’ as compared to the size of the body plays an
0 0.5 1 1.5 2 important role in the diffraction of spread waves. As the
FPSO used in the calculation is relatively large compared
with the wavelength for the entire range of the wave
frequencies considered, a shift in position of the focus
point also affects the wave force, but a larger degree of
wave spreading always leads to a reduction of the surge
force.
Some of the results for non-head on cases are presented
in Figure 6. While it is found that uni-directional waves
still produce the maximum mean drift surge forces, it is
(θ −θ ) 2 interesting to see that the mean drift sway QTFs obtained
−
from spread waves are bigger than those from uni-
e
directional waves for some frequencies. It is evident that,
at certain wave frequencies in certain wave directions,
spread waves may lead to a more severe design case for
wave loading.
The above results have shown the significance of wave
directional spreading on mean-drift forces. They also
demonstrate that the influence of wave directionality on
the FPSO depends on a number of factors, such as
spreading function, main incoming wave direction, wave
frequency, focusing position, and relative ratio of body
size and spread wave characteristics etc.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

o -3
Sway QTF β 0=15 x 10 Sway QTF
0.2 1.8

uni-directional o
o 1.6 β=15
Spread sea σ =15 o
0.18 θ o β=30
Spread sea σ =30
θ β=45
o
1.4
0.16
1.2

0.14 1

0.8
0.12

0.6
0.1
0.4

0.08 0.2

0.06
4 5 6 7 8 9
ω ω
Figure 6: QTFs of mean sway drift forces for various
bandwidths of Gaussian spreading function in main wave
direction βo=15.0o
3.4 MOORED VESSEL ANALYSIS
A vessel similar to the Schiehallion FPSO was chosen to
investigate the non-linear wave hydrodynamics in the
REBASDO project. In order to provide data to partners
for their mooring analysis, QTFs are obtained for slow-
drift forces. The length of the ship is 238m, the width is
45m, and the draft is 11.5m. The one-plane symmetric
numerical mesh created for the diffraction analysis is
shown in Figure 7. Examples of surge and sway mean-
drift forces are presented in Figure 8 and Figure 9.
0
10 11 12 13 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Figure 9: Sway QTFs for various wave headings ( ω− = 0 )
The surge QTFs seem not to vary much between the
wave headings for most frequencies. But sway QTFs
vary considerably when the wave directions are changed.
The sway QTF values increase dramatically with the
increase of the angle of wave heading away from the
head-on in the case of short waves.
4. CONCLUSIONS
The second order wave interaction of uni-directional
focused wave groups and spreading seas with FPSOs is
presented in this paper. The non-linear wave scattering
problem is solved by employing a quadratic boundary
element method. The computer program DIFFRACT
developed at the University of Oxford has been
successfully extended to deal with uni-directional and
directionally spread waves, calculating nonlinear wave
hydrodynamic loads on the ship and free surface
elevations on and around the ship.

Our numerical results for the non-linear free surface have

shown that both sum and difference frequency second-
order free surface components at the bow of the ship are
significant, and cannot be neglected if one requires
accurate prediction of the wave-structure interaction;
Figure 7: Body surface mesh for FPSO otherwise a major underestimation of the wave impact on
-4
the structure could occur. As the wave amplitude is
x 10 Surge QTF
6 increased, this second order enhancement will increase
β=0
o rapidly as the square of wave amplitude. It is interesting
o
5
β=15
o
to note that the crucial location for both maximal and
β=30
β=45
o minimal values for the non-head on case is not at bow.
4
For a wave heading at 45o, the maximum of the non-
linear free surface takes place near the mid-section of the
ship, while the minimal value occurs between the mid
3
part of the ship and the stern.
2
In the case of the spread seas, the results demonstrate that
uni-directional waves do not always lead to the
1
maximum wave forces on a ship-shaped FPSO, which
contrasts to the effect of spread seas on a circular
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
cylinder [10]. The orientation of the main wave heading
ω can have a substantial influence on the spreading effects
Figure 8: Surge QTFs for various wave headings ( ω− = 0 ) on the body. We may underestimate the wave loading if
wave directionality is not considered.

© 2005: Royal Institution of Naval Architects


5. ACKNOWLEDGEMENTS
This study is part of the EU project Reliability Based
Structural Design of FPSO Systems (REBASDO)
supported by EU under Framework 5. The collaborative
partners include Shell, Instituto Superior Tecnico, DHI,
DNV, Imperial College, Estaleiros Navais da LISNAVE
and Noble Denton. The authors would like to thank the
EU for the financial support and the useful discussions
with the partners during the review meetings. We would
also like to thank the Oxford Super Computing Centre,
University of Oxford for providing continuous
computing support for the calculation, and Professor
Chris Swan at Imperial College for facilitating the
experimental comparisons.
6. REFERENCES
1. Bateman, W.J.D., Swan, C. & Taylor, P.H. ‘On the
efficient numerical simulation of directionally spread
surface water waves’, Journal of Computational Physics,
174, 277-305, 2001.
2. Bowers, E.C. ‘Long period oscillations of moored
ships subject to short wave seas’, Trans. R. Inst. Naval
Archit. 118, 181-191, 1975.
3. Chamberlain, P.G. & Eatock Taylor, R. ‘High
frequency TLP responses’, Managed programme on
behaviour of fixed and compliant offshore structures.
(final report on project FLU80), 1993.
4. Chau, F.P ‘The second order velocity potential for
diffraction of waves by fixed offshore structures’, Ph.D.
dissertation, University College London, U.K, 1989.
5. Jonathan, P. & Taylor, P.H. ‘On irregular, non-linear
waves in a spread sea’, Journal of Offshore Mechanics
and Arctic Engineering, 119, 37-41, 1997.
6. Kim, M.H. & Yue, D.K.P. ‘The complete second-order
diffraction solution for an axisymmetric body. Part 2:
Bichromatic incident waves and body motions’, Journal
of Fluid Mechanics, 211, 557-593, 1990.
7. Swan, C., Taylor, P.H. and Van Langen, H.
‘Observations of wave-structure interaction for a multi-
legged concrete platform’, Applied Ocean Research, 19,
309-327, 1997.
8. Taylor P.H. and Vijfvinkel E.M. ‘Focused Wave
Groups on Deep and Shallow Water’, Ocean Wave
Kinematics, Dynamics and Loads on Structures, ASCE
Proc. of the 1998 International OTRC Symposium,
Houston, 420-427, 1998.
9. Tromans, P.S., Anaturk, A. & Hagemeijer, P. ‘A new
model for the kinematics of large ocean waves -
application as a design wave’, Proc. 1st Int. Offshore and
Marine CFD 2005: Southampton, UK
Polar Engineering Conference (ISOPE), Edinburgh, 3,
64-71, 1991.
10. Buldakov, E.V., Eatock Taylor, R. Taylor, P.H.
‘Diffraction of a directionally spread wave group by a
cylinder’, Applied Ocean Research, 25, 301-320, 2003.
11. Zang, J., Taylor, P.H. & Eatock Taylor, R.
‘Speculation on the adequacy of second order diffraction
theory’, 19 IWWWFB, Cortona, Italy, 2004.
12. Zang, J., Gibson, R., Taylor, P.H. & Eatock Taylor,
R., Swan, C. ‘Non-linear wave diffraction around a
moored ship’, OMAE, Vancouver, Canada, 2004.
7. AUTHORS’ BIOGRAPHIES
Jun Zang currently holds the position of Departmental
Lecturer at the Department of Engineering Science,
University of Oxford, where she is responsible for two
research projects as principal or co-investigator. Her
doctorate was obtained at the State key Laboratory for
Coastal and Offshore Engineering at Dalian University of
Technology, and she has several years of industry
experience on design of offshore platforms and onshore
structures.
Ke Wang was a Research Assistant in the Department of
Engineering Science, University of Oxford during part of
the REBASDO Project. He was responsible for the
calculation of mean drift forces for the Schiehallion ship-
type model.
Rodney Eatock Taylor is Professor of Mechanical
Engineering at the Department of Engineering Science,
University of Oxford. He has coordinated several
national research programmes in the area of marine
hydrodynamics. He is a Fellow of the Royal Institution
of Naval Architects, and currently a Vice-President of the
Royal Academy of Engineering.
Paul Taylor holds the position of University Lecturer at
the Department of Engineering Science, University of
Oxford. Prior to that he spent 20 years in the mechanical
and offshore engineering industry. He currently chairs
the SUT Group on Environmental Forces (SUTGEF).
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

NUMERICAL SIMULATION OF YAW EFFECT

Q Gao, V Shigunov and D Vassalos, University of Strathclyde, UK

SUMMARY

Numerical simulation of effect of yaw angle and ship attitude on ship hydrodynamics was carried out by a RANS
approach. Two-equation turbulence model of renormalization group (RNG Κ-ε) was used together with non-equilibrium
wall function. Volume-of-Fluid (VOF) and geometric reconstruction methods were applied to locate free surface. The
governing equations were discretized by finite volume approach with collocate grid layout. Second order upwinding
difference method was used for the discretization of convection flux. SIMPLE algorithm combined with Rhie and Chow
interpolation was used to deal with the coupling of velocity with pressure. The linear algebraic equation system was
solved by a Gauss-Seidel method with multigrid acceleration. Series 60 hull form was chosen as a validation case. The
computed free surface wave pattern, side force, moment coefficient, velocity field were compared with experimental
data. The general agreement is satisfactory.

1. INTRODUCTION
The traditional method currently used by towing tanks
for ship manoeuvrability predictions is based on
calculation of hydrodynamic derivatives of forces and
moments with respect to the individual degrees of
freedom. These derivatives are then used in
manoeuvrability simulation programs to evaluate ship
performance in real manoeuvres for design purpose.
The hydrodynamic derivatives are calculated from
experiments or using potential flow simulation methods.
Potential methods can only be used in manoeuvrability
calculations with additional care because of strong
viscous effects. Experimentally based predictions are
expensive and can be applied only to the type of hulls
they were obtained for.
Methods based on RANS solver have been becoming
progressively more popular. In this paper, the application
of RANS method to predict force and flow field around
Series 60 ship model under steady oblique motion was
reported.
The flow around ship hull under steady yaw motion is
characterised by complex flow separation, vortex
shedding, and strong free surface effect. There were few
successful studies in this area.
Cura Hochbaum [2] presented a calculation of the vessel
in steady oblique motion on 22nd ONR symposium.
Three dimensional separation and vortex shedding were
captured well and force results were reasonably
consistent compared with experiment data. However,
free surface and ship attitude effects were not included in
the calculation.
Alessandrini and Delhommeau [3] presented a paper on
22nd ONR Symposium on viscous free surface flow past
Series 60 in steady sway motion. Wave pattern
particularly for the bow wave shows a close agreement
between calculation and measurement. Lateral force was
predicted with good accuracy. However, ship attitude
effects were excluded in the calculation.
A question arises in the study of yaw effect on ship
hydrodynamics: how large is the effect of ship attitude
change and free surface?
In this paper, the effects of yaw angle and attitude change
on turbulent flow and ship hydrodynamics were studied
by RANS solver FLUENT. The numerical results were
given and compared with experimental measurement
available from IIHR [4].
2. MATHEMATICAL FORMULATION

Longo and Stern [4] carried out towing tank experiments The Reynolds averaged Navier-Stokes equations with Κ-
of yaw effect on ship flow for a 3.048m Series 60 ship ε turbulence model for closure were solved. The
model. Forces, wave pattern and model attitude were governing equations can be written as follows.
measured in detail, which were used to explain
complicated physics and provide valuable data for Continuity equation:
computational fluid dynamics validation. r
∇ ⋅ V = 0.
Patel, Ju and Lew [1] firstly studied asymmetric effect on
flow physics by finite-analytic difference scheme. They Momentum equations:
∂ r rr r
validated results on HSVA tanker and SR107 bulk
( ρV ) + ∇ ⋅ ( ρVV ) = ρg − ∇P
carrier. The agreement was encouraging, although there ∂t
were two main limitations on their numerical method.
One is a double model assumption of the free surface + ∇ ⋅τ + S
effect and the other is that only stern part of ship model
was included in the simulation.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Turbulence model: 4. NUMERICAL RESULTS AND

∂ r DISCUSSION
( ρK ) + ∇ ⋅ ( ρVK ) = P − ρε
∂t
µ 4.1 MESH
+ ∇ ⋅ (( µ + T )∇K )
σK Three-dimensional computational grids were generated.
∂ r ε ε 2 The total cell numbers are about 400000, which were
( ρε ) + ∇ ⋅ ( ρVε ) = C1 P − C 2 ρ based on the uncertainty study of resistance calculations.
∂t K K The sketch of computational domain and boundary
µT
+ ∇ ⋅ (( µ + )∇ε ) meshes is shown in figure 1.
σε

VOF equation:
∂ r
(rw ) + ∇ ⋅ (rwV ) = 0.
∂t
0

Z
r
where V is velocity vector, g gravity vector, P -2
2

r r
τ = µ [(∇V + ∇V T )] stress tensor,
0

Y
-2
pressure, 0
2
4 -2
6

ρ = rw ρ w + ra ρ a mixture density, µ = rw µ w + ra µ a
X 8
10

Figure 1: Boundary meshes

mixture viscosity, ρw and ρa water and air densities,
µw and µa water and air dynamic viscosities, S 4.2 ATTITUDE EFFECT
external body force, K turbulent energy and ε turbulent
dissipation rate. Since the model tests for force measurement were carried
out in free condition (with sinkage, trim and heel) and all
Boundary conditions were prescribed past numerical calculation were made in restrained
condition (without attitude change) or with double
The inlet is located at half a ship length ahead of bow model, it is needed to quantify the influence of free
where two velocity components were imposed. surface and attitude change due to sway motion on ship
manoeuvring performance. Therefore, three separate
Velocity components and free surface elevation were calculations were carried out for Froude number 0.316
given on side boundary, which is located at one ship and yaw angle 10°, named as case 1 and 2 for with and
length away. without sinkage, trim and heel, case 3 for double model.

Downstream outlet is at two ship lengths behind stern, Ship wave

where hydrostatic pressure is prescribed. The wave measurement was carried out in restrained
condition without sinkage, trim and heel. The
Non-equilibrium wall function was used on ship hull comparison between measurement and calculation (case
boundary to save computational effort. 2) was shown in figure 2-4.

3. TEST CASE From figure 2, it can be seen that the general agreement
between computed and measured wave profile is good.
Bow wave at port is larger than that at starboard.
Series60 hull form was chosen for validation purpose.
However, wave height is very close near stern on both
Numerical calculation includes study of free surface and
sides. The magnitude of bow wave height is about 0.035,
attitude change effect as well as yaw angle effect on ship
which is larger than that in straight-forward motion
hydrodynamic performance. In all test cases, Froude
(0.015). Therefore, resistance coefficient in oblique
number was set to 0.316. The computational cases are
motion tends to be larger than that in straight-forward
listed in table 1.
condition. The first and second transverse wavelength are
Froude Yaw Sinkage Trim Heel
different each other and with Kelvin theory, which is
number angle
related to the nonlinear interactions between bow, front
0.316 10 Yes Yes Yes shoulder, back shoulder and stern wave. The difference
0.316 10 No No No of wave height at bow and stern may attribute to bow and
0.316 10 Double model stern shape, viscous effect as well.
0.316 7.5 Yes Yes Yes
0.316 5 Yes Yes Yes
0.316 2.5 Yes Yes Yes
Table 1 The list of test cases

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

0.04
Generally, the calculation results in case 2 are consistent
0.03 with measurement. However, the computational results
Measured, port have some numerical diffusion, which result in thicker
Measured, star
0.02 Computed, port
Computed, star
boundary layer and weaker vortex strength. The
numerical results could be expected to improve by
Zeta/L

0.01 increasing mesh resolution.

1.0
0 0

-0.01

-0.01 -0.02 -0.0

U
1.00
-0.03 0.96 -0.0
0.91
0.87 -0.0
-0.02 -0.04

z
0 0.25 0.5 0.75 1 0.83
0.79
-0.0

z
X/L -0.05 0.74
0.70
Figure 2: Computed and measured wave profile -0.06 -0.0

-0.07 -0.0
rame 003  01 Feb 2005  title

The measured and computed wave patterns were given in -0.08

-0.05 0 0.05
-0.0

y
figure 3-4. It can be seen that general wave structure is -0.0

well predicted by numerical approach. However, the

numerical calculation appears to be over diffusive far Figure 5: Velocity field at x/L=0.1
from ship hull due to the mesh resolution. (Measurement [4])
1

Level z
19 0.0265
17
15
0.0224
0.0183
1.0
13 0.0143 0
11 0.0102
0.5
9 0.0062
-0.01
7 0.0021 U
1.00
Y/L

5 -0.0020
3 -0.0060 0.96 -0.02 U
1 -0.0101 0.91 1.00
0.87 -0.03 0.96
0.83 0.91
0.79 0.87
-0.04
z

0 0.74 0.83
0.70 0.79
-0.05 0.74
0.70
-0.06

-0.07
0 0.5 1 1.5
X/L -0.08
-0.05 0 0.05
y
Figure 3: Wave pattern
(Measurement [4]) Figure 6: Velocity field at x/L=0.1
1
(Calculation case 1)
Level z
19 0.0265
17
15
0.0224
0.0183
1.0
13 0.0143 0
11 0.0102
0.5
9 0.0062 -0.01
7 0.0021
Y/L

5 -0.0020
-0.02 U
3 -0.0060
1 -0.0101 1.00
-0.03 0.96
0.91
0.87
-0.04
z

0 0.83
0.79
-0.05 0.74
0.70
-0.06

-0.07
0 0.5 1 1.5
X/L -0.08
-0.05 0 0.05
y
Figure 4: Wave pattern
(Calculation case 2) Figure 7: Velocity field at x/L=0.1
(Calculation case 2)
Velocity field
The three dimensional velocity fields were compared in 0 1.0
figure 4-9 at two cross planes. One is at x/L=0.1 and the -0.02 U -0
other at 0.9. 0.99
0.92
-0.04 0.85 -0
z
z

0.78
0.72
The results from calculations case 1 and 2 show that -0.06 0.65
0.58
-0
0.51
there is a significant attitude effect on both boundary -0.08
Frame 003  01 Feb 2005  title
-0

layer thickness and cross-stream distribution. Compared -0.05 0 0.05

y
with heel, sinkage and trim has less effect on cross
Figure 8: Velocity field at x/L=0.9
velocity distribution. Heel angle is large to change the
(Measurement [4])
vortex shedding and pressure distribution on the hull
surface.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

0 1.0 4.3 YAW EFFECT

-0.02 U
0.99
0.92 Based on above analysis, a series of numerical
-0.04 0.85
calculation were carried out by using measured sinkage,
z

0.78
0.72
-0.06 0.65
0.58 trim and heel. The comparison of numerical and
0.51
-0.08 experimental results was given in figure 11-13.
-0.05 0 0.05
y
It could be seen that the agreement between calculation
Figure 9: Velocity field at x/L=0.9 and measurement is generally acceptable. However,
(Calculation case 1) computational hydrodynamic forces and moments are
0 1.0 slightly underestimated. This is partly due to limitation of
-0.02 U
0.99
mesh resolution, and partly due to the Κ-ε turbulence
-0.04
0.92
0.85 model. The studies of K-ω and turbulence stress model
z

0.78
0.72
0.65
are undergoing.
-0.06
0.58
0.51
-0.08
8
-0.05 0 0.05
y

Figure 10: Velocity field at x/L=0.9

7
(Calculation case 2)

Ct
Exp.
Forces and moments 6 Com.
The definition of force coefficient is given below:

Resistance coefficient: 5
Yaw angle
Ct=Fx/(0.5ρU2S) 0 2 4 6 8 10 12

Side force coefficient:

Cs=Fy/(0.5ρU2S) Figure 11: Resistance Coefficient

Yaw moment coefficient:

Cm= Mz/(0.5ρU2Lpp3) 30
25
Side force (N)

The comparison of force and moments at Froude number

0.316 and yaw angle 10 degrees is given in table 2. 20
15 Exp.
Case Ct Cs Cm 10 Com.
Exp. [4] 7.12 26.78 -1.83
5
Case 3 3.85 18.3 -1.08
0 Yaw angle
Case 2 6.91 24.98 -1.45
Case 1 7.01 25.26 -1.81 0 2 4 6 8 10 12
Table 2: Forces and moments coefficient (*1000)

It can be seen that the forces and moments are closely Figure 12: Side force coefficient
related to the computational conditions.
0 Yaw angle
The computed results from case 1 (with attitude effect)
0 2 4 6 8 10 12
agree reasonably well with experimental data. However,
-0.5
the computational side force and moment from case 2
(without attitude effect) are much smaller than Exp.
Cm

experimental data. This means that the attitude change -1

Com.
has a significant influence on the overall ship
hydrodynamics and cannot be neglected in larger yaw -1.5
angle. The computational results from double model
calculation are far below experimental results. That -2
means that wave effect cannot be neglected as well.

Figure 13 Yaw moment coefficient

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

5. CONCLUSION

Based on comparison of computed and measured results,

the following conclusion could be drawn:

a. The general agreement between computational

results and measurement are satisfactory. However,
increasing mesh resolution and enhanced turbulence
models are recommended in future study to improve
numerical accuracy.

b. The effects of attitude change on hydrodynamics are

significant in large yaw angles and should be taken into
account in the numerical prediction of manoeuvring
motion.

c. Free surface effect cannot be neglected for high

Froude number as well.

6. REFERENCE

1. V. C. Patel, S. Ju and J. M. Lew (1990) Viscous

Flow past a Ship in a Cross Current, 18th ONR
Symposium on Naval Hydrodynamics, Ann Arbor

2. A. C. Hochbaum (1998) Computation of the

Turbulent Flow around a Ship Model in Steady Turn and
in Steady Oblique Motion, 22nd Symposium on Naval
Hydrodynamics, Washington D. C., USA

3. B. Alessandrini and G. Delhommeau (1998) Viscous

Free Surface Flow Past a Ship in Drift and in Rotating
Motion, 22nd Symposium on Naval Hydrodynamics,
Washington D. C., USA

4. J. Longo and F. Stern (1997), Yaw Effect on Model-

Scale Ship Flows, 21st Symposium on Naval
Hydrodynamics, Norway

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

COMPARISON BETWEEN RANSE CALCULATIONS AND PANEL METHOD RESULTS

FOR THE HYDRODYNAMIC ANALYSIS OF MARINE PROPELLERS
P Becchi and C Pittaluga, CETENA S.p.A. - Hydrodynamic Dept., Italy

SUMMARY

The study of the main propeller is a priority research theme for naval hydrodynamics owing to the complexity of the
related physical phenomena and its impact on the overall ship design. CETENA is involved in R&D activities on the
modelling of the propeller flow using both potential (panel methods) and viscous (RANSE solvers) flow methods.

The work presented in this paper regards a validation study carried out for a high skewed propeller geometry and
concerning a comparison between experimental data, RANSE and panel method results were carried out. The RANSE
calculation results were carried out within the European-founded project “LEADING EDGE” (Contract No: G3RD-CT-
2002-00818).

A detailed analysis was performed in order to define the better panel grid for the propeller, in order to choose the panel
grid that lead to the better performance result and the less time consuming. This sensitivity analysis was carried out for
the design operating condition. Then validation against experimental data was carried out in open water conditions, in
terms of thrust, torque and efficiency coefficients. At the end, for a single operating condition, RANSE calculations were
carried out, in order to compare the pressure distributions over the blade sections. This kind of comparison makes it
possible to relate the panel method pressure results to the viscous phenomena that is possible to study by RANSE
calculation, especially for tip vortex, trailing vortex and streamlines over the blade.

NOMENCLATURE 1. INTRODUCTION

D Propeller diameter The study of the main propeller is a priority research

R Propeller radius theme for naval hydrodynamics owing to the complexity
ζ Non-dimensional radial position of the related physical phenomena and its impact on the
p total pressure overall ship design. For this reason the propeller design
patm atmospheric pressure should be carried out taking into account the ship shape,
g gravity shaft line and brackets, rudder, engine design. The
h head optimal propeller’s operation condition depends on
ω rotational velocity significant aspect such as the contractual ship speed, the
VA Advance velocity fuel and bearing consumption, noise and vibration level,
VR Resultant velocity at a blade section etc. This chain of different related and important aspects
RN Reynolds number doesn’t allow to fix the numerical tool employed for the
i Panel counter in chordwise direction study of the naval propeller.
j Panel counter in spanwise direction
nPANC Number of panels, chordwise direction CETENA is involved in R&D activities on the modelling
nSTRIP Number strip, spanwise direction of the propeller flow using both potential (panel
S(i, j) Area of the panel (i, j) methods) and viscous (RANSE solvers) flow methods.
s(i, j) Arch length of the panel (i, j) from the The propeller panel method PROPACE [2] was
leading edge developed in the early 90s at CETENA’s and since then
C(j) Chord length of the panel strip (j) has been constantly validated and improved. This work is
P(ζ) Blade pitch at the radial position ζ own to the feedback from an extensive campaign of
J Advance coefficient theoretical / experimental correlation studies.
n Rotational speed (rps)
KT Thrust coefficient Nevertheless the advancements made by these research
KQ Torque coefficient the potential flow method cannot take into account local
η Propeller efficiency effect such as the tip and root vortex structures, the
ρ Water density propeller wake phenomena that highly influence an
CD Drag coefficient accurate prediction of the propeller efficiency. For this
CF Friction coefficient reason it is necessary to introduce the viscous effect
LE LEADING EDGE EU project combining the potential flow theory with viscous method
MKC Morino Kutta Condition [1] based on the boundary layer approximation and / or with
IKC Iterative Kutta Condition tools related with the solution of the Reynolds Navier-
Stokes equations. So, once the reliability of this RANS
technique has been tested against experimental data,
these kind of numerical results can be used to improve

© 2005: Royal Institution of Naval Architects


the precision of potential methods that still remain a quite
fast and accurate tool for the preliminary propeller design.
The work presented in this paper regards a validation
study developed for a highly-skewed propeller in open
water condition. A commercial RANSE solver (CFX5)
[3] was validated and the results of this computation has
been employed to improve CETENA’s panel code
(PROPACE), especially from the point of view of the
viscous correction formulation and pressure coefficient
distribution in steady condition. The assessment of the
reliability of CFD calculations as regards the
determination of the 3D wake around this propeller has
been carried out and funded by the EC project
LEADING EDGE, that concerns prediction of leading
edge and tip flow for the design of quiet and efficient
screw propellers.
The paper is organised as follows: the first section
describes the propeller geometry; the second section
details the computational method, both the potential and
the viscous approach, the following sections describe the
experimental data and lastly the verification and
validation procedure.
2. PROPELLER GEOMETRY
The propeller geometry used in this work is a four bladed,
high skew, controllable pitch propeller from LEADING
EDGE EU project.
Main propeller geometry data in model scale
Propeller diameter: 0.233 m
Boss/diameter: 0.321
Number of blades: 4
Rotational Speed: 14 Hz
Maximum Skew: 27°
AE/AO: 0.729
The propeller was analysed at an off-design pitch.
Figure 1
Marine CFD 2005: Southampton, UK
3. CFD TOOLS AND COMPUTATION
SETTINGS
3.1 PANEL METHOD – PROPACE
The PROPACE code is a panel method mainly based on
the potential theory published by Hoshino [4, 5, 6]. The
propeller geometry is represented entirely, by planar and
non-planar panels with a constant distribution of source
and doublets. The pressure Kutta condition is imposed at
the blade trailing edge with an iterative procedure, while
the wake shape can be modelled considering two
different ways: constant or equal to the blade pitch
distribution.
Recently, a windows graphical user interface has been
developed in order to implement the pre-process phase of
a calculation. The PROPACE GUI makes it possible to
perform an automatic panel grid sensitivity analysis, to
handle the propeller geometry in a quick way, to provide
the 3D surface files used to define the CFD fluid domain,
etc.
Figure 2
3.1(a) Geometry of Panelization
In order to perform a PROPACE calculation, the
propeller geometry needs to be checked. In fact, a panel
method calculation requires some geometrical conditions
to be fulfilled. The blade surface must to be a close
surface: the edges have to overlap at the tip and at the
trailing edge. For this reason, the t/C value at the tip and
the blade thickness at the trailing edge have to be equal
to zero. Also the panel shape and aspect ratio at the tip
have to be checked, especially for high skewed propeller.
In this case, the pressure distribution over the blade in the
tip region can be affected by numerical errors. So, this
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

problem can be avoided setting a non-zero tip chord Based on these results, the panel grid configuration was
length, by an extrapolation procedure on the chord length set in order to provide the higher defined geometrical
distribution. discretization. The parameter set used for the calculation
is represented in the following table.
3.1(b) Panel Method Sensitivity Analysis
3.1(c) Viscous Correction Formulation
In order to define the better panel grid setting for a
PROPACE calculation, a sensitivity analysis was In order to evaluate the viscous effect on the propeller
performed. Because the panel grid setting has a particular performances, the PROPACE code uses a semi-empirical
influence on the calculation performing and results, it is formulation that makes it possible to estimate the friction
important to check the better panel configuration that drag contribution on thrust and torque. The total KT and
makes it possible to provide good results and reduced KQ results are defined in the following way:
computation time. For this reason, three different K T TOT = K T POT − K T FRIC
analyses were carried out: in the first one, the effect of K Q TOT = K Q POT + K Q FRIC
the number of points in chordwise and spanwise
direction was studied jointly to the effect of the panel where the frictional contributions are evaluated as:
distribution law in chordwise direction, to the wake TFRIC Q FRIC
K T FRIC = Z ⋅ K Q FRIC = Z ⋅
contraction model and the wake pitch distribution. In ρ⋅n ⋅D
2 4
ρ ⋅ n 2 ⋅ D5
spanwise direction, the panels were distributed with a co- being
 P(ζ )  
sinusoidal law. In the second phase, the wake contraction nSTRIP nPANC

TF = ∑ ∑ ⋅ ρ ⋅ VR ⋅ C D (i, j) ⋅ S(i, j) ⋅Sin  Arc tan 
1
 
2
model, and the panel distribution law in the chordwise
direction, were fixed, while the number of points in j=1 i =1 2   π ⋅ ζ ⋅ D 
chordwise and spanwise direction was changed from 30 nSTRIP nPANC

∑ ∑ 2 ⋅ρ⋅ V ⋅ C D (i, j) ⋅ S(i, j) ⋅

1
QF =
2
to 50. At the end, based on the previous results, a R
j=1 i =1
different panel configuration was tested, in order to
  P(ζ )  
  ⋅ (ζ ⋅ R )
check the reliability of a very highly defined grid.
Cos Arc tan  
  π ⋅ ζ ⋅ D 
As mentioned before, in CETENA, a graphical user
In this work, the CD formulation was tested, in order to
interface was developed for the PROPACE code, in order
minimize the mean difference between experimental and
to handle the sensitivity analysis in automatic way, so to
numerical results.
optimise the calculation time. In the following pictures,
some diagrams of the second sensitivity analysis are
HUB CONFIGURATION
shown.
Hub zone 1:
Length [m]: 0.1166
Number of sections: 25
Hub zone 2:
Length [m]: 0.0583
Number of sections: 25
Hub zone 3:
Length [m]: 0.1166
Number of sections: 25
Hub zone 4:
Length [m]: depending on the blade root
section length in axial direction
Number of sections: 25
Hub zone 5:
Length [m]: 0.0583
Number of sections: 25
BLADE PANEL GRID
Number of Points in:
Spanwise direction 35
Chordwise direction 50
Points distribution law:
Spanwise direction COSIN
Chordwise direction COSIN

WAKE CONFIGURATION
Number of sections in axial 200
direction
Wake Pitch COSTANT
Wake contraction model No Contraction
Table 1

Figure 3

© 2005: Royal Institution of Naval Architects


Figure 4
Considering the difference in order of magnitude
between potential and friction contribution, this analysis
has the main effect on torque results, which is
characterized by the bigger errors in KQ prevision.
3.1(d) Kutta-Joukowsky Condition
The PROPACE solver can be set with two different
Kutta-Joukowsky condition models: the Morino Kutta
Condition (MKC) and the Iterative Kutta Condition
(IKC). The difference concerns the value of the doublets
that a panel method needs to consider on the blade wake.
Wake doublet formulations:
r r
( )
MKC : (∆φ)w = φ u − φl + U ∞ ⋅ rte
r r
( )
IKC : (∆φ)w = φ u − φl + U ∞ ⋅ rte + (∆W )w
Linear equation system
Np NW N p
 ∂φ 
∑ D φ +∑ W
j=1
ij j
m =1
im (∆φ) m = ∑ Sij   , i = 1,2, K, N p
j=1  ∂n  j
The MKC suggests to evaluate this mathematical entity
directly by the value of the potential on the last panels
(respectively on the back, upper face, and on the face,
lower face) at the trailing edge.
The IKC introduces a correction ∆W aimed to provide
the minimum pressure difference at the trailing edge, but
it has a lower stability due to the fact that, in this case,
the solver tries to minimize a numerical problem and
then the solution can be affected by a non physical
optimisation. For this work, the calculation was
performed with the MKC.
3.2 RANSE METHOD – CFX 5.7
The computation has been performed with a general
purpose code, CFX 5.7 [3], this code solves the Reynolds
Averaged Navier Stokes Equations on a structured multi-
block mesh. For the present study the steady viscous
flow around the propeller has been investigated in open
water condition. The flow around the propeller is
computed in a rotating reference system attached to the
Marine CFD 2005: Southampton, UK
propeller, a K-ε turbulence model in conjunction with a
scalable wall function method has been adopted in order
to keep the grid dimensions and thus reduce the
computational time.
The propeller has been tested onto three different
operating conditions, corresponding to J = 0. 736, J = 0.5,
J = 0.2. The different J value was achieved by tuning the
advance velocity value, and keeping fixed the revolution
number. All the calculation has been performed in model
scale.
A sensitivity analysis has carried out: since a scalable
wall function was adopted the grid of computation was
set up for J = 0. 736 that correspond to the case with
smallest thickness of boundary layer.
3.2(a) Meshing approach
The code employed for the mesh generation is the
structured multi-block code ICEM CFD – Hexa[7].
The domain of computation adopted, to investigate a
propeller in open water conditions, is a cylinder shaped
domain located around the propeller and co-axial with
the propeller axis. Considering the periodicity of the
propeller flow, only one blade can been modelled, the
influence of the other blades is taken into account by the
application of a periodic boundary condition.
The propeller geometry was already available in IGES
format (LE), whilst for the generation of the domain of
computation and the periodic surface, has been employed
a numerical methodology for the parametrical generation
of these surfaces, starting from the blade section
geometry, this technique is developed within CETENA’s
Panel Code PROPACE. The mathematical model consist
of several law to connect the chord line with the axial
direction, in the 2D cylindrical section along several
given blade section; starting from the leading and trailing
edge three curves are defined towards respectively the
inflow and outflow zone. From the blade up to the
farfield cylindrical zone the skew is reduced along the
radius.
In order to supply the unperturbed boundary condition
the section where the flow is directed inward the domain
has to be located at about 2D ahead the propeller blade,
the outflow section at about 4D behind and, and
cylindrical external border at about 5D. The skew of the
whole domain is 270°.
The RANS solver solves the equation in a Multi Frame
Reference (MFR) System, so the computational domain
has to be divided into two sub-domain:
• Inner domain, that includes the fluid close to the
propeller blade and a part of the shaft line, the
flow is solved in the rotating reference system,
(propeller reference system )
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
• Outer domain, that includes the fluid between
the inner domain and the domain boundary, the
flow is solved in a steady reference system.
OUTER domain
INNER domain
Figure 5: Viscous mesh
The different domain are connected by General Grid
Interface Algorithm (GGI), this tools allow to use non-
matching node distribution for the surface grid of
adjacent region, but if the number of nodes differs too
much the algorithm couldn’t perform properly. For this
reason first volume grid for inner domain has been
generated and then volume grids for outer domain have
been built up as a coarse distribution of the former, with
coefficient 1:3 or 1:4.
The RANS study has been carried out for the Leading
Edge EU Project, particular attention was given to key
features of this the project: the flow around the centre
and the tip region of the propeller in order to capture the
incipient tip vortex structure The taking advantage of
domain subdivision the size of the grid is increased close
to the blade whilst the domain size is decreased away
from propeller region. The total mesh Size is a 1.130.000
nodes structured multiblock.
© 2005: Royal Institution of Naval Architects
INNER DOMAIN: A H-Grid topology is applied to split
the volume, so the initial block is subdivided in 6 blocks;
then into the inner domain, close to the blade, an O-Grid
topology is applied with the generation of 5 new blocks:
two blocks are applied along the leading and trailing
edge (from the hub till about 0.8÷0.9 of the radius), one
along the tip region, the last two are applied respectively
for the pressure side and suction side. The O-Grid is
again divided to reflect the characteristic features of the
blade geometry. About 830,000 nodes. As an example,
the computation J =0.736, pointed out that the y+
parameter is in the range 5÷55.
OUTER DOMAIN: Total grid size of this region is about
200.000 nodes, made up by 16 blocks organised in two O
grid index. The first O-grid is located around the inner
domain to build up the grid at the domain interface as a
coarse distribution of the inner domain one, (reduction
coefficient 1:3 or 1:4). The second O-grid is applied to
generate a radial distribution from the hub to the farfield
zone.
3.2(b) RANSE solver settings
The commercial code CFX 5 was used to perform the
simulations. The governing equations of the viscous flow
are solved for the mass and momentum conservation. In
order to simulate a propeller operating in open water,
steady calculation is carried out in a Multiple Frame of
Reference System (MFR), the Rotating reference frame
is applied to the fluid domain close to the propeller blade
in order to add additional terms compared to those in the
inertial system. A Frozen Rotor algorithm with a GGI
interface guarantees the conservation of the fluid
properties at the domain connection. An algebraic multi-
grid algorithm is employed to accelerate the convergence
of the linear solver. For brevity, some information about
the solver configuration is shown in table 2, for further
information refer to the CFX User’s guide [12].
Velocity formulation Absolute velocity dependent
Spatial discretisation Finite-volume colloc.
Convection Terms Discr Upwind stnd
Order of acc. Conv. Terms Second
Diffusion Terms Discr Upwind
Order of acc. Diff.Terms Second
Pressure-velocity Coupling Fully coupled
Type of turb. model Two-eq., K-ε
Wall function Wall func without press.grad.
y
+
(5 ÷ 55)
Criteria: Residual No. of iterations 3100
Table 2: Solver Configuration
The K-ε turbulence model with a scalable wall function
was used to keep the grid dimensions and thus the
computational time. The final mesh size is about
1.130.000 nodes, and satisfy the requirement 5< y+<55
for the mesh spacing of the first point near the wall, in
the higher J (=0.736). The boundary condition applied to
the domain border correspond to: at the inflow section
and the Far-field region the undisturbed velocity is
applied, at the outflow section a “zero” pressure gradient,
 Marine CFD 2005: Southampton, UK

a Periodic boundary condition (GGI) is given at the
Periodic surface. The convergence has been achieved
with 3100 iterations and has been carried out using
different resolution scheme up to the second accurate
scheme order, the residual (RMS) are below 10E-06.
4. VALIDATION
The validation phase was carried out comparing the
numerical data with two experimental open water
condition data sets, provided by SSPA and HSVA (LE
Project).
4.1 EXPERIMENTAL DATA
In Figure 6, the comparison between CFX5, PROPACE
and experimental data is shown.
when the experimental data (KT or KQ) are quite close to
zero.
The viscous correction investigation was carried out for
model and full scale calculation. So, it was possible to
check the different effect of the Reynolds number on the
results. For confidentiality reasons, the detailed
formulations cannot be published, but they are indicated
by the following notations:
• “ACTUAL”: concerns the formulation
actually implemented in the PROPACE code;
• “LAMINAR”: concerns a formulation
involving a friction coefficient (CF) depending
on the Reynolds number in laminar condition;
• “TURBULENT”: concerns a
formulation involving a friction coefficient (CF)
depending on the Reynolds number in turbulent
condition;
• “ENVELOPE”: concerns a formulation
involving where the friction coefficient (CF) is
the maximum between the laminar and the
turbulent condition.

Otherwise, for every propeller scale calculation, the

Reynolds number formulation effect has been
investigated. The object of this study was the
characteristic length that has to be used for the evaluation
of the RN, and then of the viscous effect.
Figure 6 V ⋅ C( j) V ⋅ s(i, j)
R N chord = R R N arc = R
ν ν
J δKT[%] δ10KQ[%]
For each spanwise panel strip, two important
0.2000 4.382% 3.952% hydrodynamic greatnesses have been considered: the
0.5000 2.761% 5.436% total chord length and the arc length. Then eight different
0.7360 -1.809% 5.439%
calculation were performed for each propeller scale
dimension.
Table 3
4.2(a) MODEL scale results
4.2 VISCOUS CORRECTION RESULTS
The model scale calculations show that the mean δKT
As mentioned before, one of the validation phases
value is included in the range from 3.50 to 9.00%, while
concerns the optimization of the viscous correction
the δKQ in the range from 0.50 to 17.00%, depending on
formula, in order to reduce the error shown especially by
the viscous correction formulation used. These results
the KQ distribution results. The Open Water curves
concern the Reynolds number calculation based on the
obtained by the PROPACE application show a mean
total chord length.
difference about 12-15% on torque and 4-5% on thrust.
Evaluating the Reynolds number with the arc length,
This percentage gap between numerical and experimental
data is calculated as follows: only the KQ results seem to be better, the δKQ is smaller
than the previous calculation, especially for the turbulent
K − K T EXP K − K Q EXP
δK T [%] = T δK Q [%] = Q and then also the envelope viscous correction formula.
K T EXP K Q EXP
As mentioned before, it is important to understand that
Then, for each viscous correction formulation considered the δKT and δKQ coefficient values are affected by the
and respect to the experimental data, these parameters actual KT and KQ value respectively. So a big δKT or δKQ
make it possible to understand if the numerical result is result could mislead. In order to be right in judging the
overestimated or not. It is important to notice that these different formulations used, the following 2D diagrams
parameters can be very high, and then not so reliable, have to be analyzed. You can see that every formulation

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

considered leads to curves quite similar to experimental

ones. The KT curves are very close each other and they ARC LENGTH REYNOLDS NUMBER - δKT[%]
overestimate the thrust coefficient in the same way, with J
ACTUAL LAMINAR TURBULENT
ENVELOPE
Formulation Formulation Formulation
a more or less constant gap about 5.00%.
0.2000 4.66% 6.31% 5.87% 5.75%
0.5000 3.59% 6.16% 5.46% 5.29%
Differently, the KQ curves are characterized by a quite 0.7360 5.20% 9.45% 8.28% 8.00%
different slope and a difference varying respect to the
ARC LENGTH REYNOLDS NUMBER - δ10KQ[%]
experimental data. This behavior is more evident on the
ACTUAL LAMINAR TURBULENT
results provided by the chord length Reynolds number J ENVELOPE
Formulation Formulation Formulation
evaluation. 0.2000 11.72% 3.85% 6.17% 6.64%
0.5000 12.89% 1.68% 5.01% 5.67%
0.7360 17.02% 0.55% 5.48% 6.42%
Table 5: Model Scale Arc length RN results

4.2(b) FULL scale results

The full scale calculations lead to results similar to the

previous ones, but in this case it is possible to see
different KQ slope and offset: the numerical prevision is
very close to the experimental data, especially for the
higher advance coefficient values. By the 2D diagrams
and the following tables, you can see a better δKQ
Figure 7 behavior: the mean difference from the experimental
curves is reduced of 5.00%. In this scale calculation the
Reynolds number formulation seems to be not so
important. The two KQ behaviors have the same slope,
and also the numerical difference is very small.

The better formulation for the KQ evaluation seems to be

the laminar one, which is characterized by about 1.00%
of difference between the two Reynolds number
formulations.

The KT curves are affected by an increase of the error,

about 3.00%.

Figure 8

Using the arc length for the calculation of the Reynolds

number, the KQ curve presents a slope more similar to
the experimental curve slope, but in this case there is a
constant difference from the experimental data, greater
than in the case of the chord length Reynolds number.

CHORD LENGTH REYNOLDS NUMBER - δKT[%]

ACTUAL LAMINAR TURBULENT
J ENVELOPE
Formulation Formulation Formulation
0.2000 4.66% 6.33% 5.63% 5.63%
0.5000 3.59% 6.18% 5.09% 5.09%
Figure 9
0.7360 5.20% 9.47% 7.68% 7.68%
CHORD LENGTH REYNOLDS NUMBER - δKT[%]
CHORD LENGTH REYNOLDS NUMBER - δ10KQ[%] ACTUAL LAMINAR TURBULENT
J ENVELOPE
ACTUAL LAMINAR TURBULENT Formulation Formulation Formulation
J ENVELOPE
Formulation Formulation Formulation 0.2000 6.55% 7.54% 7.07% 7.07%
0.2000 11.72% 3.87% 7.25% 7.25% 0.5000 6.44% 7.98% 7.24% 7.24%
0.5000 12.89% 1.72% 6.55% 6.55% 0.7360 9.77% 12.31% 11.10% 11.10%
0.7360 17.02% 0.63% 7.72% 7.72% CHORD LENGTH REYNOLDS NUMBER - δ10KQ[%]
Table 4: Model Scale Chord length RN results J
ACTUAL LAMINAR TURBULENT
ENVELOPE
Formulation Formulation Formulation
0.2000 8.06% 3.40% 5.65% 5.65%
0.5000 7.75% 1.11% 4.32% 4.32%
0.7360 9.48% -0.24% 4.45% 4.45%
Table 6: Full Scale Chord length RN results

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

ARC LENGTH REYNOLDS NUMBER - δKT[%] In both these two formula, the resultant velocity at the
ACTUAL LAMINAR TURBULENT single spanwise position is given by:
J ENVELOPE
Formulation Formulation Formulation
VR = VA (ζ ) + (ω ⋅ ζ ⋅ R )
2 2
0.2000 6.55% 7.69% 7.15% 7.15%
0.5000 6.44% 8.21% 7.37% 7.37%
0.7360 9.77% 12.69% 11.30% 11.30%
5.1(a) J=0.200
ARC LENGTH REYNOLDS NUMBER - δ10KQ[%]
ACTUAL LAMINAR TURBULENT
J ENVELOPE
Formulation Formulation Formulation CFX5 calculation
0.2000 8.06% 2.70% 5.27% 5.28%
0.5000 7.75% 0.12% 3.78% 3.78%
At J=0.200 the CP distribution obtained by the RANSE
0.7360 9.48% -1.69% 3.67% 3.67%
calculation shows a typical low pressure distribution
Table 7: Full Scale Arc length RN results
close to the leading edge area. In this operating condition,
the propeller is characterized by a significant angle of
attack along most of the blade sections. This involves a
vortical phenomenon at the leading edge, originating on
the back and separating from the blade surface at about
r/R=0.900, as shown in Figure 11.

Figure 10

5. CFX5-PROPACE COMPARISON

5.1 BLADE PRESSURE DISTRIBUTION

In order to evaluate the propeller performances by the

light of the hydrodynamic behaviour of every blade
sections, in the following the CP distribution over the
blade is plotted with the same legend range for the CFX5
and PROPACE results. A direct comparison is shown in
2D diagrams, where the CP distribution is plotted for
three different radial sections.

The pressure coefficient considered is evaluated in

different way by the two solver because of their
boundary condition. In fact, the velocity distribution over
the blade surface is zero for the RANSE calculation (no
slip condition) and then the pressure coefficient can be
estimated by the pressure field with the formulation:
C P RANSE =
(p − p at ) − ρ ⋅ g ⋅ h
1
⋅ ρ ⋅ VR2
2

Otherwise, the panel methodology provides the velocity

field over the blade because of the ideal fluid hypothesis
on which this theory is based. So, the application of the
Bernoulli’s equation leads to this pressure coefficient
formulation:
2
 V 
C p PROPACE = 1 −  
 Figure 11: J=0.2, CFX
 VR 

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

PROPACE calculation
The CP distribution obtained by the PROPACE
calculation is shown in the picture below. The
comparison with the CFX5 results shows that in this
operating condition there are some problems in propeller
analysis.
It must be reminded that the CP color map related to the
PROPACE result shows a wide zone characterized by a
positive (dark green) pressure on the back. It is important
to notice that in this region there is very slight difference
between CFX and PROPACE results because the zero
value is indicated as the line dividing the positive (dark
green) to the negative (light green) area.

Because of the high angle of attack of the blade sections, The CP distribution on the face confirms the numerical
and consequently by the leading edge vortex starting in problems encountered in the central blade region,
the lower zone of the blade and quite close to it for all the especially at the leading edge.
leading edge extension, the computation carried out by
the panel method is affected by some numerical
problems. The most important are the fulfilling of the
Kutta-Joukowsky condition at the trailing edge and the
vortical phenomena that cannot be estimated at the
leading edge. Otherwise, the CP distribution over the
back side seems to be quite similar to the CFX5 result:
from r/R = 0.700 up to r/R = 0.900, the CP distribution at
the leading edge is not exactly foreseen. The effect of the
presence of a vortical phenomenon, in fact, makes the
pressure coefficient to be very small, as it can be seen in
the following 2D diagram, but it cannot be evaluated by a
potential code. For similar reasons, the pressure
difference at the trailing edge points out some numerical
problems, especially on the last panels on the trailing
edge.

Figure 13

Figure 12: J=0.2, PROPACE

© 2005: Royal Institution of Naval Architects


5.1(b) J=0.500
CFX5 calculation
As expected, in this operating condition, the CP
distribution on the back is thinner than the previous result.
The separating vortex at the blade tip is also thinner and
it follows the distribution of the lower CP. The blade
section works with reduced angle of attack, as it can be
seen on the 2D CP diagrams. The leading edge shows to
have a not negligible effect on the pressure distribution
over the blade surface. In fact, looking at the r/R = 0.900
CP distribution, the pressure coefficient at the trailing
edge is affected by a local peak of negative pressure, due
to the nearness of the leading edge to the blade surface.
Obviously, it is a viscous phenomenon that cannot be
expected from a potential code application.
Figure 14: J=0.5, CFX
Marine CFD 2005: Southampton, UK
PROPACE calculation
PROPACE calculation provided the following pressure
distribution. Like in J=0.200 operating condition, in this
case the leading edge vortex has not been predicted and
then the CP distribution at the leading edge shows some
differences from CFX5 results, especially in the central
zone. At the trailing edge, the pressure gap between back
and face has a coherent behavior with the previous J
calculation, but particularly reduced.
Figure 15: J=0.5, PROPACE
Figure 16: J=0.5, PROPACE
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
Figure 17
5.1(c) J=0.734
CFX5 calculation
For J=0.734 the CP distribution over the blade shows a
very slight region characterized by a non zero angle of
attack. This area is affected by a low CP distribution at
the leading edge on the back side, and a narrow high CP
strip on the face.
The blade sections closest to the hub seem to be affected
by a slight negative angle of attack, and this is the reason
a little negative CP distribution over the face occurs close
to the hub-blade intersection (Figure 18).
© 2005: Royal Institution of Naval Architects
Figure 18: J=0.736, CFX
At the outer radial positions, the blade section CP
distributions show to work in a slight angle of attack
condition. This is the reason a little pressure peak is
visible at the blade sections’ leading edge.
The effect of the leading edge closeness to the blade is
clearly shown in r/R=0.900 CP diagram. At the trailing
edge, the pressure distribution is affected by a local peak
due to the leading edge transit, Figure 21.
Figure 19 : J=0.736, CFX
PROPACE calculation
The CP distribution obtained by the PROPACE
calculation shows a pressure behavior over the blade
very similar to the RANSE results. In this case, the
operating condition is quite close to the project J value,
then the hydrodynamic behavior of the blade sections is
not particularly affected by angle of attack. The 2D CP
diagrams shows that up to the blade tip, the blade
sections work in shock-free mode.
 Marine CFD 2005: Southampton, UK

Figure 20: J=0.736, PROPACE Figure 21

6. CONCLUSIONS To this aim, the investigation carried out on the viscous

correction formula for the panel method allowed to point
The work detailed in the present paper allowed a precise out that it is possible to reduce this accuracy lost with
and detailed validation of the panel code PROPACE, different formulation to take into account the viscous
developed at CETENA for the study of the naval main effects.
propeller. A comparison study has been carried out
referring to numerical investigations with a commercial 7. ACKNOWLEDGEMENTS
RANSE solver founded by the European Project
LEADING EDGE. The work presented in this paper has been partly funded
by the EC though the project LEADING EDGE,
A first comparison has been carried out for the prediction of leading edge and tip flow for the design of
hydrodynamics coefficient in open water condition for a quiet and efficient screw propellers.
highly skewed propeller. The two codes have pointed out
a similar behavior for the trust coefficient KT, a Contract No: G3RD-CT-2002-00818
difference in a range 4-5% have been found with respect
to the available experimental data. Differently the torque
coefficient KQ shows different accuracy: a lost of 4-5%
between RANSE and experimental, whilst the accuracy
decrease to 12-15% for the panel – experimental
comparison.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

8. REFERENCES 9. AUTHORS’ BIOGRAPHIES

1. L. Morino - L.T. Chen, ‘A General Theory of Paolo Becchi holds the current position of researcher at
Unsteady Compressible Potential CETENA. He graduated in 2000 at the University of
Aerodynamics’, NASA, CR-2464, Dec. 1974 Genoa, Department of Naval Architecture and Marine
2. P. Becchi - A. Traverso, ‘Manuale Utente Engineering. He concluded Ph.D. course in 2004 at the
programma PROPACE Vers. 3.4’, Report Genoa’s University. Since March 2002 he is working at
CETENA 8420 CETENA, Hydrodynamics Department, especially
3. ‘CFX-5.7 Theory Manual’, www.ansys.com/cfx concerning hydrodynamic propeller analysis with CFD
4. T. Hoshino, ‘Hydrodinamic Analysis of calculation (panel method) and experimental test at the
propellers in steady flow using a surface panel Genoa’s University.
method’, Proceeding of the Spring Meeting, The
Society of Naval Architects of Japan, May 1989 Chiara Pittaluga, holds the current position of
5. T. Hoshino, ‘Hydrodynamic Analysis of researcher at CETENA. She studied Civil Engineering at
propellers in steady flow using a surface panel the Hydraulics Department of the University of Genoa
method (2nd Report: Flow Field around from which she graduated in 1999. Since then she is
Propeller)’, Proceeding of the Spring Meeting, working in the Hydrodynamics Department of CETENA;
The Society of Naval Architects of Japan, her work is mainly concerned with CFD calculations.
November 1989
6. T. Hoshino ‘A surface panel method with
a deformed wake model to analyse
hydrodynamic characteristics of propellers in
steady flow’, MTB195 April 1991
7. 2000 ICEM CFD Engineering, ICEM CFD
Version 4.3, Berkeley, USA.
8. H. Abbott – A. von Doenhoff, ‘Theory of Wing
Sections’
9. J. S. Carlton, ‘Marine Propellers and
Propulsion’, Butterworth Heinemann
10. H. E. Saunders, ‘Hydrodynamics in ship design’,
The Society of Naval Architects and Marine
Engineers
11. J. P. Breslin - P. Andersen, ‘Hydrodynamics of
ship propellers, Cambridge University Press
12. Cheng-I Yang, ‘Prediction of Propeller Blade
Pressure Distribution with a Panel Method’, Rep.
DTRC 90/013 May 1990
13. J. T. Lee, ‘A Potential Based Panel Method for
Analysis of Marine Propellers in Steady Flow’,
Ph D. Thesis, MIT 1987
14. P. Becchi - A. Traverso, ‘Analisi di sensibilità
sul programma PROPACE’, Report CETENA
8020

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

CONSIDERATION ON DEVIATIONS IN TORQUE PREDICTION FOR PROPELLERS

AND WATERJETS WITH RANS CODES
N W H Bulten, Wärtsilä Propulsion Netherlands BV / Technical University Eindhoven, The Netherlands
I A Oprea, Wärtsilä Propulsion Netherlands BV, The Netherlands

SUMMARY

RANS codes are often used to predict thrust and torque of propellers and waterjets. The results show quite well
correlation with measurements. It has been noticed however that torque is over-predicted by 2 to 5% in most cases.

In this paper an explanation is given for this phenomenon. The prediction of thrust and torque is related to lift and drag
of 2D profiles. Results of detailed RANS calculations of 2D profile sections are compared with experimental data for lift
and drag. With the normal k-ε turbulence model, lift is predicted well, but drag is over-predicted. This is due to an error
in the magnitude of the stagnation point pressure. The effect of increased 2D-section drag on the total thrust and torque
is evaluated for different blade angles.

3D RANS calculations of the DTRC 4119 propeller and a LIPS Jets waterjet installation are presented to show the actual
deviations in torque prediction.

NOMENCLATURE A typical problem in the performance prediction with

CFD is the determination of the required torque. This has
D diameter [m] a direct consequence for the prediction of the power
D drag [N] consumption and the efficiency, and it is therefore one of
F force [N] the most important performance indicators.
J advance Ratio = vship/nD [-]
Kq propeller torque coefficient [-] During the last few years, numerous CFD calculations
Kt propeller thrust coefficient [-] have been made of various propellers, ducted propellers
L lift force [N] and waterjet installations at Wärtsilä Propulsion
n angular velocity [1/s] Netherlands. Analysis of the flow is made with the
P power [W] commercial CFD code Star-CD. The CFD code uses the
P/D propeller pitch [-] finite volume method to solve the discretised set of
p static pressure [N/m2] RANS equations. The calculations showed the general
Q volume flow [m3/s] trend that torque of propellers and mixed-flow pump
q 2D source strength [m2/s ] impellers was over-predicted.
r radius [m]
v velocity [m/s] The paper can be split into three parts. First detailed CFD
β blade angle [deg] analyses of 2D test sections will be presented. These
ε drag over-prediction factor [-] calculations give an indication of the obtained accuracy
νΤ turbulent viscosity [m2/s] with the current method. The following section will
ρ density [kg/m3] address the transformation from lift and drag to thrust
θ angle [rad] and torque. Effect of errors in profile drag prediction will
be calculated for various pitch settings (equivalent to
1. INTRODUCTION impeller blade angles). The results of the actual CFD
calculations of a propeller and a waterjet will be shown
For a very long time development of propulsion systems in sections 4 and 5. Conclusions are formulated in the
for ships has been based on model scale experiments. last section.
Prediction of full-scale performance requires an
extrapolation method. It is known however that these 2. CALCULATIONS ON TEST SECTIONS
methods do not always give accurate performance
predictions. Use of computational fluid dynamics (CFD) In order to get better insight in the accuracy of the
methods based on Reynolds Averaged Navier-Stokes calculations of a complete 3D propeller or waterjet
(RANS) equations can possibly eliminate the problems impeller is it helpful to make detailed calculations of 2D
of the extrapolation methods. Introduction of CFD in the profile sections first. Calculations of 2D profiles can be
maritime world has been rather slow due to the moderate validated with experimental data available in open
accuracy of the performance predictions based on RANS literature. Geometries, which allow analytical solution of
codes in the past. the pressure distribution, are also suitable for validation
of CFD methods. Even analytical solutions based on
irrotational flow may be useful, if the effect of boundary
layer development is negligible.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Three geometries have been analysed. The first geometry
is a half body test case, which allows analytical solution
of the pressure distribution along the surface. This
geometry is used for detailed study of the flow at the
stagnation point. The remaining two profiles are NACA
profiles, which are analysed as isolated profile
(NACA0012) or in a cascade (NACA 65-410).
away. The third region is an intermediate region, which
fills the space between the first region and the outer
region. The complete mesh and a detailed view of the
mesh near the stagnation point are shown in figure 2.
Only half of the geometry is meshed due to symmetry of
the flow field.

2.1 2D HALF BODY TEST CASE

2.1(a) Analytical solution of pressure distribution

The geometry of the half body is based on the dividing

streamline, which is formed when a uniform flow and a
source are combined. The stream function for this
geometry is given by [1]:

q
ψ = θ + vr sinθ (1)
2π

where q is the strength of the source and v the uniform

velocity. A sketch of the geometry and some streamlines
is shown in figure 1.

r Figure 2: Mesh of Half Body and Detailed View at

θ
Stagnation Point

At the upstream boundary an inlet velocity condition has

Figure 1: Streamlines along half body
The analytical solution of the pressure distribution along
the surface can be calculated from the stream function, if
the flow is assumed to be irrotational. The resulting non-
dimensional pressure distribution is calculated as:
been applied. Here the three velocity components and the
turbulence intensity and the turbulent length scale are
prescribed. On the other boundaries constant pressure
conditions have been used. A symmetry condition is used
at the centreplane. Calculations have been made with the
standard k-ε turbulence model. Coupling between mass
and momentum equations is achieved with the simple
algorithm.
2.1(c) Results of pressure distribution

p − p0  sin θ  2 sin(2θ )  The results presented here are based on a systematic

Cp = = −   +  (2) study of the effect of variation of turbulence intensity and
1  π −θ  π −θ 
ρv 2
  turbulent length scale. The turbulence intensity is varied
2 between 0.5% and 2.0%. The length scale is set to values
between 0.01 m and 0.3 m. The maximum of 0.3 m is
where θ is defined according to figure 1. According the equivalent to 5% of the width of the inlet. Figure 3 shows
expectation, the non-dimensional pressure at the the comparison of a numerical analysis with the
stagnation point (θ=π) is 1.0. analytical solution. This analytical solution does not take
effects of boundary layer development into account
2.1(b) Set-up of numerical model however.
A 2D mesh has been made consisting of three regions.
The first region is a thin layer around the half body to
ensure a high quality mesh for the solution of the
boundary layer. The second region is a rectangular outer
region to place the boundary conditions sufficiently far

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

A clear trend of the overall inlet turbulence level on the

Pressure distribution along half body stagnation point pressure is observed. Moreover a small
second order effect of the variation of turbulence
1.2 intensity is noticeable. The error in stagnation pressure
Analytical solution can be around 25% for a reasonable turbulence level at
1
CFD result
the inlet.
0.8 Stagnation point pressure

0.6 1.35

0.4
Cp [-]

1.3

0.2
1.25
0
0 30 60 90 120 150 180
1.2
-0.2

Cp [-]
-0.4 1.15

-0.6
Angle α=180-θ [degrees]
Figure 3: Comparison of Analytical Solution and
Calculated Pressure Distribution along Half Body
Difference between the calculation and the exact solution
at the stagnation point is clearly noticeable. The error of
the pressure coefficient found in the calculation is 19%.
The deviation of the minimum pressure is much smaller
though. The relative error is 3.7%. The deviation in the
minimum pressure can be attributed to the development
of the boundary layer along the surface, which is
neglected in the analytical solution. The deviation at the
stagnation point remains an error in the CFD calculation
however.
2.1(d) Effect of Turbulence Model Input Parameters
Turbulence is governed by the turbulence intensity (TI)
and the length scale l. The length scale is related to the
amount of dissipation. Small eddy lengths lead to larger
dissipation. An increase of turbulence intensity or length
scale will both give a higher turbulent viscosity. All input
parameters can be combined to a single parameter, which
represents the turbulence level:
νT 2 9
= C 1µ 4 ⋅ ⋅ TI ⋅ l =
v in 3 20
where νT is the turbulent viscosity, vin the inlet velocity
and Cµ a closure coefficient of the k-ε turbulence model
(equal to 0.09).
Figure 4 shows the results for a number of calculations
with varying turbulence intensity and length scales. In
this figure the stagnation point pressure is plotted as
function of the turbulence level at the inlet boundary.
1.1
TI = 0.5 %
1.05 TI = 1.0 %
TI = 1.5 %
TI = 2.0 %
1
0 0.5 1 1.5 2 2.5 3
Turbulent viscosity at inlet/V_in
Figure 4: Stagnation Point Pressure as Function of
Turbulence Level at Inlet
Explanation of the effect of the turbulence level on the
stagnation point pressure requires a detailed analysis of
the applied turbulence model. This is beyond the scope
of this paper. It can be noted however that the cause of
the error can be found in the turbulent production term
(see [2]). This equation is used in both k-ε as well as k-ω
turbulence models. Therefore it is to be expected that
both two-equation turbulence models will show similar
behaviour.
Over-prediction of the pressure in the stagnation point
will result in a larger value of the pressure drag of a
profile. So it is to be expected that total drag of a profile
is over-predicted with a CFD method, which uses the
standard k-ε turbulence model.
⋅ TI ⋅ l (3) 2.2 NACA 0012 PROFILE
Calculation of the flow along a half body is of pure
theoretical use. In order to get better insight about the
practical implications of the inaccuracies in the flow
prediction a number of well-known NACA profiles have
been analysed. In this section the results of calculations
of an isolated NACA0012 profile will be presented.
Experimental data of lift and drag is available for this
profile.

© 2005: Royal Institution of Naval Architects


2.2(a) Set-up of Numerical Model
The mesh for the numerical analysis is based on a
structure with three regions, which is similar to the 2D
half body mesh to a certain extent. The inner region
consists of an O-grid around the complete profile section.
A very small radius is modelled at the trailing edge of the
profile to have sufficient curvature for the O-grid
topology. Around the O-grid a second region is created,
which has an approximately rectangular outer shape.
Both regions can be rotated, in order to vary the angle of
attack of the profile. In this way the mesh near the profile
is identical for all calculated conditions. The mesh of the
complete domain and the detailed view of the mesh
around the profile is shown in figure 5.
Figure 5: Mesh of Complete Domain and Detailed View
around NACA0012 Profile
At the upper and lower boundary of the domain two
types of boundary conditions can be applied: wall
boundary conditions or constant pressure boundary
conditions. The first type can be used if the experimental
data is obtained from wind tunnel tests. The second type
is suitable for the calculation of lift and drag in an
unbounded region. For sufficiently large numerical
domains both types will give the same results. At the
inlet boundary a uniform velocity profile is prescribed.
Turbulence intensity is set to 0.5% and the length scale is
set to a small fraction of the tunnel height. All
calculations are made with the standard k-ε turbulence
model.
Marine CFD 2005: Southampton, UK
2.2(b) Results of Lift and Drag Calculations
Figure 6 shows a comparison of the calculated and
measured lift and drag coefficients for the isolated
NACA0012 profiles. Agreement is good for the lift over
the whole range of angles of attack. The comparison of
the calculated and measured drag shows a clear offset.
This trend is in line with the error in the stagnation point
pressure prediction. Similar results are found for
NACA0006 and NACA0015 profiles.
Lift and drag comparison NACA0012
1 0.04
Cl_exp
Cl_CFD
Cd_exp
Cd_CFD
0.75 0.03
CD [-]
CL [-] 0.5 0.02
0.25 0.01
0 0
0 1 2 3 4 5 6 7 8 9
Section angle of attack [degree]
Figure 6: Comparison of Measured and Calculated Lift
and Drag for NACA0012 Profile
2.3 NACA 65-410 CASCADE
If a waterjet impeller is analysed with aid of 2D sections,
then the cascade effect should be taken into account.
Extensive experimental data is available on tests with
NACA 65 compressor blade profiles [3]. Tests were done
with NACA 65 profiles with various camber lines and a
maximum thickness of 10% of the chord. Cascades of
profiles can be described with two additional parameters.
These are the solidity and the blade angle. The solidity
represents the distance between two profiles in relation to
the chord length. The blade angle is defined as the angle
between the profile base line and the line connecting all
leading edges. Calculations are made with a NACA65-
410 profile with a blade angle of 70 degrees and a
solidity of 1.0. Comparison of calculated and measured
pressure distribution along the profile surface showed
good agreement for various angles of attack.
Comparison of lift and drag for the NACA65-410
cascade shows similar trends as the isolated NACA0012
profile. This is shown in figure 7.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

 P 
Lift and drag coefficients NACA65_410 cascade
β = atan D  (5)
(s/l=1.0, beta=70)  0.7 ⋅ π 
0.8 0.1  
Measurements Typical values of the blade angle in waterjet mixed flow
0.7 CFD_results 0.09 pumps are in the range of 18 to 25 degrees. This is
Measurements equivalent with a P/D range of 0.7 to 1.0.
0.6 CFD_results 0.08

0.5 0.07 Over-prediction of profile drag in a CFD analysis can be

expressed with a factor ε:
0.4 0.06
DCFD = Dexact (1 + ε ) (6)

Cd [-]
Cl [-]

0.3 0.05

0.2 0.04 This factor ε can be up to about 0.25, based on the results
presented in the previous section. Effect of over-
0.1 0.03 prediction on thrust and torque can already be noticed in
0 0.02
figure 8.

-0.1 0.01

-0.2 0
-4 0 4 8 12 16 20
Angle of attack [degrees]

Figure 7: Comparison of Measured and Calculated Lift

and Drag for NACA65-410 Profiles in Cascade with
Solidity of 1.0 and Blade Angle of 70 Degrees

2.4 CONCLUDING REMARKS Figure 8: Sketch of Forces Acting on Profile

Over-prediction of drag can be related to the over- Besides this graphical representation is it also possible to
prediction of static pressure at the stagnation point. This calculate the relative error in axial and tangential
is due to an over estimation of the turbulence production, direction. With eqn. (6) substituted into eqn. (4) the
and this is apparent in commonly applied two-equation relative error of the axial force becomes:
turbulence models.
− sin β ⋅ (1 + ε )
L
It can be concluded that the trend of over-prediction of Fax _ CFD cos β ⋅
= D (7)
drag is applicable to both isolated as well as cascade L
Fax _ exact
profiles. Lift is predicted significantly better for both cos β ⋅ − sin β
configurations. D

where L/D is the lift over drag ratio. The relative error in
3. TRANSFORMATION FROM LIFT AND tangential direction yields:
DRAG TO THRUST AND TORQUE
+ cos β ⋅ (1 + ε )
L
Ftan _CFD sin β ⋅
Torque and thrust of a propeller or waterjet impeller are = D (8)
based on the experienced tangential and axial forces Ftan _exact L
sin β ⋅ + cos β
acting on the blade profile sections. These forces can be D
related to the lift and drag of these profiles. Figure 8
shows a sketch of the different forces. The The actual relative errors in axial and tangential force are
transformation of the forces is defined as: plotted in figures 9a and 9b for a blade angle of 20 and
25 degrees as function of the lift over drag ratio. The
Fax = cos β ⋅ L − sin β ⋅ D drag over-prediction factor is set to 10%, 20% and 30%
(4) to show the sensitivity.
Ftan = sin β ⋅ L + cos β ⋅ D
These diagrams show clearly that the effect on axial
where β is the blade angle. This angle can be derived force is very small even for a quite large over-predicted
from the pitch at 0.7 radius with: drag. On the other hand the error in tangential force
remains significant for realistic values of ε (0.25) and
L/D (35). Comparison of both diagrams also reveals a

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

clear influence of the blade angle. Smaller angles result turbulence model seems to be impossible. Application of
in a larger error in tangential force. a correction factor might be an acceptable solution for
this fundamental problem.
Relative error in axial and tangential force for
blade angle of 20 degrees
4. CFD ANALYSIS OF DTRC 4119
12%
F_ax - Epsilon = 0.1 PROPELLER
F_ax - Epsilon = 0.2
10% F_ax - Epsilon = 0.3
For the analysis of the flow along an open propeller the
F_tan - Epsilon = 0.1
well-known geometry of the DTRC 4119 propeller has
F_tan - Epsilon = 0.2
been used. Flow around this propeller has been
8%
F_tan - Epsilon = 0.3
investigated experimentally. Results of performance
measurements and LDV are presented in [4].
F_CFD/F_exact [-]

6%
4.1 SET-UP OF NUMERICAL MODEL

4% For the generation of the mesh a semi-automated mesh

generator for propellers has been used. This specific
mesh generator has been developed in order to be able to
2%
mesh a large variety of different propeller geometries.
Close to the blade an O-grid is applied to have high
0% quality hexagonal cells in the blade boundary layer. The
0 10 20 30 40 50 remainder of the domain between two blades is filled
with additional hexagonal cells. Due to the uniform axi-
-2%
symmetrical inflow is it sufficient to model only 1
L/D ratio [-] propeller blade with periodic boundaries to simulate the
Figure 9a: Relative Errors in Tangential and Axial Forces flow of the complete propeller. The surface mesh is
due to Over-Prediction of Profile Drag. shown in figure 10.
Blade angle is 20 degrees

Relative error in axial and tangential force for

blade angle of 25 degrees
12%
F_ax - Epsilon = 0.1
F_ax - Epsilon = 0.2
10% F_ax - Epsilon = 0.3
F_tan - Epsilon = 0.1
F_tan - Epsilon = 0.2
8%
F_tan - Epsilon = 0.3
F_CFD/F_exact [-]

6%

4%

2%

Figure 10: Surface Mesh of 4119 Propeller

0%
0 10 20 30
-2%
L/D ratio [-]
Figure 9b: Relative Errors in Tangential and Axial
Forces due to Over-Prediction of Profile Drag.
Blade angle is 25 degrees
The error in tangential force remains quite large for
configurations with realistic values of the blade angle,
the lift over drag ratio and the relative error in drag
prediction. Extreme accurate predictions of required
torque with an error of less than 0.25% with normal k-ε
(mesh has been copied for visualisation purposes only)
40 50
Upstream of the propeller an inlet boundary is applied,
where a constant uniform inflow velocity is prescribed.
At the outer radius and the downstream plane of the
numerical domain a constant pressure boundary
condition is applied. This type of boundary condition
enables mass fluxes through the boundary, depending on
the flow field solution near the boundary.
Rotation of the propeller is implemented with the quasi-
steady Multiple Frame of Reference (MFR) method. Part
of the mesh is placed in a rotating frame of reference in
this method.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

For all calculations the k-ε turbulence model is used. For

DTRC 4119 propeller performance
spatial discretisation a second order method is applied.
0.8
Coupling between mass and momentum is established
with simple algorithm. Convergence speed is increased
with an algebraic multi-grid method. 0.7

4.2 RESULTS OF PROPELLER ANALYSIS 0.6

Performance of the propeller is based on the evaluation 0.5

of thrust and torque. Both are derived from the forces

Kt, 10Kq [-]

acting on the blades. These forces take into account both
0.4
the normal pressure forces as well as the wall shear stress
forces. Figure 11 shows the pressure distribution on the
suction side of the blades for an advance ratio J=0.833. 0.3

KT_CFD
0.2
KT_EXP
10Kq_CFD
0.1 10Kq_EXP
Eta_CFD
Eta_EXP
0
0 0.2 0.4 0.6 0.8 1 1.2
J [-]

Figure 12: Comparison of Calculated and Measured

Propeller Performance

5. CFD ANALYSIS OF LIPS JETS

WATERJET INSTALLATION

5.1 SET-UP NUMERICAL MODEL

The mesh of the complete waterjet installation is based

on two separate numerical models of the waterjet mixed-
flow pump and the inlet duct. The mesh for the inlet is
created with aid of the CFD waterjet inlet analysis tool.
This tool automatically creates a multi-block mesh with
Figure 11: Pressure Distribution at Suction Side at
hexagonal cells, according to a certain topology. Results
J=0.833
of the validation process of this inlet analysis tool have
been presented (see [5], [6]).
Propeller performance is presented in the usual non-
dimensional coefficients for advance ratio, thrust and
In a similar way as the validation project for the waterjet
torque. These coefficients are defined as:
inlet ducts, a project has been executed for the validation
v
J = ad of the numerical model of the mixed-flow waterjet pump
nD ([7], [8]). This provided information about the required
mesh density for the pump and the stator bowl. In order
Thrust
Kt = (9) to create a high quality mesh, it is decided to define an
ρ n 2D 4 O-grid around the impeller blades. This topology is
continued along the hub surface. The channels between
Torque the blades are filled with a multi-block hexagonal mesh.
Kq =
ρ n 2D 5 The leading edge regions are meshed with a very fine
Figure 12 shows the comparison of the calculated and the mesh in order to capture the suction peak accurately.
measured propeller thrust and torque. Special attention is paid to the tip clearance region. The
gap between the rotating impeller and the stationary
Agreement of the propeller thrust is quite good at the seatring is filled with some layers of extrusion cells, to
best efficiency point (J=0.9) and acceptable over the allow flow over the blade tips. Comparisons with
remainder of analysed conditions. The torque is over- calculations without tip gap showed a notable difference;
predicted, which is according to the expectations. The therefore inclusion of the tip gap cells has become
trend of both Kt and Kq as function of J is reproduced default approach.
well within the CFD model.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

The inlet and pump mesh are connected with a non-
matching coupling method at a straight plane just
upstream of the pump. The final mesh of the complete
waterjet is shown in figure 13. Near the walls of the inlet
ducting a thin layer of extrusion cells is used, similar to
the tip gap cells. This controls the level of y+ values in
the boundary layer.
First a comparison of the predicted volume flow through
the waterjet installation is made. The flow is determined
by integration of the velocity at the nozzle exit plane.
Table 1 shows the calculated volume flow and the
estimated flow from the waterjet performance prediction
software (WPPS) as used at Wärtsilä Propulsion
Netherlands. This software provides very accurate
predictions, and it is in use for many years already.
Differences in flow prediction are less than 1% for all
conditions. Agreement is thus very good. The CFD
results can be regarded to be accurately enough for
further analysis.

V_ship Q_WPPS Q_CFD Difference

[knots] [m3/s] [m3/s] [%]
31 12.96 12.93 -0.23 %
35 13.11 13.08 -0.19 %
37 13.26 13.19 -0.54 %
Figure 13: Mesh of Complete Waterjet Installation 39 13.41 13.31 -0.72 %
41 13.55 13.49 -0.45 %
At the inflow plane an inlet boundary condition is Table 1: Volume Flow Prediction Based on Performance
applied, which takes into account the non-uniform Prediction Software and CFD Calculations
velocity distribution of the hull's boundary layer. At the
side planes, the bottom plane and the outflow plane is a Thrust and torque are determined in the same way as for
constant pressure boundary condition prescribed. This the open propeller. Forces of the wall cells connected to
enables inflow and outflow depending on the flow field the impeller are summarised to get the torque and axial
near the boundary. At the nozzle outlet plane constant force. Figure 14 shows the static pressure distribution on
atmospheric pressure is prescribed too. This has the a part of the complete waterjet installation. A detailed
consequence that the flow through the waterjet system is analysis of the results reveals a variation of the pressure
part of the solution. This flow depends on the produced distribution on the different blades. This is due to the
head of the pump at given RPM and the total resistance strong non-uniform inflow velocity distribution to the
of the inlet duct and the nozzle. pump. This phenomenon is well known in waterjet
applications [9].
All numerical settings are equal to the open propeller
calculations. This means that all calculations are made
with the k-ε turbulence model. Rotation of the impeller is
implemented via the multiple frame of reference method.
Mass and momentum are coupled via simple algorithm
and the equations are solved with a second order scheme
and an algebraic multi-grid method.

5.2 RESULTS OF WATERJET ANALYSIS

Variation in the operating condition can be obtained by

variation of the ship speed or the pump revolutions for a
given geometry. Compared to open propellers, variation
of the ship speed has only little influence on the final
operating point of the waterjet pump. This is due to the
fact that the mixed-flow pump is an internal flow Figure 14: Pressure Distribution at Inlet Duct and
machine, whereas the propeller can be regarded as an Impeller
external flow machine.
Figure 15 shows the required power based on the
In this paper results of calculations with a constant pump waterjet performance software and the CFD results. The
RPM and varying ship speed are presented. It is to be offset between prediction and CFD results can be seen
expected, based on simple waterjet propulsion theory and clearly. At the design point of the waterjet the over-
the characteristics of the mixed-flow pump, that the prediction is 2.4%. This is contributed to the error in
delivered thrust and the required power will decrease prediction of the stagnation point pressure.
with increasing ship speed and constant pump RPM.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK
Power comparison
110.0%
prediction software
107.5%
CFD prediction
105.0%
Power/Power_design_point [%]
102.5%
100.0%
97.5%
95.0%
92.5%
90.0%
30 35 40 45
Ship speed [knots]
Figure 15: Comparison of Required Waterjet Power
Based on CFD Calculations and Waterjet Performance
Prediction Software
6. CONCLUSIONS
• Lift of 2D profiles can be predicted very accurate
with RANS codes with standard k-ε turbulence
model. Drag is over-predicted quite significantly
however. Deviations can be up to 25%.
• Transformation of lift and drag to thrust and torque
redistributes the error in drag prediction. The major
influence is found in torque and only little effect is
seen on thrust. This redistribution depends on the
actual blade angle, or pitch setting, however.
• Open water calculations of open propellers give
quite accurate thrust prediction at the best efficiency
point of the propeller. Over the remainder of the
operational range also an acceptable accuracy in
thrust prediction is found. Torque is over-predicted
over the whole range of operating points however.
• Calculations of the flow through a complete waterjet
installation provide accurate results. Prediction of
the volume flow through the system shows
deviations of less than 1%. Torque, and
consequently power, is over-predicted by about
2.5% at the design point. This offset remains more
or less constant over a large range of operating
conditions.
7. ACKNOWLEDGEMENTS
The author would like to thank Wärtsilä Propulsion
Netherlands for supporting the publication of this paper,
and for the provision of the performance data and the
computational resources.
© 2005: Royal Institution of Naval Architects
8. REFERENCES
1. Fox, R.W., McDonald, A.T., 'Introduction to fluid
mechanics', Third edition, John Wiley & Sons, New
York, 1985
2. Moore, J.G., Moore, J., 'Controlling over-production
of turbulence in two-equation models by limiting the
anisotropy of the Reynolds normal stresses, ASME
Fluids Engineering Division Summer Meeting, 1997
3. Emery, J.C., Herrig, L.J., Erwin, J.R., Felix, A.R.,
'Systematic two-dimensional cascade tests of NACA
65-series compressor blades at low speeds', NACA
report 1368, Langley Aeronautical Laboratory,
Langley Field, 1958
4. Jessup, S.D., 'An experimental investigation of
viscous aspects of propeller blade flow', PhD thesis,
Washington, D.C., 1989
5. Verbeek, R., Bulten, N.W.H., 'Interpretation of
model scale test results with aid of CFD
calculations', Proc. Waterjet Propulsion III
conference, Gothenburg, 2001
50 6. Bulten, N.W.H. Verbeek, R., 'Design of optimal inlet
duct geometry based on vessel operational profile',
Proceedings FAST 2003 conference, Session A2, pp
35-41, Ischia, Italy, 2003
7. Bulten, N.W.H., Van Esch, B.P.M., 'Calculation of
radial forces due to non-uniform entrance flow in a
mixed-flow waterjet pump', Proc. 12th Int. Conf. On
Fluid Flow Technologies, Budapest, 977-983, 2003
8. Bulten, N.W.H., Verbeek, R., 'CFD simulation of the
flow through a waterjet installation', Proc. Waterjet
Propulsion IV conference, London, 2004
9. Verbeek, R., Bulten, N.W.H., 'Recent developments
in waterjet design', RINA Waterjet Propulsion II
conference, Amsterdam, 1998
9. AUTHORS’ BIOGRAPHIES
Norbert Bulten works at the Propulsor Technology -
Hydrodynamics (PTH) department of Wärtsilä
Propulsion Netherlands BV. He is responsible for CFD
and FEM projects for waterjets and propellers. He is
preparing a PhD thesis at the Technical University
Eindhoven (TU/e) about the effects of non-uniform
inflow to a waterjet pump
Iulia Oprea follows a Marie Curie fellowship and she is
located at the Propulsor Technology - Hydrodynamics
(PTH) department of Wärtsilä Propulsion Netherlands
BV. She is working on a project about implementation
and validation of CFD calculations for propellers.
Marine CFD 2005: Southampton, UK

INFLUENCE OF SCALE EFFECTS ON THE HYDRODYNAMIC CHARACTERISTICS

OF PROPELLERS
S-B Müller and M Abdel-Maksoud, University Duisburg-Essen, Germany

SUMMARY

To analysis the scale effects on the thrust and torque coefficients of propellers, the viscous flow around a propeller
geometry was calculated for 3 different advance ratios at 6 Reynolds numbers. The thrust and torque coefficients of the
full-scale were estimated using the CFD-results and the extrapolation procedure of the International Towing Tank
Conference (ITTC).

The calculated thrust and torque coefficients using the CFX-5.7 program show much higher dependence on the Reynolds
number than the estimated values using the ITTC-procedure. While a remarkable increase of the calculated thrust
coefficient based on CFD-results takes place with a growing Reynolds number, the estimated increase of the thrust
coefficient using the ITTC-method is very limited.

The comparison of the normalised pressure and wall-shear stress distributions at different Reynolds numbers is very
helpful in understanding the reasons for the change of the thrust and torque coefficients with respect to the Reynolds
number. The normalised pressure on the suction side of the propeller blade decreases when increasing the Reynolds
number. The same is valid for the wall-shear stress. This reduction in the wall-shear stress is stronger in the leading edge
region.

NOMENCLATURE provided by Mewis and Klug [1]. Since the dimensions

of the test facilities in the towing tanks are not expanded
δKQ torque coefficient difference (--) to consider these ship sizes, the used scale factor
δKT thrust coefficient difference (--) increases. With the increase of the scale factor, the
δP pressure difference (N m-2) problem of extrapolating the measured model results to
η efficiency (--) the full-scale is intensified substantially. This is to be
η-full efficiency of full-scale ship (--) attributed to the fact that the used extrapolation
η-model efficiency of model (--) procedures are applied in a new range, for which only
limited experience exists.
λ scale factor (--)
CFD computational fluid dynamics (--)
The progress in the area of numerical methods for
CTh thrust loading coefficient (--)
viscous flow computation opens new perspectives and
c/D ratio of profile chord length to
possibilities to gain extensive information about the
propeller diameter (--)
characteristics of the flow, e.g. the distribution of the
D propeller diameter (m)
wall-shear stress at the propeller blade. This information
dh/D hub diameter relationship (--)
is very important for the correct prediction of the
ITTC International Towing Tank Conference
propulsion characteristics of a new design. Such
J advance ratio coefficient (--)
information cannot be generated experimentally or with a
KQ torque coefficient (--)
help of potential flow-theory based methods.
KQ-full torque coefficient of full-scale ship (--)
KQ-model torque coefficient of model (--)
2. PROBLEM DEFINITION
KT thrust coefficient (--)
KT-full thrust coefficient of full-scale ship (--)
In an industrial project the thrust and the torque
KT-model thrust coefficient of model (--)
coefficients of a propulsion system are predicted mainly
n revolution per unit time (s-1)
experimentally in model scale. The carrying out of open
P pressure (N m-2)
water tests in towing tanks is highly optimized. As a
P/D pitch ratio (--)
consequence the costs and time required for the
Rn Reynolds' number (--)
experimental investigation of the hydrodynamical
SST Shear Stress Transport
characteristics of a propulsion system are clearly
t/c profile thickness-ratio (--)
minimized. The improvement in the measuring technique
V velocity (m s-1)
makes it possible that the accuracy of the measuring data
and the reliability of the measuring systems achieve a
1. INTRODUCTION
high level. However, the extrapolation of the results of
the measurement to full-scale remains a problem which
When ship sizes are increased, particularly in the case of
so far has been treated insufficiently.
container ships, the diameter of the appropriate
propellers becomes larger. An overview of the propeller
diameters used by current container ship projects was

© 2005: Royal Institution of Naval Architects


In the standard ITTC 1978 procedure for performance
prediction, two corrections are used; one for the thrust
and one for the torque coefficient. These corrections
consider the influence of the Reynolds number (Rn), the
profile thickness-ratio (t/c) and the pitch ratio (P/D).
Thus, the ITTC procedure is hardly able to consider the
local flow conditions like e.g. the blade contour and skew
of the propeller blade. That is also not the aim of the
ITTC-procedure, which has to be oriented towards
practical solutions for reasons of costs. The relevance of
the problem has been recognized by various scientists.
An example of this is the work of Meyne [2] regarding
the development of the procedure of the equivalent
profile.
3. NUMERICAL INVESTIGATION
For the investigation of the scale effects for propellers,
the viscous flow around a propeller geometry was
calculated at different Reynolds numbers, see Tables 1, 2
and 3. The investigated propeller is a modified geometry
of the Wageninger B-Series propeller B4.55. It has a P/D
ratio of 0.9. The hub diameter relationship (dh/D) is 0.25.
The profile thickness relation at the dimensionless radius
0.7 is 0.03867. The ratio of profile chord length to
propeller diameter (c/D) is 0.2948. The investigations
were carried out for the following propeller diameters:
model: 0.25 m, full-scale: 4, 6, 8, 10 and 12 m. The
numerical computations were carried out for three
advance ratios J = 0.8, 0.7 and 0.6.
In the all computations the flow is considered to be fresh
water at a temperature of 25 C°. In order to keep the
advance ratio constant for the full-scale, the inflow
velocity to the propeller was varied. In the model scale
both the rotation and the inflow velocity were varied.
The computed thrust loading coefficient (CTH) is
included in Tables 1, 2 and 3.
The numerical investigations were carried out for open
water test condition, i.e. with parallel inflow. Therefore,
only one propeller blade was regarded in the
computations. The interaction effect with other blades
was considered by the application of a periodic boundary
condition in space. The dimensions of the computation
domain must be large enough to avoid the influence of
the boundary conditions on the numerical results. The
diameter of the calculation domain was 5.5 D; its length
is 20 D. The number of grid points is approx. 9x105. The
number of points was constant over all computations.
The grid lines on the blade are represented in Figure 1.
The numerical computations of the viscous flow around
the propeller were carried out with the help of the
program CFX-5.7 by ANSYS. The procedure
implemented in CFX-5.7 for the computation of the flow
in rotating systems corresponds to the existing procedure
in the TASCflow code, which is validated in numerous
investigations for propeller flows [3 - 8].
Marine CFD 2005: Southampton, UK
A rotating coordinate system is applied in the
computation, which turns with the same number of
revolutions and direction of rotation as the propeller. The
x-axis of the coordinate system is the axis of rotation of
the propeller. In the numerical computations, the whole
computing domain and not only the surrounding region
to the propeller turns. The SST turbulence model was
applied in all computations. The computed streamlines
for the Reynolds number 1.24x107 are represented for
J = 0.8 in Figure 2. In the computation for all Reynolds
numbers and advance ratios, a stable convergence
behavior could be achieved.
4. RESULTS OF THE STUDY
4.1 THRUST AND TORQUE COEFFICIENTS
The ITTC procedure was applied in the study to
extrapolate the thrust and torque coefficients of the
propeller model to full-scale. The extrapolated values
were compared with the CFD results calculated for the
full-scale case. A comparison between the calculated
CFD results and the data of the Wageninger B-Series
propeller B4.55 is given in Figure 3.
The computed thrust and torque coefficients as well as
the efficiencies of the propeller for the different
Reynolds numbers and advance ratios are contained in
Tables 1-3. All coefficients in the tables are included
without rounding the numbers, in order to avoid small
differences between the coefficients being lost through
rounding.
The Reynolds numbers (Rn) of the propeller model at
advance ratios 0.6, 0.7 and 0.8 were 2.52 x105, 2.93 x105
and 3.33x105 respectively. These Reynolds numbers were
in the range which is usually used in model tests. The
largest investigated propeller diameter is 12 m; the
smallest (model) is 0.25 m. This corresponds to a scale
factor (λ) of 48.
As can be seen in Tables 1-3, the computed thrust and
torque coefficients show a strong dependence on the
Reynolds number. While with increasing Reynolds
number the torque coefficient decreases, the thrust
coefficient of the propeller increases. As expected this
leads to an increase in the efficiency of the propeller.
However, this increase in efficiency depends also on the
thrust loading condition. At a high thrust loading
condition, the increase in the propeller efficiency is much
smaller than at lower thrust loading. (compare Tables 1
and 3)
The differences between the numerically estimated thrust
and torque coefficients as well as the propeller
efficiencies for model and full-scale at different advance
ratios are included in the Tables 4, 6 and 8. The
computed values of the model propeller were used as a
basis. The same data is also used as model values for
applying the ITTC procedure. The estimated results for
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
extrapolated coefficients by applying the ITTC procedure
can be found in the Tables 5, 7 and 9.
The ratios between the estimated thrust and torque
coefficients as well as the propeller efficiencies for
model and full-scale data were also included in Figure 4,
5 and 6.
Figure 5 shows that the increase in the calculated thrust
coefficient using the applied CFD method is much higher
than the results of the ITTC procedure. The scale effect
increases with the decrease of the thrust loading. The
dependency of the scale effect on the thrust loading is
much higher for the torque coefficient than for the thrust,
see Figure 4. At J = 0.8 the calculated reduction of torque
coefficient using the applied CFD method is much higher
than the results of the ITTC procedure. On the contrary at
J = 0.6, the calculated reduction of torque coefficient
using the CFD method is lower than the ITTC procedure.
The computed changes of the efficiencies according to
the results of the CFD and the ITTC procedures are
shown in Figure 6. Due to the high values of the thrust
coefficients estimated with the help of CFD, the
efficiency determined by the CFD results is clearly much
higher than the corresponding values using the ITTC
procedure.
It can be seen that the changes in the thrust coefficient
over the Reynolds number show very different
tendencies. While the CFD results show a clear
dependence of the thrust coefficient on the Reynolds
number, the determined coefficient is nearly constant in
the ITTC procedure. That the tendency of the calculated
CFD results does not correspond to the results of the
ITTC procedures for extrapolation of the thrust
coefficient is understandable, since the ITTC procedure
is based on the acceptance that no, or only a very small,
dependence of the thrust coefficient on the Reynolds
number exists.
4.2 PRESSURE DISTRIBUTION
The dependency of the thrust and torque coefficients as a
function of the Reynolds number is a result of the change
of the local pressure and wall-shear friction on the blade.
The pressure distribution on the suction and on the
pressure side for model scale is shown in Figure 7. To be
able to compare the pressure distribution as well as the
wall-shear friction for different operation conditions of
the propeller, the values presented were made
dimensionless by using the local relative velocity to the
blade at each propeller radius.
The pressure distribution on the suction face shows a
negative pressure over a large area of the propeller blade
with exceptions at the leading and trailing edge. The
pressure reduction on the propeller blade is increased
with the increase of the propeller loading. Compare the
results in Figure 7 for J = 0.8 and 0.6. Positive pressure
© 2005: Royal Institution of Naval Architects
area exists on the pressure side with the exception of the
tip and the root region of the blade. The highest pressure
areas are located near the leading and the trailing edges.
The pressure distribution on the suction and on the
pressure side for the other Reynolds numbers shows
similar tendencies as seen in Figure 7, but the size of the
positive and negative pressure areas changes respective
to the Reynolds number. In order to be able to quantify
these differences, the dimensionless pressure
distributions calculated at model scale were subtracted
from the corresponding pressure values at each Reynolds
number, see Figures 8-10. The largest differences occur
near the edges of the propeller blade. While the pressure
on the root and the tip regions of the suction side is
reduced, the pressure on the leading and the trailing edge
is increased.
When increasing the Reynolds number, the differences of
pressure on the suction face of the blade are clearly
higher, see Figures 8-10. The figures show the reduction
of the pressure on the suction side of the blade, which is
responsible for the increase in the propeller thrust.
The change of pressure distribution on the pressure side
with respect to the Reynolds number can be seen in
Figures 8-10. The increase in pressure takes place in an
area near the trailing edge region. This pressure increase
is high in the root region and it decreases toward the tip.
In contrast to this, a pressure reduction occurs in the
region of the leading edge. With an increase in the
Reynolds number, the area of the negative pressure on
the root of the blade increases. This area is clearly
recognizable in Figures 11-13. However, the area of the
pressure increase on the trailing edge is much larger than
the low pressure area on the leading edge. Therefore, it
must be expected that more thrust will be generated from
the pressure side of the blade.
Detailed figures for the change of the pressure
distribution for the leading and trailing edges as well as
for the tip of the blade are included in order to focus on
the change of the pressure at this regions due to the
increase of the Reynolds number, see Figures 11-14.
4.3 WALL-SHEAR STRESS DISTRIBUTION
The wall-shear stress distribution on the suction and
pressure side for the model scale at the three advance
ratios investigated is shown in Figure 15. The absolute
values of the wall-shear stress were made dimensionless
by the same manner as for the pressure. Figure 15 shows
that the wall-shear stress is very high near the leading
edge and the tip of the blade due to the increase of the
peripheral speed in a radial direction with increasing
propeller radius. Additionally, it can be seen that a local
increase of the wall-shear stress takes place within the
root region near the leading edge .

The wall-shear stress distribution on the suction and
pressure side at the other Reynolds numbers shows
similar tendencies, as described in Figure 15. In order to
be able to regard these differences exactly, the calculated
dimensionless wall-shear stresses of the model scale
were subtracted from the appropriate results for the
Reynolds numbers investigated at the same advance ratio.
In Figures 16-18 it can be observed that the
dimensionless wall-shear stresses decrease with an
increasing Reynolds number. This can be seen more
clearly in the leading edge region.
The reduction of the wall-shear stress within the leading
edge region with an increasing Reynolds number can be
seen clearly in Figures 19-21. The same tendency occurs
in the range of the trailing edge. But the reduction of the
wall-shear stress within the trailing edge region is much
lower than that for the leading edge.
The decrease in the wall-shear stress on the tip region
with an increasing Reynolds number is shown in Figures
22-24. The differences in wall-shear stress between the
Reynolds numbers investigated in the full scale and the
model are considerable.
5. CONCLUSIONS
In order to investigate the scale effects on the thrust and
torque coefficients of the propeller the viscous flow
around a propeller geometry was computed at 6
Reynolds numbers and 3 advance ratios. The thrust and
torque coefficients of the full-scale propeller were
computed using the CFX-5.7 method and the ITTC
procedure for the extrapolation of thrust and torque
coefficients from the model to full-scale.
The changes of the thrust and torque coefficients of the
investigated propeller geometry show that different
tendencies of dependency of the results on the Reynolds
number. While the CFD results supply a clear increase in
the thrust coefficients, the corresponding data in the
ITTC procedure show a very weak dependence.
The comparison of the dimensionless distributions of
pressure and the wall-shear stress at the Reynolds
numbers investigated clarifies the reason for the change
of the thrust and torque coefficients of the propeller in
dependence on the Reynolds number. The dimensionless
pressure on the suction face of the blade drops with
increasing Reynolds number. The dimensionless wall-
shear stress shows in general a high reduction, especially
in the trailing edge and tip regions.
Since only one propeller geometry is investigated in this
study, the validity of the tendencies determined
according to the CFD results is too restricted to establish
generally valid statements.
Marine CFD 2005: Southampton, UK
Moreover a comparison with the results of full-scale
measurement is needed, which however can not be easily
realized due to technical complications and the expense
of full-scale tests. Yet in any case the results of the
computation show that a large need for research in this
field exists, whose completion can be of substantial use
both for shipyards and for propeller manufacturers.
6. ACKNOWLEDGEMENTS
The authors wish to thank Mr. Dipl-Ing. H.-J. Heinke,
SVA Potsdam, for suppling the geometry data. Special
thanks also apply to Dr.–Ing. K. Meyne for the sources of
literature and computer programs for the extrapolation
procedures of the characteristic values of the propeller.
7. REFERENCES
1. Mewis, F., Klug, H., ‘Very large container ships,
difficulties and potential from the hydrodynamic
standpoint’, International Symposium on Naval
Architecture and Ocean Engineering, Shanghai, China,
September 2003.
2. Meyne, K., ‘Investigation of the effect of the boundary
layer flow and friction on the propeller thrust and torque
coefficients’, STG-Jahrbuch, 1972.
3. Abdel-Maksoud, M., Heinke, H.-J., ‘Investigation of
the viscous flow around modern propulsion systems’,
STG-Jahrbuch, 2000.
4. Abdel-Maksoud, M., Rieck, K., ‘Calculation of the
pressure reduction in the tip vortex core of a skew
propeller in model and full scale’, 4th Numerical Towing
Tank Symposium, Hamburg, Germany, September 2001.
5. Abdel-Maksoud, M., Heinke, H.-J., ‘Scale effects on
ducted propellers’, 24th Symposium on Naval
Hydrodynamics, Fukuoka, Japan, July 2002.
6. Lübke, L., Abdel-Maksoud, M., ‘Calculation of the
wake field of full-scale’, STG-Jahrbuch, 2002.
7. Abdel-Maksoud, M., ‘Numerical and experimental
study of cavitation behaviour of a propeller’, Sprechtag
Kavitation, Schiffbautechnische Gesellschaft, Hamburg,
January 2003.
8. Abdel-Maksoud, M., Hellwig, K., Blaurock, J.,
‘Numerical and experimental investigation of the hub
vortex flow of a marine propeller’, 25th Symposium on
Naval Hydrodynamics, St. John’s, Canada, August 2004.
8. AUTHORS’ BIOGRAPHIES
Sven-Brian Müller holds the current position of a
scientific assistant at the Institute of Ship Technology
and Transportation Systems (IST), University Duisburg-
Essen.
Moustafa Abdel-Maksoud holds the current position of
the professor for hydrodynamic and ship design at the
Institute of Ship Technology and Transportation Systems
(IST), University Duisburg-Essen.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

9. TABLES AND FIGURES

Table 1: Operation conditions and characteristics of the investigated propeller geometry at J = 0.8
D V n KT KQ η Rn CTH
Nr.
[m] [m/s] [1/s] [--] [--] [--] [--] [--]
1 12 9.6 1 0.098012930 0.016962055 0.735724161 1.11E+08 0.099117348
2 10 8.0 1 0.097715984 0.017004814 0.731650765 7.73E+07 0.098905299
3 8 6.4 1 0.097339146 0.017062785 0.726352996 4.95E+07 0.098640153
4 6 4.8 1 0.096797714 0.017145521 0.718827255 2.78E+07 0.098259928
5 4 3.2 1 0.095906109 0.017281219 0.706613653 1.24E+07 0.097630282
6 0.25 4 20 0.091757117 0.017916611 0.652069705 3.33E+05 0.094738651

Table 2: Operation conditions and characteristics of the investigated propeller geometry at J = 0.7
D V n KT KQ η Rn CTH
Nr.
[m] [m/s] [1/s] [--] [--] [--] [--] [--]
1 12 8.4 1 0.142187842 0.022763765 0.695883519 1.10E+08 0.186759582
2 10 7 1 0.141849079 0.022797472 0.693199106 7.62E+07 0.186423766
3 8 5.6 1 0.141405163 0.022842576 0.689665287 4.88E+07 0.185984628
4 6 4.2 1 0.140771215 0.022908128 0.684608734 2.74E+07 0.185357859
5 4 2.8 1 0.139754645 0.023016275 0.676471304 1.22E+07 0.184357951
6 0.25 3.5 20 0.135054050 0.023534848 0.639314261 2.93E+05 0.179794549

Table 3: Operation conditions and characteristics of the investigated propeller geometry at J = 0.6
D V n KT KQ η Rn CTH
Nr.
[m] [m/s] [1/s] [--] [--] [--] [--] [--]
1 12 7.2 1 0.184337761 0.028024693 0.628123195 1.08E+08 0.328615931
2 10 6 1 0.183960694 0.028049910 0.626274824 7.53E+07 0.328084231
3 8 4.8 1 0.183462311 0.028083885 0.623822520 4.82E+07 0.327381890
4 6 3.6 1 0.182747186 0.028131927 0.620329742 2.71E+07 0.326373476
5 4 2.4 1 0.181597539 0.028212254 0.614672180 1.20E+07 0.324758428
6 0.25 3 20 0.176332910 0.028609964 0.588555526 2.52E+05 0.317466395

Table 4: Variation of the thrust and torque coefficient based on CFD results, model-Reynolds’ number 3.33E+05
Nr. D δKQ δKT KQ-full / KQ-Model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000955 0.006256 94.67 106.82 112.82907
2 10 -0.000912 0.005959 94.91 106.49 112.20438
3 8 -0.000854 0.005582 95.23 106.08 111.39193
4 6 -0.000771 0.005041 95.70 105.49 110.23779
5 4 -0.000635 0.004149 96.45 104.52 108.36474

Table 5: Variation of the thrust and torque coefficient based on ITTC procedure, model Reynolds’ number 3.33E+05
Nr. D δKQ δKT KQ-full / KQ-model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000739 0.000799 95.87 100.87 105.21
2 10 -0.000676 0.000730 96.23 100.80 104.75
3 8 -0.000594 0.000642 96.68 100.70 104.15
4 6 -0.000482 0.000520 97.31 100.57 103.35
5 4 -0.000309 0.000334 98.28 100.36 102.12

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Table 6: Variation of the thrust and torque coefficient based on CFD results, model-Reynolds’ number 2.93E+05
Nr. D δKQ δKT KQ-full / KQ-model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000771 0.007134 96.72 105.28 108.84843
2 10 -0.000737 0.006795 96.87 105.03 108.42854
3 8 -0.000692 0.006351 97.06 104.70 107.87579
4 6 -0.000627 0.005717 97.34 104.23 107.08485
5 4 -0.000519 0.004701 97.80 103.48 105.81202

Table 7: Variation of the thrust and torque coefficient based on ITTC procedure, model Reynolds’ number 2.93E+05
Nr. D δKQ δKT KQ-full / KQ-model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000753 0.000813 96.80 100.60 103.93
2 10 -0.000690 0.000745 97.07 100.55 103.59
3 8 -0.000608 0.000657 97.42 100.49 103.15
4 6 -0.000496 0.000535 97.89 100.40 102.56
5 4 -0.000323 0.000348 98.63 100.26 101.65

Table 8: Variation of the thrust and torque coefficient based on CFD results, model-Reynolds’ number 2.52E+05
Nr. D δKQ δKT KQ-full / KQ-model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000585 0.008005 97.95 104.54 106.72284
2 10 -0.000560 0.007628 98.04 104.33 106.40879
3 8 -0.000526 0.007129 98.16 104.04 105.99213
4 6 -0.000478 0.006414 98.33 103.64 105.39868
5 4 -0.000398 0.005265 98.61 102.99 104.43742

Table 9: Variation of the thrust and torque coefficient based on ITTC procedure, model Reynolds’ number 2.52E+05
Nr. D δKQ δKT KQ-full / KQ-model KT-full / KT-model η-full / η-model
[m] [--] [--] [%] [%] [%]
1 12 -0.000764 0.000825 97.33 100.47 103.23
2 10 -0.000701 0.000757 97.55 100.43 102.95
3 8 -0.000619 0.000669 97.84 100.38 102.60
4 6 -0.000507 0.000547 98.23 100.31 102.12
5 4 -0.000334 0.000360 98.83 100.20 101.39

Figure 1: Numerical grid at the propeller blades Figure 2: Streamlines on the propeller blades, J = 0.8

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Figure 3: Open water test diagram

Figure 4: Ratio of the full-scale KQ coefficient relative to the model in %

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Figure 5: Ratio of the full-scale KT coefficient relative to the model in %

Figure 6: Ratio of the full-scale efficiency coefficient relative to the model in %

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 3.33E+05; J = 0.8

suction side pressure side

Rn = 2.93E+05; J = 0.7

suction side pressure side

Rn = 2.52E+05; J = 0.6

suction side pressure side

Figure 7: Pressure distribution

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.24E+07

suction side pressure side

Rn = 4.95E+07

suction side pressure side

Rn = 1.11E+08

suction side pressure side

Figure 8: Pressure differences relative to Rn = 3.33E+05; J = 0.8

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.22E+07

suction side pressure side

Rn = 4.88E+07

suction side pressure side

Rn = 1.10E+08

suction side pressure side

Figure 9: Pressure differences relative to Rn = 2.93E+05; J = 0.7

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

suction side pressure side

Rn = 4.82E+07

suction side pressure side

Rn = 1.08E+08

suction side pressure side

Figure 10: Pressure differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.24E+07

leading edge trailing edge

Rn = 4.95E+07

leading edge trailing edge

Rn = 1.11E+08

leading edge trailing edge

Figure 11: Pressure differences relative to Rn = 3.33E+05; J = 0,8

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.22E+07

leading edge trailing edge

Rn = 4.88E+07

leading edge trailing edge

Rn = 1.10E+08

leading edge trailing edge

Figure 12: Pressure differences relative to Rn = 2.93E+05; J = 0.7

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

leading edge trailing edge

Rn = 4.82E+07

leading edge trailing edge

Rn = 1.08E+08

leading edge trailing edge

Figure 13: Pressure differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

tip, suction side tip, pressure side

Rn = 4.82E+07

tip, suction side tip, pressure side

Rn = 1.08E+08

tip, suction side tip, pressure side

Figure 14: Pressure differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 3.33E+05; J = 0.8

suction side pressure side

Rn = 2.93E+05; J = 0.7

suction side pressure side

Rn = 2.52E+05; J = 0.6

suction side pressure side

Figure 15: Wall shear stress

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.24E+07

suction side pressure side

Rn = 4.95E+07

suction side pressure side

Rn = 1.11E+08

suction side pressure side

Figure 16: Wall shear stress differences relative to Rn = 3.33E+05; J = 0.8

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.22E+07

suction side pressure side

Rn = 4.88E+07

suction side pressure side

Rn = 1.10E+08

suction side pressure side

Figure 17: Wall shear stress differences relative to Rn = 2.93E+05; J = 0.7

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

suction side pressure side

Rn = 4.82E+07

suction side pressure side

Rn = 1.08E+08

suction side pressure side

Figure 18: Wall shear stress differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.24E+07

leading edge trailing edge

Rn = 4.95E+07

leading edge trailing edge

Rn = 1.11E+08

leading edge trailing edge

Figure 19: Wall shear stress differences relative to Rn = 3.33E+05; J = 0.8

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.22E+07

leading edge trailing edge

Rn = 4.88E+07

leading edge trailing edge

Rn = 1.10E+08

leading edge trailing edge

Figure 20: Wall shear stress differences relative to Rn = 2.93E+05; J = 0.7

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

leading edge trailing edge

Rn = 4.82E+07

leading edge trailing edge

Rn = 1.08E+08

leading edge trailing edge

Figure 21: Wall shear stress differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.24E+07

tip, suction side tip, pressure side

Rn = 4.95E+07

tip, suction side tip, pressure side

Rn = 1.11E+08

tip, suction side tip, pressure side

Figure 22: Wall shear stress differences relative to Rn = 3.33E+05; J = 0.8

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Rn = 1.22E+07

tip, suction side tip, pressure side

Rn = 4.88E+07

tip, suction side tip, pressure side

Rn = 1.10E+08

tip, suction side tip, pressure side

Figure 23: Wall shear stress differences relative to Rn = 2.93E+05; J = 0.7

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Rn = 1.20E+07

tip, suction side tip, pressure side

Rn = 4.82E+07

tip, suction side tip, pressure side

Rn = 1.08E+08

tip, suction side tip, pressure side

Figure 24: Wall shear stress differences relative to Rn = 2.52E+05; J = 0.6

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

PREDICTIONS OF THE THRUST AND TORQUE PERFORMANCE FOR TWO

PROPELLER BLADES USING COMPUTATIONAL FLUID DYNAMICS
K Randle and P W Bull, Haslar,QinetiQ Ltd, UK

SUMMARY

Numerical computations of the flow around two typical propeller blades have been carried out and compared with the
equivalent measurements obtained for open water thrust and torque performance characteristics of the two propellers.
The objectives of the predictions were to examine the numerical and modelling parameters required for reliable
comparison with the measured data for the complex three-dimensional turbulent flows. The two propeller blades were
the standard DTRC 4119 and a more advanced skewed blade that is one of a systematic series of propellers designed at
QinetiQ Haslar. Computational results obtained from two different methods are compared with measured data to
examine the numerical sensitivity of the evaluation of the thrust and torque. The first method uses a well-established
RANS based technique for propeller flows, MACH0, and the second method uses a commercial RANS CFD code,
CFX5.

NOMENCLATURE solving the fundamental equations of motion. This

approach can potentially remove the difficulties
C local propeller chord associated with the Reynolds number scaling and can
D propeller diameter obtain detailed velocity and pressure distributions with
fm maximum camber of blade section significantly reduced time and cost.
iT blade section rake, axial displacement of blade
relative to blade generator line 2. BACKGROUND
J advance ratio
(J = V/nD) This paper describes preliminary work carried out within
KT thrust coefficient a research programme funded by the MoD at QinetiQ
(KT = T/(ρn2D4)) Haslar for the analysis of different propeller designs. The
KQ torque coefficient objectives of the programme are to develop the capability
(KQ = Q/(ρn2D5)) to predict the full viscous flow around marine propulsion
n propeller revolutions per second systems including cavitation and noise. These include:
P local propeller thickness
Q propeller torque • The steady and unsteady forces and moments on the
R propeller radius propulsion system.
r local propeller radius • The steady and unsteady flow characteristics
tm maximum blade section thickness including details such as tip vortex and blade wake
T propeller thrust flows.
V free stream speed • The interaction between the propulsion system and
ϑs projected skew angle the appendage including shafts, brackets, rudder and
η efficiency hull.
(η = (J/2π) KT/KQ) • Cavitation onset.
ρ fluid density • Extent, thickness and time variations of cavitation.
• Near and far field noise.
1. INTRODUCTION
The objectives of this paper are to compare
Design of a propeller requires knowledge of the fluid computational results obtained using two different CFD
flow around the propeller blades to enhance performance methods with measured data for the steady thrust and
parameters such as thrust and torque and to reduce torque for two representative propellers. The first method
cavitation and noise. Detailed knowledge of the fluid is an in-house RANS CFD code, MACH0 that has been
velocity and pressure distributions enables the design of validated for propeller flows and the second method is a
efficient and effective propulsion systems for a range of commercial RANS CFD code, CFX5 that has been
flow conditions. validated for ship and submarine flows. This comparison
has been carried out as a preliminary exercise prior to a
Although the thrust, torque, cavitation and noise can be more detailed benchmark exercise to establish the more
readily measured using a scale model in a water tunnel or appropriate method for further development.
towing tank there are difficulties in applying results
obtained for the model to the actual ship or submarine
due to the change in Reynolds number. A complementary
approach is to use Computational Fluid Dynamics (CFD)
to predict the fluid velocity and pressure distributions by

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

3. PROPELLER TEST CASES

The following propeller geometries were chosen for the

comparison:

• Standard three bladed propeller without skew

• Five bladed propeller with skew

The geometric characteristics of the two blades are given

in Table 1 with a graphical representation of each
propeller shown in Figure 1 and Figure 2 for the two
propellers respectively.

Standard Propeller
r/R C/D P/D ϑs iT/D tM/C fM/C
0.2 0.320 1.105 0 0 0.2055 0.01429
0.3 0.3625 1.102 0 0 0.1553 0.02318
Figure 2: Skewed propeller
0.4 0.4048 1.098 0 0 0.1180 0.02303
0.5 0.4392 1.093 0 0 0.09016 0.02182
0.6 0.4610 1.088 0 0 0.06960 0.02072 4. GRID GENERATION
0.7 0.4622 1.084 0 0 0.05418 0.02003
0.8 0.4347 1.081 0 0 0.04206 0.01967 The generation of a suitable grid for the application of a
0.9 0.3613 1.079 0 0 0.03321 0.01817 CFD method to propeller geometries can be carried out
0.95 0.2775 1.077 0 0 0.03228 0.01631 using a variety of different techniques. However, to
1.0 0.0 1.075 0 0 0.03160 0.01175
compare the results obtained from two different CFD
Skewed Propeller
r/R C/D P/D ϑs iT/D tM/C fM/C codes it is beneficial to use the same grids in each code.
0.2 0.1963 1.0808 0 0 0.04960 0 The in house code, MACH0, is a single block structured
0.3 0.2459 1.3971 -6.5649 -0.01774 0.04018 0.014545 code that limits the type of grid generation method that
0.4 0.2862 1.5179 -10.0000 -0.02703 0.03233 0.028212 can be used. Such limitations do not apply to CFX5 that
0.5 0.3168 1.5628 -10.3125 -0.02787 0.02606 0.031779 can use a wide variety of different grids and can combine
0.6 0.6337 1.5390 -7.5000 -0.02027 0.02138 0.030330 them together to obtain the flow characteristics around
0.7 0.3447 1.4510 -1.5623 -0.00423 0.01827 0.025525
0.8 0.3348 1.3064 7.5000 0.02027 0.01674 0.018182
complex configurations.
0.9 0.2949 1.1407 19.6875 0.05321 0.01474 0.008921
0.925 0.2759 1.1029 23.2227 0.06276 0.01379 0.006104 A simple ‘skewed H’ grid configuration was used for
0.95 0.2496 1.0672 26.9531 0.07284 0.01248 0.003157 both propeller blades that consisted of a single structured
0.975 0.2089 1.0337 30.8789 0.08345 0.01045 0.000889 block containing hexahedral cells that were produced in
1.0 0.0522 1.0029 35.000 0.09459 0.00027 0.0 cylindrical polar co-ordinates. The grid generation
Table 1: Geometric characteristics of the two propeller process uses blade section data to define a surface grid
blades on the blade pressure and suction sides. Suitable
clustering in the axial direction is used to define the
The measured thrust and torque data used for comparison shape of the leading edge. The leading and trailing edges
with the CFD codes for standard propeller is given in are extended in the upstream and downstream axial
reference 1. The measured thrust and torque data for the directions that follow the local blade pitch angles. This
skewed propeller for comparison with the CFD codes forms a surface grid that aligns to the wake from the
were obtained in the Ship Tank at QinetiQ Haslar. trailing edge of the blade. Figure 3 shows some of the
details of the surface grid produced for the standard
blade.

Figure 3: Axial surface grid for DTRC 4119

Figure 1: Standard propeller

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK
The surface grid is interpolated in the radial direction to
provide suitable clustering to the hub and tip of the blade.
The surface grid is also extended in a constant direction
to a far field boundary. Finally the volume grid is
produced by interpolating in the blade-to-blade direction
with suitable clustering to define the boundary layer on
the blade surfaces. Figure 4 and Figure 5 show
representative radial and blade-to-blade grid surfaces for
the standard blade. Only some of the grid lines are
displayed to improve the clarity of the picture.
Figure 4: Representative grid surface on the radial
direction
Figure 5: Representative grid surface in the blade-to-
blade direction
The skewed H grid generation method was used to obtain
different grid resolutions of 500K, 1M and 2M
hexahedral cells for the standard blade to ensure that
reasonable grid convergence could be achieved.
Subsequently, 1M cells were used to obtain results for
the skewed blade.
5. NUMERICAL METHODS
The two CFD methods used to obtain the flow
parameters for the propeller blades have distinctly
different characteristics.
© 2005: Royal Institution of Naval Architects
The first method, MACH0, is based on a pseudo
compressibility time marching technique with a finite
volume discretisation scheme for the hexahedral cells. A
second order accurate central-difference scheme is used
with additional fourth order artificial dissipation to
provide numerical stability. An implicit pre-processing
algorithm is used to accelerate the convergence of the
solution to steady state. The Baldwin-Lomax algebraic
turbulence model is used to account for the turbulent
flow. A logarithmic function is used to define the wall
boundary condition on the blade surfaces.
The second method, CFX5, is based on a fully coupled
time marching technique with a finite element
discretisation scheme. The hexahedral cells are further
subdivided to form control volumes around each cell
vertex. A number of different difference schemes are
available in CFX5:
• Upwind
• Blended
• High resolution
• Second order central
• Second order upwind
The upwind and blended schemes provide very robust
and stable first order discretisation but the second order
high resolution, central and upwind schemes provide
improved accuracy. The convergence is accelerated
using algebraic multi-grid techniques. A range of
turbulence models is available in CFX5:
• One equation models
• Two equation k-ε and k-ω turbulence models
• Reynolds stress models
Advanced wall functions are used to define wall
boundary conditions for the blade that depend upon the
turbulence model.
6. STANDARD PROPELLER GRID
RESOLUTION STUDY
The numerical simulations were set up with constant
inflow velocity and varying the revolutions of the blade
varied advance coefficient (J), based on inlet velocity.
The effect of the hub on thrust and torque was initially
investigated to compare with the measured data. Changes
to the hub boundary condition did not significantly affect
the results and as such the hub boundary condition was
modelled as free slip in the CFX5 numerical method. The
outer boundary was also modelled as a free slip. The exit
boundary condition was modelled as an average
atmospheric pressure boundary. The blade surfaces were
modelled with non-slip boundary conditions. A single
blade passage was modelled with the blade-to-blade
information transferred using periodic boundary
conditions, upstream and downstream of the blade
surfaces.
 Marine CFD 2005: Southampton, UK

The CFX5 data sets were generated for the standard
blade using a physical time step of 5.0x10-4 for 500k and
2.0x10-4 for 1M and 2M cells, using the high resolution
advection scheme and the k-ε turbulence model.
The simulations were carried out until there was
convergence in the order of 3 orders of magnitude for the
RMS residual mass and momentum equations and the
steady state thrust and torque was invariant. Each of the
simulations was stopped at 300 iterations for direct
comparison. All of the simulations were carried out in
parallel using domain decomposition to partition the
computing load across multiple processors.
This process was repeated for a range of J for each of the
three grid resolutions. The results in Table 2 to Table 4
detail the coefficients obtained for varying grid
resolutions, advance coefficient and numerical method.
The thrust and torque is obtained by integrating the
pressure and wall shear stress on each blade surface for
each computational cell.
Details of the measured data and the MACH0 results are
given in reference 1 and 2.
direction there were high aspect cells present in the tip
region. This caused some solver divergence when
partitioned across processors. This divergence issue was
eliminated by further refinement in the radial direction to
2M cells, and the partitioned grids were robust.
The results indicate that, in the design region with
J=0.833 there is good correlation with the in house code
(MACH0) results, the commercial numerical code
(CFX5) results and with the measured data.
Grid independence was confirmed from 1M to 2M cells
by investigation of the velocity and pressure distributions
over the blade surface and in the wake field. Indicative
plots of pressure on the blade suction surface are given in
Figure 6 to Figure 12. Figures for J=0.5 and J=0.833
show consistent pressure contours with small variation
due to the change in the grid resolution. However, the
figures for J=1.1 show a much larger variation in the
pressure contours. This is likely due to the blade sections
local angle of attack being very close to zero degrees at
this advance ratio.

KT
J 0.5 0.833 1.1
MACH0 0.3120 0.1470 0.0280
CFX 500k 0.3080 0.1531 0.0311
CFX 1M - 0.1511 -
CFX 2M 0.3016 0.1512 0.0311
Measured 0.285 0.146 0.034
Table 2: Comparison of thrust coefficient for the
standard blade

KQ
J 0.5 0.833 1.1
MACH0 0.0531 0.0287 0.0107
CFX 500k 0.0507 0.0295 0.0112
Figure 6: Pressure contours, J=0.5, 500k, standard blade
CFX 1M - 0.0288 -
CFX 2M 0.0481 0.0288 0.0106
Measured 0.0477 0.0280 0.0106
Table 3: Comparison of torque coefficient for the
standard blade

ηo
J 0.5 0.833 1.1
MACH0 0.4673 0.6790 0.4573
CFX 500k 0.4839 0.6877 0.4866
CFX 1M - 0.6965 -
CFX 2M 0.4995 0.6956 0.5145
Measured 0.489 0.692 0.575
Table 4: Comparison of open water efficiency for the
standard blade

It was noted during the process of carrying out the grid

resolution study that obtaining a robust converged
solution relied upon there being fewer high aspect ratio Figure 7: Pressure contours, J=0.5, 2M, standard blade
cells in the tip region. With the method used to provide
grid refinement from 500k to 1M cells in the theta

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Figure 8: Pressure contours, J=0.833, 500k, standard Figure 11: Pressure contours, J=1.1, 500k, standard blade
blade

Figure 12: Pressure contours, J=1.1, 2M, standard blade

Figure 9: Pressure contours, J=0.833, 1M, standard blade
7. SKEWED BLADE ADVANCE RATIO
STUDY

From the previous study, the number of cells required in

the H-grid for grid independence was established at 1M
cells. The same boundary conditions were set up for the
skewed blade as for the standard blade, noting the
reverse direction of rotation.

During the simulation process, in order to obtain a

converged robust and accurate solution the skewed blade
was first run with the upwind differencing advection
scheme option in CFX5. After 500 iterations the upwind
scheme had reached a converged solution. The
simulation was restarted from the values obtained from
the upwind solution using the second order, high-
resolution scheme option. Subsequent simulations were
carried out with approximately 300-400 iterations using
Figure 10: Pressure contours, J=0.833, 2M, standard the upwind scheme and the high-resolution scheme
blade applied for a further 500 iterations. This two-step process
was run for a series of simulations, over a range of
advance ratio.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

Figure 13 and Figure 14 illustrate typical convergence

time histories for residual momentum and mass and
turbulence quantities respectively. Note the jump at 360
iterations where the numerical simulation was restarted
from upwind advection solution values and continued
with high-resolution advection.

Figure 15 and Figure 16 show representative time

histories of the blade thrust and torque given by the
normal pressure distribution on the blade surface.
Because the normal pressure force on the blade is
dominant, the tangential component has been omitted in
these figures for clarity in the display of the results.
However, it was computed and included in the
calculation of the coefficients.

Figure 14: Turbulence convergence iteration history,

J=1.2, skewed blade

Figure 13: Momentum and mass convergence iteration

history, J=1.2, skewed blade

Figure 15: Thrust (normal to blade) convergence iteration

history, J=1.2, skewed blade

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

These results indicate that the two numerical codes have

similar results in the design region of the blade. It was
again noted that the off design accuracy of the results for
each of the numeric methods was not in close agreement
to the measured data.

A more detailed comparison is shown in Figure 17 which

compares the measured data and the numerical
simulation results for the propeller coefficients across a
range of J. This indicates good correlation between
MACH0 and CFX5.

The error bars on either side of the measured result are

indicative of 5% error as a typical value for the
measurement errors. The slight reduction in KT and KQ at
low values of J (below J=0.8), which results in an
elongated ‘S’ characteristic in the measured values, is not
present in either of the numeric simulations.

Figure 18 to Figure 21 show examples of the pressure

distribution over the pressure and suction faces of the
skewed propeller blade for varying J. The pressure face
is shown on the left hand side of the picture and the
suction face on the right. The changes in the pressure
Figure 16: Torque (normal to blade) convergence field structure can be seen for each advance ratio,
iteration history, J=12, skewed blade especially for the suction surface. Each figure has a
different scale for the pressure.
Table 5 to Table 7 show the calculated high resolution
results for the skewed propeller, comparing the measured 8. DISCUSSION AND CONCLUSIONS
data and numerical solutions for MACH0 and CFX5 for
the J values of J=0.8, 1.0, 1.2 and 1.4. The measured It has been demonstrated that the commercial code
results and the MACH0 results were obtained from CFX5, employed in this study, has yielded comparable
reference 3. results to measured data and to the results obtained from
the in house code MACH0 for thrust and torque of two
KT open water propeller cases.
J 0.8 1.0 1.2 1.4
MACH0 0.3753 0.2730 0.1683 0.0616 It was noted during the work that the robustness of the
CFX5 solver and results for steady state cases were not
CFX 0.3714 0.2695 0.1665 0.0561
dissimilar to MACH0 for off-design J. The robustness of
Measured 0.3534 0.2774 0.1812 0.0754
the commercial code was affected by the quality of the
Table 5: Comparison of thrust coefficients for the grid, especially in the tip region.
skewed blade
However there were discrepancies in both numerical
KQ methods with measured data at off-design J. This may be
J 0.8 1.0 1.2 1.4 due to the flow physics, where at low J, the blade section
MACH0 0.0802 0.0635 0.0450 0.0245 angle of attack is higher than the angle of stall for the
CFX 0.0781 0.0630 0.0450 0.0227 blade, possibly leading to tip separation and blade
cavitation, and at high value of J, in the zero thrust region
Measured 0.0781 0.0660 0.0482 0.0263
of the blade, there is a zero average and negative angle of
Table 6: Comparison of torque coefficients for the attack on the many blade sections. The blade section
skewed blade angles of attack for J=0.6 and 1.5 are given in Figure 22
and Figure 23
ηo
J 0.8 1.0 1.2 1.4 These conditions, along with the use of the k-ε model for
MACH0 0.5960 0.6845 0.7143 0.5607 turbulence are possibly not suited to steady state, and
CFX 0.6051 0.6813 0.7072 0.5501 transient calculations may lead to more comparable
Measured 0.5760 0.6685 0.7175 0.6400 improved results. Improvements in the modelling of the
Table 7: Comparison of open water efficiency for the physics by way of a more optimal turbulence model and
skewed blade improvements in the grid geometry are expected to
improve off design results.

© 2005: Royal Institution of Naval Architects

 Marine CFD 2005: Southampton, UK

0.8

0.6
KT, 10KQ, Eta

0.4

0.2

0
Kt Ship Tank Kt CFX 10 Kq CFX
10 Kq Ship Tank Efficiency Ship Tank Efficiency CFX
10 Kq MACHO Efficiency MACHO Kt MACHO

-0.2
0.5 0.7 0.9 1.1 1.3 1.5
Advance ratio (J =Va/nD)

Figure 17: Comparison of non-dimensional propeller curves, skewed blade

Figure 18: Pressure distribution, skewed blade, J= 0.8 Figure 20: Pressure distribution, skewed blade, J= 1.2

Figure 19: Pressure distribution, skewed blade, J= 1.0 Figure 21: Pressure distribution, skewed blade, J= 1.4

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

Science, School of Engineering Sciences, Ship

Blade Section Angle Of Attack at J=0.6
Science, January, 2002.
3. Unpublished QinetiQ Report.
40 Skew ed blade DTRC 4119
11. AUTHORS BIOGRAPHIES
30
Karl Randle is a Naval Architect in the Surface Ship
Angle (º)

20 Capability Group at QinetiQ Ltd, Haslar Marine

Technology Park. He graduated from the University of
10
Southampton with an MEng in Ship Science and joined
QinetiQ in 2002. He has worked on a range of projects
0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 from physical model testing, concept design, ship trials
-10 and numerical simulation. He is a graduate member of
r/R RINA and working towards gaining chartered status.
Figure 22: Blade section angle of attack for J=0.6, Peter Bull holds the current position of Technical Leader
(standard and skewed blades) for Computational Fluid Dynamics (CFD) in the
Blade Section Angle Of Attack at J=1.5 Submarine Capability Group at QinetiQ Ltd, Haslar
Marine Technology Park. He is responsible for the
application, validation and development of CFD
40 Skew ed blade DTRC 4119 techniques in marine hydrodynamics. He is currently a
member of the ITTC Resistance Committee.
30
Angle (º)

20

10

0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-10
r/R

Figure 23: Blade section angle of attack for J=1.5,

(standard and skewed blades)

In this preliminary study the grid definition in the tip

region was not ideal and further improvements have been
made in the grid generation process, refining the grid in
all the axial, radial and theta directions for further work
and the benchmarking exercise.

Further work is ongoing, including a benchmark of the

two numerical codes using equivalent grids and
turbulence model.

9. ACKNOWLEDGEMENT

The authors wish to thank Chris Jenkins (Dstl) and the

UK MoD for their continued support, facilitating the
development of CFD methods for underwater
applications.

10. REFERENCES

1. Jessup SD “Propeller blade flow measurement

using LDV” PhD Thesis, ASME Fluid Engineering
Division Summer Meeting, Take Tahoe, Nevada,
1994.
2. Stanier MJ “Numerical prediction of propeller
scale effect” PhD Thesis, University of
Southampton, Faculty of Engineering and Applied

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

BEHAVIOUR OF SHIP FUNNEL EXHAUST IN THE WAKE OF A BLUFF BODY

P R Kulkarni, S N Singh and V Seshadri, IIT Delhi, India

SUMMARY

The cluttered topside of a typical naval ship features short funnels that are located in the vicinity of taller structures that
are aerodynamically bluff bodies. This highly unfavorable funnel configuration causes the exhaust smoke to exit into the
wake of the bluff bodies, thereby resulting in trapping of the exhaust into recirculation zones and cause undesirable
smoke contamination on the topside operational areas. This paper presents an experimental and a CFD study of the
funnel of a scale model of a typical naval ship, operating in the wake of a bluff body and demonstrates that CFD is a
powerful tool capable of providing a means of visualising the path of the exhaust under different operating conditions
very early in the design spiral of the ship. The aerodynamic study comprised of flow visualisation in the wind tunnel
and CFD simulations of the near field smoke dispersion in the flow field that is disturbed by bluff bodies of the
superstructure/mast located in the vicinity. The flow visualization study of the plume trajectory was undertaken to gain
an understanding of the behavior of the exhaust smoke exiting into the wake of the bluff body and to study the
interaction between a bluff body wake and the ship’s exhaust. The exhaust from the funnel was thereafter simulated
using the computational code FLUENT version-6.0. Closure was achieved by using the standard k-ε turbulence model.
The CFD simulations show a very good agreement with the flow visualisation photographs. CFD analysis has identified
the large velocity gradients, recirculation zones and strong vortex fields in the wake of the bluff bodies of the
superstructure block/mast that cause the smoke downwash. The momentum of the exhaust smoke has a major effect on
the downwash.

NOMENCLATURE aerodynamically bluff bodies, like masts,

superstructure/bridge blocks etc. On naval ships, the need
K Velocity ratio = Ve / Vw to locate weapons on topsides and the masts that house a
range of sensors and the associated equipment make the
Vx , Vy, Vz Components of wind velocity in x, topside extremely cluttered. The competition for topside
y and z directions on superstructure space has seen a reduced funnel size that is required to be
located in the vicinity of the mast/superstructure that are
Vw Ambient cross flow velocity (m/s) aerodynamically bluff bodies and results in violation of
good design practices of a funnel for avoiding smoke
Ve Exhaust velocity (m/s) nuisance (like increasing funnel height and avoiding the
bluff bodies in the vicinity of the funnels), thereby

(Vx ) + (Vy )
2 2
Vxy
(V ) + (V )
2 2
Vyz
y z
(Vx ) + (Vz )
2 2
Vxz
1. INTRODUCTION
An investigation of the behavior of exhaust smoke
emitted from the funnel located in the wake of a bluff
body assumes significance in the design superstructures
on new design ships for identifying the recirculation
zones and undertaking modifications to the topside
layout. The knowledge of the funnel exhaust behavior
and avoiding the problem of smoke nuisance requires
continuous development of the topside design and fine-
tuning the design options, which will enable the ship
designer to achieve the ultimate objective of optimising
the superstructure design and further, save costly changes
later. The superstructures of naval ships are
geometrically much more complex as compared to their
predecessors or the counterparts in the passenger or the
cargo ships as they feature many appendages which are
making them prone to the problem of smoke trouble.
Unfavorable funnel configuration and layout on the
topside produce turbulence and distortion of air motion.
They in turn cause the exhaust smoke to exit into the
wake of the bluff bodies, thereby resulting in trapping of
the exhaust into recirculation zones and cause
undesirable smoke contamination on the topside
operational areas. The downwash of the exhaust causes
the funnel gases to disperse downward toward the deck
more rapidly than upward. This has many adverse
consequences like the sucking of the hot exhaust into the
GT intake and the ships ventilation system apart from
temperature contamination of topside electronic
equipment and interference of the smoke with flight
operations. The evolution of the funnel shape and topside
configuration on passenger and naval ships over the last
hundred years and a comprehensive review of the
problem of smoke nuisance on ships by various
researchers since 1930’s has been presented in the review
paper by Kulkarni et.al. [1]. Traditionally, the funnel
performance has been investigated using scale models in
wind tunnel at a relatively advanced stage of design.
Wind tunnel investigations of the problem have been
reported in the literature since 1940’s [2 to 11]. Bahm
et.al [12] have reported an analytical approach to the

© 2005: Royal Institution of Naval Architects


problem. CFD investigations of the smoke problem on
ships and offshore structures are being reported in the
recent publications for the past five years [13 to 25] and
the trend indicates that such CFD simulations particularly
for cruise vessels, ferries and naval ships will be a part of
the design process in foreseeable future.
Figure: 1: Superstructure configuration
This paper presents a CFD study of the funnel exhaust on
a scale model of a typical naval ship, operating in the
wake of a bluff body and demonstrates that CFD is a
powerful tool capable of visualising the trajectory of the
exhaust under different conditions very early in the
design spiral of the ship. The aerodynamic study
comprised of flow visualisation in the wind tunnel and
CFD simulations of the near field smoke dispersion in a
flow field that is disturbed by the wake of bluff bodies of
the superstructure/mast in the vicinity. The analysis was
carried out for a 1:50 scale model funnel configuration of
typical topside of a generic frigate shown in Fig. 1. This
configuration comprised of two funnels (a forward funnel
and an aft funnel). The superstructure block comprising
the bridge block plus a mast, which is an
aerodynamically a bluff body, is located ahead of the
forward funnel at an upstream distance of one funnel
height. The flow visualization studies in the wind tunnel
were undertaken to gain an understanding of the plume
trajectories from a funnel operating in the wake of the
bluff body and to study the interaction between a bluff
Pitot Tube
WIND
Smoke Generator
Air from
blower
Control Valve Orifice Plate
Figure 2: Experimental setup
Marine CFD 2005: Southampton, UK
body air wake (of the superstructure/mast) and the ship’s
exhaust. The exhaust from the funnel was thereafter
simulated using the computational code FLUENT
version-6.0.
2. APPROACH ADOPTED
The results presented in this paper are a part of an
experimental and numerical study undertaken with an
aim of gaining an understanding of the flow field around
the topside of naval ships and the interaction between a
bluff body air wake (of the funnel and
superstructure/mast) and the ship’s exhaust. The
experimental study comprised of velocity measurements
[10] and flow visualisation photographs [11] of the flow
around a scale model of different variants of a simplified
superstructure of a generic frigate, which yielded a large
body of benchmark data. This data provided the physical
quantities that could directly be correlated to the results
of the numerical simulations by the CFD code FLUENT
version 6.0, wherein the closure was achieved by using
the standard k-ε turbulence model along with grid
refinement and grid adaptation techniques The qualitative
and quantitative comparison [24] shows a reasonably
good agreement of the experimental data with the results
from CFD simulation. Having established the capability
of the CFD code FLUENT version 6.0 to predict the flow
and performance characteristics around the ship’s
superstructures by demonstrating a reasonably good
agreement between the predicted and the experimental
results [24], the same turbulence model and solution
techniques were used for undertaking parametric
investigation of the exhaust smoke-superstructure
interaction on the ship’s topside.
Illuninating Lamp
1:50 scale model
Mirror superstructure
Smoke Generator
Air from
compressor
Control Valve
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
3. MODELLING CRITERIA AND
EXPERIMENTAL SETUP
In the context of dispersion of exhaust plume from ship’s
funnel, the phenomenon of interest is near field
dispersion of the jet in a disturbed flow field created by
bluff bodies of the superstructure, where it is only the
initial plume rise phase that needs to be modelled. In the
near field, the plume path is inertia dominated rather than
buoyancy dominated and hence, the buoyancy forces can
be ignored in the simulation. Simulation of the exhaust
smoke over a model superstructure in the wind tunnel is
done by injection of iso-thermal (unheated) air from the
funnels. The modelling criteria of the an iso-thermal
plume in the wind tunnel as discussed by Kulkarni et.al
[11] shows that scaling the momentum in the near field
study requires only velocity ratio (K) to be matched apart
from the Reynolds number on a geometrically similar
model. In case of bluff bodies such as the superstructure
of ships, the flow patterns around the topside do not
change significantly for Reynolds numbers above 10,000
[7]. This is because the effects of the turbulence flow
phenomenon remain the same over a wide range of
Reynolds number. In the present study, the flow
visualisation in the wind tunnel on a scale model of the
superstructure was carried out at as high a velocity as
possible and K was maintained same in both the model
and the prototype to ensure kinematic similarity.
The schematic arrangement of the experimental set up for
flow visualization studies is shown in Fig. 2. The 1:50
scale model of the superstructure was placed in the wind
tunnel with a test section of 0.75 m x 0.45 m cross
section and having a length of 5 m. A standard Pitot-
static tube inserted into the wind tunnel test section,
upstream of the model measured the free stream velocity.
Air at ambient temperature was injected through the
funnels of the model superstructure to represent the
exhaust of ship’s exhaust smoke. A separate air delivery
system was provided for the forward and the aft funnel,
which included control valves and orifice plate flow
meters. Flow visualization was undertaken by injecting
smoke along with air from two smoke generators
provided in the pipe network upstream of the funnel exit.
The sidewalls of the wind tunnel test section were made
of Perspex sheets, which allowed visualisation of the
flow field over the superstructure model. A 1000 W lamp
Wall
Wall
Inlet
(funnel exits)
Inlet
Figure 3: Computational Domain with boundary conditions
© 2005: Royal Institution of Naval Architects
was clamped over the model outside the wind tunnel,
which illuminated the top edge of the plume and a mirror
was laid horizontally on the deck of the superstructure to
light up the underside of the plume. The video and still
photographs of the smoke exiting from the funnel was
captured using the SONY DCR VX2000E digital video
camera that was mounted on fixed tripod outside the
perspex window on the side of the tunnel. The flow
visualization study in the wind tunnel was carried out for
velocity ratios of 1 and 2 by maintaining the exhaust
velocity at 11.8 m/s and suitably choosing the wind
velocity.
4. CFD SIMULATION
The commercial CFD code "FLUENT" version 6.0,
based on the finite volume technique was used to study
the problem in the present investigation. The
superstructure was modelled in GAMBIT. The
computational domain and the boundary conditions
prescribed are shown in Fig. 3. The uniform wind speed
Vw over the height of the domain inlet was set at 10 m/s
and the exhaust velocity Ve was varied to achieve the
velocity ratio (K) values of 1, 2 and 3 respectively.
Gosman [25] reports that the available RANS (Reynolds
Averaged Navier Stokes) models range from a large
number of variants of the well-known k – ε approach.
The k – ε model is probably the simplest type that is
practically useful. The more complex models tend to
better represent the effects of turbulent anisotropy, which
can be important in some applications, including
turbulent dispersion and buoyancy effects. However, they
usually offer insufficient benefits in return for substantial
extra effort required to solve them. As suggested by
Gosman [25], closure was achieved by using the standard
k-ε turbulence model. The standard k-ε turbulence model
is a two-equation model and is well described in the
literature and is widely used in wind engineering and for
fluid flow analysis through complex geometries. The
computational domain was dicretised using the 3D
tetrahedral mesh. In the first stage, the computational
domain was meshed using a coarse grid to solve for the
flow pattern and plume path.
Outlet
Wall
Wall
(Superstructure and funnel))
 Marine CFD 2005: Southampton, UK

Centerline
Plane

(a) (b)
Figure 4: (a) Definition of centerline plane; (b) Adapted and refined mesh with 9,94,679 tetrahedral cells

(a) K = 1 (b) K = 2
Figure 5: Comparison of downwash on superstructure at velocity ratio of 1 and 2

(a) K = 1 (b) K = 2
Figure 6: Plume trajectory in the wake of a bluff body at velocity ratios of 1 and 2

The following boundary conditions were applied to the
computational domain (Fig. 3)
a) In general, for a ship, the only known flow is that of
the approaching natural wind. The flow around the
superstructure is a secondary process resulting from
the interaction of the external wind with the total
structure of the platform. The flow domain was
bounded at the entry by this specified upstream
boundary. At the entry of the ambient air, the INLET
boundary condition was specified with a velocity of
wind Vw to be 10 m/s.
b) At the exit of the plume, the INLET boundary
condition was imposed for the air with the exhaust
velocity Ve suitably chosen so as to achieve velocity
ratios (K) of 1, 2 and 3.
c) At the exit of the domain, the OUTLET boundary
condition was applied.
d) In the bottom of the domain and on the
superstructure model, no-slip, adiabatic WALL
boundary condition was applied.
e) For the exterior, i.e. the sides of the computational
domain, since it is to be compared with the wind
tunnel experimental data, a no-slip, adiabatic WALL
boundary condition, as applicable to the wind tunnel
was applied.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK
(a) K = 1
(c) K = 3
Figure 7: Pathlines of plume trajectory from CFD simulation at velocity ratios of 1, 2, 3 and 4
The key requirement of the numerical grid is the
adequate resolution of all the separated flow structures so
that the detailed flow characteristics can be calculated in
the region of interest. The regions of interest in the
current study are the wake of the bluff bodies located on
the deck such as the superstructure/mast blocks and the
funnel. Solution-adaptive refinement feature of FLUENT
was used to adapt the grid with an aim of efficiently
reducing the numerical error in the digital solution. The
wakes in the flow represent a total pressure deficit, and
jets are identifiable by a region of relatively high-velocity
fluid. These flow features using the parameters of
pressure and velocity was used to adapt the grid. To
improve the mesh, it was further refined using volume
adaptation with the criterion that the maximum cell
volume change should be less than 50%. One such
adapted grid on the centerline plane is shown in Fig. 4(b).
The solution from the previous (coarse) mesh was
mapped on to this new (adapted) mesh and the
calculations were re-started. The calculations were
carried out using UPWIND (first order) algorithm, with
default settings recommended in FLUENT for such
calculations. Since the thermal gradients were not
considered in the analysis, the flow and heat transfer
were decoupled (i.e. there are no temperature dependent
properties or buoyancy forces) and the equations were
solved for iso-thermal flow by turning ‘off’ the energy
equation to yield a converged flow field solution. The
solution converged to a normalized residual level of 10-6
was carefully post processed for analysis and the results
are presented in the subsequent sections.
© 2005: Royal Institution of Naval Architects
(b) K = 2
(d) K = 4
5. RESULTS AND DISCUSSION
5.1 OBSERVATIONS FROM FLOW
VISUALISATION STUDIES
The photographs from flow visualisation study are
presented in Fig. 5 and 6. At K = 1, it is observed that the
smoke gets trapped in the strong recirculation zone
immediately downstream of the superstructure/mast
block and comes down on the deck as shown in Fig. 5(a),
thus exhibiting a severe downwash. Close-up photograph
of the trajectory of the plume in the wake of the bluff
body at K = 1 (Fig. 6(a)) shows that that the plume path
bends backwards from the vertical.
The trajectory of the exhaust smoke as K is increased to
2 is shown in Fig. 5(b). The smoke does exhibit a
tendency to get sucked into the wake of the mast as
captured in the photograph shown in Fig. 6 (b). However,
the momentum is sufficient to enable it to overcome the
wake region and avoid the downwash as seen in Fig.
5(b).
5.2 COMPARISON BETWEEN FLOW
VISUALISATION PHOTOGRAPHS AND
CFD STUDIES
The visualisations of the plume trajectory in the form of
pathlines from CFD simulation are shown in Fig. 7. The
qualitative comparison between the flow visualisation by
CFD simulation and the flow visualisation photographs
from wind tunnel studies at K = 1 (Fig. 6(a) and 7(a)) and
at K = 2 (Fig. 6(b) and 7(b)) shows a very good
prediction by CFD. The bending of the plume as well as
its trajectory as visualised from the pathlines from CFD

simulation at K = 1 and 2 (Fig. 7(a) & (b)) compares very
well with the flow visualisation photographs. These
agreements confirm that the CFD code using the standard
k-ε turbulence model can predict the flow and
performance characteristics reasonably well.
5.3 ANALYSIS OF THE FLOW FIELD BY CFD
The results of the CFD analysis are presented using the
vector plots of Vyz, Vxz, and Vxy to study the flow structure
and pathlines of the trajectory of the exhaust from the
funnels for flow visualisation. The pathlines were
released from the funnel exit. The flow around the
superstructure was analysed in the horizontal plane - ‘Q’
at a height of 0.9 h (‘h’ is the height of the funnel)
defined in Fig. 8, transverse planes (numbered 1 to 4)
defined in Fig. 11 and the centerline plane defined in Fig.
4(a)
Q
Figure 8: Definition of plane ‘Q’
Figure 9: Vector plot of Vxy on plane ‘Q’ at K = 1
Marine CFD 2005: Southampton, UK
The air stream around the forward funnel is disturbed due
to the presence of the superstructure and the mast
upstream of the funnel. Therefore, the forward funnel
emits the smoke into the wake of the superstructure and
the mast. The flow visualization photographs (Fig. 5(a)
and 6(a)) show that that the plume bends backwards from
the vertical. The superstructure and the tall mast shield
the smoke from the free stream airflow and therefore,
there is significant aerodynamic blockage caused by the
mast to form a low-pressure zone in its wake. Smoke
from the forward funnel is emitted right into this wake of
the mast in the region of pressure deficit. The low-
pressure zone in the wake therefore causes the smoke to
bend backwards from the vertical towards the mast as
seen in Fig. 6(a) and the smoke from the forward funnel
experiences severe downwash at K = 1 as seen in Fig.
5(a).
0.9h
(a)
(b)
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
Fig. 10: Vector plot of Vxz on centerline plane at K = 1
Vw = 10 m/s
1 2
Fig. 11: Definition of planes 1, 2, 3 and 4
The vector plots of Vxy for K = 1 in plane ‘Q’ (Fig. 9(a)
and (b)), indicate the presence of two counter rotating
vortices behind the superstructure/mast block. The
vortices are found behind the forward funnel as well. An
enlarged view of these vortices of Vxy in plane ‘Q’ in Fig.
9(b) clearly shows the interaction of the mast wake with
the forward funnel. The wake-affected region due to the
presence of the bluff bodies (the superstructure/mast
block) as well as the funnels is evident in the vector plot
of Vxz in the centerline plane of the superstructure shown
in Fig. 10. The vector plot of Vxz in the centerline plane
(Fig. 10) also shows that the wake of the mast shields the
exhaust jet of the forward funnel from the free stream
and hence, the jet rises freely up to the height of the mast.
This is confirmed from the plot of pathlines (Fig. 7(a)
and (b)) as well as the flow visualisation photographs
(Fig. 6(a) and 6(b)).
Analysis of the vector plot of Vyz (Fig. 12) explains the
process of shielding and the flow structure immediately
downstream of the superstructure/mast block. At plane-1
located between the forward funnel and the
superstructure/mast block (Fig. 11), two pairs of
symmetrical contra-rotating vortices behind the
superstructure/mast block are seen (Fig. 12(a)). The first
pair of the vortices is attributed to the mast and is present
up to the height of the mast. The second pair of vortices
is due to the presence of the superstructure block. The
region of strong intensity of these two pair of vortices
coincides with the shape of the superstructure/mast
© 2005: Royal Institution of Naval Architects
3 4
block, indicating that the bluff body shape affects the size
and the location of the vortices. Downstream of the
forward funnel, as seen in planes 2, 3 and 4 (Fig. 12 (b)
to (d)) only one pair of the contra-rotating vortices is
observed. The vortices are very strong in the region
immediately behind the bluff body (planes 1 and 2) as
shown in Fig. 12 (a) and (b) respectively. However, as
indicated by the vector plot of Vyz in plane 3 and 4 (Fig.
12(c) and (d)), at increasing downstream distance, the
vortices appear to dissipate and become weak. The effect
of the wake of the superstructure/mast located upstream
of the forward funnel is observed to influence the flow
structure downstream of the aft funnel as well, as seen in
the vector plots in planes 3 and 4.
The vector plots of Vxy at plane ‘Q’ (Fig.9) Vxz at
centerline plane (Fig. 10), as well as the vector plots of
Vyz at planes 1, 2, 3 and 4 (Fig. 12) thus confirm that the
presence of the mast upstream of the funnel causes
blockage in the air stream and shedding of eddies and
vortices in the wake region. These vortices cause a
significant recirculation zone in the wake of the
superstructure/mast block as well as the funnel. As the
smoke is emitted right into this recirculation zone in the
wake of the mast, the eddies and the vortices shed by the
mast bring the smoke down and cause the exhaust from
the forward funnel to be sucked into this region as
observed during the flow visualisation studies (Fig. 5(a)
and 6(a)). These vortices behind the superstructure/mast
block also explain the phenomenon of the plume bending

backwards from the vertical (Fig. 6(a)). Further, at K = 1,
the momentum of the smoke is not sufficient to
overcome its trapping in the wake, and as a result, as
seen in Fig. 5(a), there is a severe downwash and the
smoke comes down on the deck.
The trajectory of the exhaust smoke as K is increased to
2 is shown in Fig. 5(b). The increased momentum at K =
2 along with the fact that the tall mast provides an
(a) Plane-1
(c) Plane-3
Figure 12: Vector plot of Vyz at K = 1 on planes 1. 2. 3 and 4
Marine CFD 2005: Southampton, UK
aerodynamic blockage and shields the smoke from the
free stream airflow, the plume continues to have a
vertical path after exit. The smoke does exhibit a
tendency to get sucked into the wake of the mast as
captured in the photograph shown in Fig. 6 (b). However,
the momentum is sufficient to enable it to overcome the
wake region.
(b) Plane-2
(d) Plane-4
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
Fig. 7 presents a CFD study of the effect of the
momentum of the plume on the behavior of the exhaust
emitted in the wake of a bluff body at velocity ratios of 1,
2, 3 and 4. At K = 1, Fig. 6 (a) & 7(a) show that if the
momentum of the exhaust is insufficient, the exhaust
cannot overcome its trapping in the wake and thereafter,
results in a severe downwash (Fig. 5(a)). At K = 2, the
increased momentum of the plume is able to overcome
the eddies and vortices in the wake of the
superstructure/mast block. The plume escapes the wake
region without significant downwash, though there is a
tendency for the smoke to be sucked in, as indicated by
the photograph from flow visualisation (Fig. 6(b)) and
the pathlines in Fig. 7(b). Further increase in momentum
at K= 3 and 4 (Fig. 7(c) & (d)) shows that there is
absolutely no trapping of the smoke or downwash. This
confirms that the increasing momentum of the exhaust is
necessary to overcome the strong recirculation zone in
the wake of the bluff bodies on the topside of the ship,
particularly those in the vicinity of the funnel and further,
the increased upward momentum of the funnel exhaust
gases enables it to overcome the effect of low pressure
region produced by the wind blowing around the mast in
the wake region, thus ensuring that the downwash does
not occur.
6. CONCLUSIONS
a) The qualitative comparison between the flow
visualisation photographs in the wind tunnel and the
flow visualisation from path lines obtained from
CFD at velocity ratios of 1 and 2 shows a very good
agreement. Further, these agreements confirm that
the CFD code using the standard k-ε turbulence
model can predict the flow and performance
characteristics reasonably well.
b) The flow behind the bluff bodies on topside of a ship
like the superstructure block/mast is charecterised by
large velocity gradients, recirculation zones and
strong vortex fields. The analysis of the flow
structure downstream of the bluff body indicates that
it generates a vortex trail that depends on the shape
of the bluff body and the extent of its streamlining.
The strength of these trailing vortices is a major
contributor of the downwash.
c) The location as well as the size of the bluff body
upstream of the funnel results in the funnel ejecting
the exhaust gases into a strong recirculation zone,
which causes the smoke to come down.
d) A velocity ratio of at least 2 should be maintained to
avoid the problem of downwash. At velocity ratio
greater than 2, the increased momentum ensures that
the smoke stays well clear of the deck.
e) In case of the exhaust smoke– superstructure
interaction on the naval ships, wherein short funnels
are located in the vicinity of taller structures that are
aerodynamically bluff bodies, it is the momentum of
the exhaust that decides the behavior of the smoke
nuisance problem.
© 2005: Royal Institution of Naval Architects
f) This study further goes on to show that CFD is a
powerful tool capable of predicting the larger scale
features of the exhaust smoke-superstructure
interaction, which is particularly advantageous for
the investigation of modern naval ships with
complex topside layouts and allows the detection of
shortfalls in design and to find efficient means to
eliminate them. This can be done very early in the
design spiral, thus avoiding costly modifications at a
later stage.
7. REFERENCES
1. KULKARNI PR, SINGH SN, SESHADRI V, “Study
of Smoke Nuisance Problem on Ships – A Review”
communicated to IJME, RINA, London.
2. NOLAN ROBERT W, “Design of Stacks to Minimise
Smoke Nuisance”, Trans SNAME Vol. 54, 1946, pp
42-82
3. BURGE CH, OWER E, “Funnel Design and Smoke
Abatement”, Trans. Institute of Marine Engineers
(London), Vol 62, 1950, pp 119
4. ACKER H.G., “Stack Design to Avoid Smoke
Nuisance,” Trans SNAME, Vol. 60, 1952, pp 566-594.
5. OWER E, THIRD A.D, “Superstructure Design in
Relation to the Descent of Funnel Smoke”, Trans.
Institute of Marine Engineers (London), Vol 1, 1959,
pp 109-138
6. THIRD AD, OWER E, “Funnel Design and Smoke
Plume,” Trans. Institute of Marine Engineers (London).
Vol. 72, 1962.pp 245-272
7. MICHAEL K JOHNS, VAL HEALEY J, “The
Airwake of a DD 963 Class Destroyer”, Naval
Engineers Journal, May 1989, pp 36-42.
8. SESHADRI V, SINGH SN, “Wind Tunnel Studies to
Obviate the Problem of Unwarranted Rise in Air
Intake Temperatures of Gas Turbines in INS Delhi of
Project 15”, Applied Mechanics Department, IIT Delhi,
2000
9. SESHADRI V, SINGH SN, KULKARNI PR, “A
Study of the Problem of Ingress of Exhaust Smoke into
the GT Intakes in Naval Ships”, Proc. of Int. Engineer
Officers Conference, INS Shivaji, Lonavala, India, Jan
2005, pp168-175
10. KULKARNI PR, SINGH SN, SESHADRI V, “Flow
Visualisation Studies of Exhaust Smoke-
Superstructure Interaction on Naval Ships”, accepted
by ASNE for publication in Naval Engineers Journal
edition of 2005
11. KULKARNI PR, SINGH SN, SESHADRI V,
“Experimental Study of the Flow Field Over
Simplified Superstructure of a Ship”, communicated to
IJME, RINA, London.
12. BAHAM GARY J, D MC CULLUM, “Stack Design
Technology for Naval and Merchant Ships”, trans
SNAME, Vol 85, 1977, pp 324-349
13. HOLDO AE, “Modelling Helicopter Landing
Conditions onboard Offshore Structures”, Application
of Fluid Dynamics in the Safe Design of Topsides and
 Marine CFD 2005: Southampton, UK

Superstructures (1997), Institute of Marine Engineers, College of Engineering, with a PG Diploma in Naval
London, pp 71-77 Construction from IIT Delhi and M.Sc. in Naval
14. TAI TC, “Simulation of DD-963 Ship Airwake by Architecture from University College London. He has
Navier-Stokes Method”, J of Aircraft 32/6, 1995, pp also served onboard various Indian Naval Ships and the
1399-1401. Naval Dockyards. His fields of interest include Warship
15. TAI TC, “Effect of Ship Motion on DD-963 Ship Design, Ship Hydrodynamics, Ship Dynamics, Ship
Airwake Simulated by Multizone Navier-Stokes Aerodynamics and Computational Fluid Dynamics. He is
Solution”, 21st Symposium on Naval Hydrodynamics presently perusing his doctoral research titled “An
(2001), pp 1007-1017. Aerodynamic Study of Exhaust Smoke-Superstructure
16. RADOSAVLJEVIC D, GEBRA JM, “CFD: Design Interaction on Naval Ships” for the award of a PhD.
Assessment Case Studies in Offshore and Marine
Industries”, Application of Fluid Dynamics in the Safe Prof. SN Singh is Professor in the Department of
Design of Topsides and Superstructures, Institute of Applied Mechanics, Indian Institute of Technology,
Marine Engineers, London, 1997, pp 37-48 Delhi with over 21 Years experience in teaching and
17. EL MOCTAR, SCOTT GATCHELL, VOLKER R&D. He is an Aeronautical Engineer with an M.Tech. in
BERTRAM, “RANSE Simulations for Aerodynamic Aeronautical Engineering from IIT-Kanpur and a Ph.D.
Flows around Ship Superstructures”, 4th Num. Towing in Fluid Engineering from IIT-Delhi. His areas of interest
Tank Symposium, Hamburg, 2001 are Fluid Mechanics, Internal Flows, Computational
18. EL MOCTAR, V BERTRAM, “Computation of Fluid Dynamics, Two Phase Flows, Flow
Viscous Flow around Fast Ship Superstructures”, 24th Instrumentation and Ship Aerodynamics.
Symposium on Naval Hydrodynamics, 2003, pp 940-
949. Prof.V.Seshadri is Professor in the Department of
Applied Mechanics, Indian Institute of Technology
19. TAI TC, “Simulation of LPD Airwake by Navier-
Delhi. He is a Mechanical Engineer with ScM. and Ph.D.
Stokes Method ”, 6th Asian Congress of Fluid
degrees from Brown University, U.S.A. With over 31
Mechanics, Singapore, 1995.
Years of experience in teaching and R&D, he has held
20. TAYLOR K, SMITH AG, “CFD Prediction of Exhaust various administrative positions including that of Head,
Plumes and Interaction with Superstructures”, Applied Mechanics Department and Deputy Director
Application of Fluid Dynamics in the Safe Design of (Admn) at the Institute. His areas of specialisation are
Topsides and Superstructures, Institute of Marine Pipeline Engineering, Development and Calibration of
Engineers, London, 1997,pp 56-61 Fluid Devices, Computational Fluid Dynamics, Coal Ash
21. RAMAMURTI R, W.C. SANDBERG “Unstructured Handling & Transportation, Biofluid Mechanics and Ship
Grids for Ship Unsteady Airwake on the LPD-17: A Aerodynamics.
Successful Validation”, Naval Engineers Journal,
ASNE, 114, No. 4, Fall, 2002, pp 41- 53.
22. REDDY KR, TOFFOLETTO R, JONES KRW,
“Numerical Simulation of Ship Airwake”, J of
Computers & Fluids, 29 (2000), pp 451-465.
23. JIN E, YOON J, KIM Y, “A CFD based parametric
Study on the Smoke Behavior of a typical Merchant
Ship”, PRADS’01, Shanghai, 2001, pp 459-465.
24. KULKARNI PR, SINGH SN, SESHADRI V,
“Comparison of CFD Simulation of Exhaust Smoke-
Superstructure Interaction on a Ship with Experimental
Data”, NPTS 05, (Naval Platform Technology
Seminar), Singapore, 2005
25. GOSMAN AD, “Developments in CFD for Industrial
and Environmental Applications in Wind Engineering”,
J of Wind Eng. & Industrial. Aerodynamics, 81 (1999),
pp 21-39.
26. FLUENT INDIA, “Fluent V-6.0 Users Guide”, Fluent.
Inc., 2004

9. AUTHORS’ BIOGRAPHIES

Cdr. PR Kulkarni belongs to the Corps of Naval

Constructors of the Indian Navy and is presently Faculty
at the Naval Construction Wing, Department of Applied
Mechanics at Indian Institute of Technology, Delhi. He is
a graduate of Mechanical Engineering from the Naval

© 2005: Royal Institution of Naval Architects

Marine CFD 2005: Southampton, UK

THE EFFECT OF SHIP SHAPE AND ANEMOMETER LOCATION ON WIND SPEED

MEASUREMENTS OBTAINED FROM SHIPS
B I Moat, M J Yelland and R W Pascal, Southampton Oceanography Centre, UK
A F Molland, University of Southampton, UK

SUMMARY

Wind speed measurements obtained from ship-mounted anemometers are biased by the distortion of the airflow around
the ship's hull and superstructure. These wind speed measurements are used both in numerical weather prediction and in
climate studies and need to be known as accurately as possible. This paper presents results from CFD models used to
quantify and correct airflow distortion effects.

Three-dimensional CFD studies of the mean airflow over various research ships and a generic tanker/bulk carrier have
been performed. The bias in the wind speed measurements is highly dependent upon anemometer position and ship
shape. Even for anemometers in well-exposed locations on research ships the wind speed may be biased by about 10 %.
Anemometers located above the bridge of tankers/bulk carriers may not be as well exposed and could be accelerated by
over 10 % or decelerated by 100 %.

CFD results are compared to in situ wind speed measurements made from a number of anemometers above the bridge of
the research ship RRS Charles Darwin. The CFD-predicted wind speeds agreed with those measured to within 4 %.

1. INTRODUCTION Large Eddy Simulation code GERRIS [10] to study the

Several thousand merchant ships are recruited to the
World Meteorological Organisation (WMO) Voluntary
Observing Ship (VOS) programme to report the
meteorological conditions at the ocean surface. These
reports include wind speed and direction, air and sea
surface temperature, cloud cover and sea state. Wind
speed measurements obtained from anemometers on
these ships are biased by the distortion of the airflow by
the ships hull and superstructure. Quantifying this bias is
important for accurate wind speed measurements needed
for ocean/atmosphere model forcing, satellite validation
and for climate change studies. Previous studies have
been carried out to investigate flow over ship
superstructures in respect of smoke dispersion [1, 2] or
over the aft deck of warships for landing helicopters [3,
4]. The current work focuses on studying the general
flow pattern over ship’s superstructures with particular
attention to the correction of wind speed measurements
made from fixed anemometers.
unsteady flow around the R/V Tangaroa. In all cases the
ship geometries were very detailed.
This paper will describe the CFD code VECTIS (Section
2). In situ measurements used to validate the CFD
simulations will be described in Section 3. Results from
previous flow simulations over the RRS Charles Darwin
(Figure 1) and RRS Discovery (Figure 2) will be used to
highlight the changes in wind speed created by the
presence of research ships (Section 4.1). In addition
recommendations will be made on locating anemometers
to minimise the effects of flow distortion in wind speed
measurements.

Computational fluid dynamics (CFD) has been employed

to correct the wind speed measurements obtained from
research ships [5 to 10]. Kahma and Leppäranta [5]
applied potential flow theory to model the flow over a 2-
dimensional ship model. Potential flow models simulate
the flow of an ideal fluid and do not reproduce many
features of a real flow, e.g. flow separation. Nevertheless,
their study gave the first insight into the magnitude of the
flow distortion at anemometer sites on ships. With the
increase in computing power more realistic flow models
have recently been used. Yelland et al. [6, 7] used the 3-
dimensional CFD code VECTIS to predict the airflow
distortion at anemometer sites on a number of research
ships. Dupuis [8] used a 3-dimensinal CFD model and
predicted wind speed increases of about 20 % at the main
mast site on the RV L’Atalant. Popinet et al. [9] used the
foremast platform
Figure 1: The airflow directly over the bow of the RRS
Charles Darwin. The shade of the velocity vectors
represents the speed of the flow.
Section 4.2 will describe the work of Moat et al. [11, 12]
in studying the airflow over a typical tanker/bulk carrier
(Figure 3). The problems associated with simulating the
airflows over a container ship will be discussed in
Section 4.3. The results of these studies will be used to
make recommendations for locating anemometers on
ships (Section 5).

© 2005: Royal Institution of Naval Architects


foremast platform
Figure 2: As Figure 1, but for a flow over the RRS
Discovery.
2. COMPUTATIONAL METHOD
The CFD simulations were performed using the VECTIS
software package [13]. VECTIS is a commercial three-
dimensional Reynolds Averaged Navier-Stokes solver
originally designed to study the fluid flow within engines.
Nevertheless, the code has successfully been used since
1993 to model the airflow over many research ships [6,
7]. The benefit of using VECTIS over other commercial
codes is the speed at which the mesh can be created. For
complicated geometries typical meshes of 500,000 cells
can be created in less than an hour.
The finite volume code VECTIS is second order accurate.
The VECTIS studies are only intended to reproduce the
steady state mean flow characteristics, not accurate
simulations of the turbulence structure. Therefore the
standard k ~ ε [14] and RNG k ~ ε [15] turbulence
closure models were used to approximate the turbulence.
Eason [16] showed that the RNG model was generally as
accurate as higher order turbulence models in studying
the mean airflow over bluff body cubes.
The detailed ship geometries are created from digitised
2-dimsional ship plans. The digitised plans are then
converted into a 3-dimensional geometry using the pre-
processing software FEMGEN [17]. The creation of the
geometry can take up to 2 weeks. A computational
domain is defined around the geometry with the ship in
the centre. The size of the domain is dependent upon the
ship size and its orientation to the flow. For flows
directly over the bow (head to wind) typical domain sizes
are 600 m in length, 300 m wide and 150 m high for a
ship of 90 m in length. The width of the domain can
increase to over 1000 m for flows over the ship’s beam.
In general the ratio of the frontal area of the ship to the
area of the inlet provides a blockage by the ship of less
than 1 %.
VECTIS is based on a regular Cartesian mesh within
which the number of cells can be increased in regions of
interest, such as anemometer locations, and around sharp
edges. The exact shapes of the geometries are preserved
in the mesh generation process. ‘Law of the wall’
functions were used to describe the thin boundary layers
close to surfaces. The computational cells close to the
Marine CFD 2005: Southampton, UK
solid surfaces were sub-divided to increase the mesh
resolution. The problems associated with regular
Cartesian grids and properly resolving the thin boundary
layers close to complex geometries was not an issue for
the research ship studies, as the anemometer locations are
at a great enough distance from the solid walls ( ≈ 2 m) to
not be affected by the thin boundary layer formation. For
the simulations of flow over the simplified tanker (Figure
3) anemometers may be located close to the bridge top.
Therefore the boundary layers were accurately resolved
to model the complex flow above the bridge. The y+
value varied between 35 and 300, where y + is the
characteristic wall co-ordinate for the boundary layer.
All VECTIS simulations presented were 3-dimensional
and steady state. No attempt was made to accurately
model the flow within the unsteady wake regions. The
number of computational cells used in the simulations
varied from 200,000 to 600,000. Early simulations were
run on an SGI Indigo UNIX workstation and took up to 4
weeks to converge. Current simulations are run on the
HPC facility at the Southampton Oceanography Centre.
This provides a platform on which flow simulations
using three times the number of cells used in the early
computations can be run in less than 2 weeks.
stern
bow
Figure 3: As Figure 1, but for a flow over the simplified
tanker geometry.
The inlet wind speed profiles for the research ship studies
were defined as atmospheric boundary layers typical of
open ocean conditions. The wind speed profile, U ZN ,
varied logarithmically with height, z, and was defined
using:
u*  z 
U zN = ln  (1)
kv  z 0 
where u* is the friction velocity, kv is the von Kármán
constant (0.4) and z 0 is the roughness length. The
subscripts 10 and N refer to a height above the sea
surface of 10 m, and equivalent neutral stability
conditions. The wind speed profile can be defined from
Eq. 1 by calculating values of u* and z 0 . The friction
velocity, u* , was calculated using:
u*2 = C D10N U10N
2
(2)
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

where C D10N is the drag coefficient which varies with accelerated flow region and predicts a maximum increase
wind speed and is defined by an empirical bulk formula of 35 %, which was reasonably close to the maximum
[18]: observed in the wind tunnel. The flow in the decelerated
region counter to the mean flow direction at heights of
1000C D10N = 0.61+ 0.063U10N (3) z/H<0.2 is predicted well.
1
accelerated
The roughness length, z 0 , was calculated by combining wind tunnel
k~eps

hieght, z/H
Eq. 1 and 2 and using a measurement height of 10 m and RNG k~eps
specifying the required wind speed at 10 m. Boundary 0.5
layer profiles and uniform wind speed profiles at typical
wind speeds of 7 ms-1 were used in the simulations. Even decelerated
though the CFD solutions were modelled at sufficiently
low wind speeds so that density changes are minimal, a 0
-0.5 0 0.5 1 1.5
compressible solution was always specified since it
produces a more stable solution [19]. normalised wind speed

Figure 4: A comparison of VECTIS results with the wind

VECTIS simulations of the flow over a typical merchant
tunnel measurements of [21].
ship (Figure 3) were performed using various mesh
densities, turbulence closure schemes, geometry size and
The second test case was the comparison with the
inlet wind speed profiles. The results for the changes in
boundary layer flow over a surface mounted cube [22].
the flow field above the ship’s bridge are presented in
Measurements of the velocity above the cube are
[20] and will be summarised here. The mesh size stated
compared to the VECTIS result in Figure 5. The
was scaled by the bridge top to deck height, H. The
Reynolds number, based on the cube height, was
findings showed that there were possible changes in wind
Re =4 × 104. Unfortunately the measurements were not
speed of < 1 % using minimum cell sizes between
0.018H and 0.04H; < 2 % between the RNG k ~ ε and very extensive with only four measurements between the
standard k ~ ε turbulence closure schemes; and < 3 % in cube top and height of z/H=0.12. The RNG k ~ ε
scaling the geometry. The shape of the wind speed turbulence closure scheme reproduces the flow pattern in
profile has the largest influence (4 %) on the wind speed the decelerated region well.
above the bridge. 0.4

wind tunnel
3. VALIDATION OF CFD 0.3 k~eps
height, z/H

RNG k~eps
3.1 COMPARISONS WITH PREVIOUS WIND 0.2
TUNNEL DATA
0.1 decelerated accelerated
Two test cases were used to validate the VECTIS flow
simulations. Both are wind tunnel studies of the flow 0
-0.5 0 0.5 1 1.5
over surface mounted cubes and were obtained from the normalised wind speed
European Research Community on Flow, Turbulence and
Combustion (ERCOFTAC) database. The first case is a Figure 5: A comparison of VECTIS with the wind tunnel
fully developed channel flow [21] and the second is a measurements of [22].
boundary layer flow [22]. Both sets of measurements
were made using a two component Laser Doppler 3.2 COMPARISONS WITH IN SITU WIND
Anemometer (LDA). Comparisons of VECTIS SPEED DATA
simulations using the standard k ~ ε and RNG k ~ ε
turbulence closure models are made with the wind tunnel Wind speed measurements were obtained using
measurements. In all cases the wind speed profiles were anemometers above the bridge of the RRS Charles
normalised by the inlet wind speed. A negative Darwin (Figure 1) during the SCIPIO cruise [23] in the
normalised velocity indicates a flow counter to the mean Indian Ocean. Although not a true representation of the
flow direction. All heights were normalised by the height flow over a typical VOS, the ship’s structure makes it
of the surface mounted cube, H, used in the study. The ideal for studying bluff body flows when the wind is
VECTIS simulations are based on a minimum mesh blowing on to either beam. This is a summary of the
density of 0.02H above the cube. work described in [24].

The channel flow of Martinuzzi and Tropea [21] was
reproduced using VECTIS and are compared to the
VECTIS results in Figure 4. The Reynolds number,
based on the channel height, was Re =105. The RNG
k ~ ε closure model closely simulates the shape of the
Wind speed data were obtained for 58 days between May
and July 2002. The ship was equipped with 7
anemometers. A HS sonic was located on the foremast
platform. A temporary 6 m mast equipped with an R2
Sonic anemometer, 4 Vector cup anemometers and a

© 2005: Royal Institution of Naval Architects


Windmaster sonic anemometer was located above the
bridge top. The instrument accuracy was: the HS sonic
anemometer (< ±1 % for winds below 45 ms-1); the R2
sonic anemometer (<1 % rms); the Windmaster sonic
anemometer (1.5 % for winds below 20 ms-1) and the
Vector cup anemometers (1 %, ± 0.05 ms-1). The HS, R2
and Windmaster sonics output 3-component wind speed
measurements at 20 Hz, 21 Hz and 0.1 Hz respectively.
The Vector cup anemometers were sampled at 0.1 Hz.
Pre- and post-cruise calibrations of the HS sonic, R2
sonic and Windmaster sonic were performed to examine
any change in the accuracy of the instrumentation during
the experiment. The post-cruise HS and Windmaster
calibrations showed there was no change in their
calibration during the cruise. The post-cruise R2 sonic
calibration suggested a 2 % overestimate of the wind
speed for relative wind directions over either beam. The
correction was applied to the wind speed data measured
by this instrument.
An estimate of the free stream, or undistorted, wind
speed was required in order to quantify the biases in the
measured wind speed for flows directly over either beam.
The HS anemometer was used to normalise the wind
speed measurements above the bridge because; it was the
best-exposed instrument and it was located on the
foremast, well away from the bridge top, i.e. the area
under investigation. To correct for the effects of airflow
distortion at the HS anemometer site CFD simulations of
the airflow over both beams of a detailed representation
of the RRS Charles Darwin were performed. Corrections
of 7.3 % and 3.7 % were applied to the HS sonic in situ
wind speed data for flows over the port and starboard
beam respectively.
The normalised wind speed profile measured above the
bridge of the ship for a flow directly over the port beam
is compared to CFD results in Figure 6. Both profiles
predict a deceleration in wind speed close to the bridge
top and the accelerated region above. In general there is
good agreement (4 % or better) between the two profiles.
0.6
0.5
decelerated region
height, z/H
0.4
0.3 CFD
in situ
0.2
0.1
0 amount the airflow is distorted at anemometer sites. For
-0.4 -0.2 0 0.2 0.4 0.6 0.8
normalised wind speed
Figure 6: Comparison of CFD and in situ wind speed
measurements (adapted from Moat [24]).
Marine CFD 2005: Southampton, UK
4. CFD RESULTS
4.1 RESEARCH SHIPS
VECTIS simulations of the airflow have been performed
over 11 research ships (American, British, Canadian,
French and German) [7]. Anemometers on research ships
are usually located outside of wake regions and in well-
exposed locations, typically on a foremast in the bows of
the ship. Even so wind speed data collected from
different ships and even data from different instruments
on the same ship have disagreed. VECTIS CFD models
have successfully been used to correct for this [6, 7] and
this work will be summarised here.
VECTIS simulations of the air flow over research ships
were performed using a full-scale ship with Reynolds
numbers varying between 6.81 ×10 7 to 1.17 ×10 8 , based
on the ship length. Wind speed at the anemometer sites
are normalised by the free stream, or undisturbed, wind
speed at the height of the anemometer. This is obtained
from the CFD simulations at a large distance abeam of
the anemometer location, typically 250m or more. This is
important to achieve an absolute bias from the free
stream when boundary layer profiles are used.
An example of the wind speed bias present in
measurements made from well-exposed anemometers is
presented in Figure 7. For these instrument positions, the
wind speed measurements can be biased high by up to
7 % and biased low by up to 9 %. Other anemometer
locations may be biased to a greater extent due to their
position relative to the ship superstructure and the
platform it is located on.
15
wind speed bias (%)
10 PORT STARBOARD
5
0
-5
-10
-15 CHARLES DARWIN
-20 RRS DISCOVERY
-25
-90 -60 -30 0 30 60 90
relative wind direction (degrees)
Figure 7: Wind speed bias at well-exposed foremast
anemometer sites on two research ships.
accelerated
flow The shape of a research ship has a large effect on the
1 1.2 1.4 instance, the RRS Discovery (Figure 2) has a streamlined
shape with the foremast platform located well away from
the bridge superstructure. The wind speed measurements
at anemometer sites located on this platform are only
decelerated by a few percent. In contrast the foremast on
the RRS Charles Darwin is close to a block like
superstructure (Figure 1). Consequently these wind speed
measurements are decelerated by up to 9 %.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK
The results of these VECTIS studies have been taken into
account in the design of the new UK research ship the
RRS James Cook.
4.2 TANKERS AND BULK CARRIERS
Little work has been undertaken to quantify the effect of
flow distortion on wind speed measurements obtained
from anemometers located on VOS. This is due to the
several thousand ships participating in the VOS
programme making it unrealistic to study each individual
ship and the variation in ship type, size and shape. A
simple linear model was developed by Moat et al. [11] to
describe the principal dimensions of a tanker and bulk
carrier. These relationships are very similar to those
found more recently by Kent et al. [25] using a much
larger sample of ships. In addition, Moat [11] showed
that tankers and bulk carriers were similar in shape and,
providing that there are no deck cranes present, the same
model can describe their principal dimensions. The mean
flow over a simplified representation of a tanker/bulk
carrier (Figure 3) model of 170 m was studied. The
dimensions of the ship are shown in Table 1.
Bridge Bridge Bridge Freeboard Breadth
to deck to sea length
(m) (m) (m) (m) (m)
13.5 19.4 13.5 5.9 27.3
Table 1: The dimensions of a simple representation of a
tanker geometry of overall length of 170 m.
Firstly, flow visualisation studies were performed in a
wind tunnel to understand the complexity of the flow to
be modelled (Figure 8). A scaled 1:46 generic tanker
model was placed in the low speed section of the
Southampton 2.13 m by 1.52 m wind tunnel. At deck
level a vortex was formed in front of the deck house
block. Above the bridge top the air separated at the sharp
leading edge and created a recirculation region close to
the bridge top with accelerated air above. The
decelerated region increases in depth with distance from
the upwind leading edge and did not reattach to the
bridge top.
H
Figure 8: A wind tunnel study of the flow over the bridge
of a simplified tanker/bulk carrier. The flow is from left
to right.
© 2005: Royal Institution of Naval Architects
CFD studies were performed over the same 1:46 scale
tanker model (Figure 3). A normalised wind speed
profile at a distance of x/H=0.3 back from the leading
edge of the bridge is shown in Figure 9, where H is the
bridge top to deck height. The wind speed was
normalised by the free stream wind speed simulated from
a second VECTIS simulation with no model present.
Wind speeds from anemometers placed close to the
bridge top (at heights of z/H<0.2) can be decelerated by
up to 100 % and may even reverse in direction. Above
this decelerated region the wind speeds are accelerated
by over 10 % and return to within 2 % of the free stream
wind speed at a height of z/H=2.5.
2.5
2 decelerated flow
height, z/H
1.5
1
0.5 bow-on
0
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
normalised wind speed
Figure 9: A vertical profile of the normalised wind speed
above the bridge of the tanker (adapted from [12]).
4.3 CONTAINER SHIPS
A container ship geometry was made by adding an extra
block to the tanker geometry in order to represent the
containers loaded forwards of the deck house block.
Moat [11] found that the large upwind obstacle of the
containers influenced the downstream flow above the
bridge. In addition, it is unknown what effect the
irregular loading of the containers will have on the
airflow across them and consequently the flow above the
bridge. This will be the subject of future work.
5. APPLICATION OF RESULTS
Anemometers on research ships and VOS should be
located as high as possible above the deck, ideally on a
foremast in the bows of the ship. If the anemometer is to
be located above the bridge of the ship, it should be
placed as high as possible above the front edge. Previous
studies suggest that instruments should be located at a
distance of over three mast diameters from cylindrical
masts and spars [26]. The airflow in front of platforms is
generally decelerated; therefore, anemometers located on
platforms should be sited above the platform rather than
in front [12].
VOS vary a great deal in size and type and until recently
the anemometer positions were unknown. With the
recent inclusion of these ship parameters in the WMO
Publication No. 47 metadata [25] the results from CFD
models can be used to examine the effects of airflow
distortion on the wind speed reports from anemometers
on tankers and bulk carriers.

6. CONCLUSIONS
Comparisons with independent wind tunnel data and with
in situ wind speed measurements have determined that
CFD is a valid research tool to investigate the mean air
flow over ships. For anemometers located outside the
wake of upstream obstacles the results agreed to within
4 % (or better).
Wind speed measurements from anemometers on ships
can be biased by the presence of the ships hull and
superstructure. The size of the bias is dependent upon the
anemometer position and the relative wind direction, i.e.
the angle of the ship to the wind. Measurements from
well-exposed anemometers on research ships may only
be biased by about 10 %.
The mean flow above the bridge of typical tankers and
bulk carriers is defined by flow separation at the upwind
leading edge, with a decelerated region close to the
bridge top. Wind speed measurements made from
anemometers above the bridge can be biased high by
over 10 %, or low by up to 100 %. Predicting and
correcting the bias in wind speed measurements reported
from fixed anemometers located on merchant ships
participating in the VOS programme will be the subject
of future work.
Anemometers on ships should be positioned as high as
possible above the deck and if possible located in the
bows of the ship. It is not recommended to locate
anemometers directly in front of platforms or structures.
Anemometers above the bridge of a merchant ship
should be located as high and as far forewards as
possible, ideally above the front edge of the bridge.
The design of a ship will affect the amount the airflow is
distorted. A comparison of two research ships with
different superstructure design has shown that a block-
like superstructure, close to anemometers located on the
foremast in the bow of the ship, can significantly effect
the wind speed measurements. If possible, it is
recommended that the superstructure of research ships
should be streamlined or located as far as possible from
the foremast to reduce its influence on the upstream
airflow.
7. ACKNOWLEDGEMENTS
The authors would like to thank Val Swail
(Meteorological Service of Canada) and Dr. Dave
Hosum (Woods Whole Oceanographic Institution, USA)
for Partial Funding throughout this project.
8. REFERENCES
1. MOCTAR, O. E. and BERTRAM, V., ‘Computation
of viscous flow around fast ship superstructures’, 24th
Symposium on Naval Hydrodynamics, Fukuoka, Japan,
2002, 68-77.
Marine CFD 2005: Southampton, UK
2. JIN, E., YOON, J. and KIM Y., ‘A CFD based
parametric study of the smoke behaviour of a typical
merchant ship’, Practical design of ships and other
floating structures, Y-S Wu, W-C Cui and G-J Zhou
(Ed.) Elsevier Science Ltd. 2001, 459 – 465.
3. TAI, T. C. and CARICO, D., ‘Simulation of DD-963
ship airwake by Navier-Stokes method. J. of Aircraft,
32(6), 1995, 1399-1401.
4. CHENEY, B. T. and ZAN S. J., ‘CFD code validation
data and flow topology for the technical co-operation
program AER-TP2 simple frigate shape’. National
Research Council Canada, Institute for Aerospace
Research, Canada, Report No. LTR-A-035, 1999, 32 pp.
5. KAHMA, K. K., and LEPPÄRANTA, M., ‘On errors
in wind speed observations on R/V Aranda’, Geophysica,
17(1-2), 1981, 155-165.
6. YELLAND, M. J., MOAT, B. I., TAYLOR P. K.,
PASCAL, R. W., HUTCHINGS J. and CORNELL V. C.,
‘Wind stress measurements from the open ocean
corrected for airflow distortion by the ship’, J. of Phys.
Oceanogr., 28(7), 1998, 1511-1526.
7. YELLAND, M. J., MOAT, B. I., PASCAL, R. W. and
BERRY, D. I., ‘CFD model estimates of the airflow over
research ships and the impact on momentum flux
measurements’, J. of Atmos. and Ocean. Tech., 19(10),
2002, 1477-1499.
8. DUPUIS H., GUERIN, C., HAUSER, D., WEILL A.,
NACASS, P., DRENNAN, W. M., CLOCHE, S. and
GRABER, H. C., ‘Impact of flow distortion corrections
on turbulent fluxes estimated by the inertial dissipation
method during the FETCH experiment on R/V
L’Atalante’, J. of Geophys. Res., 108(C3), 2003, 8064,
doi: 10.1029/2001JC001075.
9. POPINET S., SMITH, M. and STEVENS, C.,
‘Experimental and numerical study of the turbulence
characteristics of airflow around a research vessel’, J. of
Atmos. and Ocean. Tech., 21 (10), 2004, 1575-1589 pp.
10. POPINET, S, ‘The GERRIS flow solver. Release
0.6.0. Freely available at http://gfs.sourceforge.net/ ,
2003
11. MOAT, B. I., YELLAND, M. J., PASCAL, R.W. and
MOLLAND, A. F., ’An overview of the airflow
distortion at anemometer sites on ships’, Accepted by the
Int. J. of Climatology.
12. MOAT, B. I., YELLAND, M. J. and MOLLAND, A.
F., ‘Quantifying the airflow distortion over merchant
ships: part II: application of model results’, submitted to
the J. of Atmos. and Ocean. Tech., 2005.
© 2005: Royal Institution of Naval Architects
Marine CFD 2005: Southampton, UK

13. RICARDO, ‘VECTIS Computational Fluid
Dynamics (Release 3.8) user manual’, Ricardo
Consulting Engineers Ltd., Shoreham-by-Sea, UK, 2004,
578 pp.
14. LAUNDER, B. E. and SPALDING D. B., ‘The
numerical computation of turbulent flows’, Computer
Meth. in Appl Mech. and Eng., 3, 1974, 269 – 289 pp.
15. YAKHOT, V., ORSZAG, S. A., THANGAM, S.,
GATSKI, T. B., and SPEZIALE, G., ‘Development of
turbulence models for shear flows by a double expansion
technique’, Physics of Fluids, A4(7), 1992, 1510-1520.
16. EASON, G., ’Improved Turbulence models for
Computational Wind Engineering. PhD. Thesis,
University of Nottingham, UK., 2000, 219 pp.
17. FEMSYS, ‘FEMGV User manual’, Femsys Ltd.,
Leicester, United Kingdom, 1992, 598 pp.
18. SMITH, S. D., ‘Wind stress and heat flux over the
Ocean in gale force winds’, J. of Phys. Ocean., 10, 1980,
709-726.
26. GILL, G. C., OLSSON, L. E., SELA, J. S. and
SUDA, M., ‘Accuracy of wind measurements on towers
and stacks’, Bull. Amer. Meteor. Soc., 48, 1967, 665-674.
9. AUTHORS’ BIOGRAPHIES
Ben Moat holds the current position of Research Fellow
at the Southampton Oceanography Centre, UK. He is
responsible for the CFD ship modelling.
Margaret Yelland holds the current position of Senior
Scientific Officer at the Southampton Oceanography
Centre, UK. She has overall responsibility for the project.
Anthony Molland holds the current position of
Professor of Ship Design at the School of Engineering
Sciences, University of Southampton, UK.
Robin Pascal is an Engineer at the Southampton
Oceanography Centre, UK. His responsibilities include
the implementation of ship based meteorological
measurements.

19. Carrol, B, Personal communication, Ricardo

Consulting Engineers, 2002.

20. MOAT, B. I., ‘Quantifying the effects of airflow

distortion on anemometer wind speed measurements
from merchant ships’ PhD. Thesis, University of
Southampton, UK, 2003, 163 pp.

21. MARTINUZZI, R. and TROPEA, C., ‘The flow

around surface-mounted, prismatic obstacles placed in a
fully developed channel flow’, J. of Fluids Eng., 115,
1993, 85-92.

22. MINSON, A. J., WOOD, C. J., and BELCHER, R. E.,

‘Experimental velocity measurements for CFD
validation’, J. of Wind Eng. and Ind. Aero., 58, 1995,
205-215.

23. NEW, A., and CO-AUTHORS, ‘RRS Charles

Darwin Cruise 141, 1st June – 11th July 2002 Satellite
Calibration and Interior Physics of the Indian Ocean:
SCIPIO’. SOC Cruise Report No. 41, Southampton
Oceanography Centre, Southampton, UK. 2003.

24. MOAT, B. I., YELLAND, M. J. and MOLLAND, A.

F., ‘Quantifying the airflow distortion over merchant
ships: part I: validation of CFD model’, submitted to the
J. of Atmos. and Ocean. Tech., 2005.

25. KENT, E. C., WOODRUFF, S. D. and BERRY, D. I.,

‘WMO publication of metadata and an assessment of
observation heights in ICOADS’, submitted to the J. of
Atmos. and Ocean. Tech., 2005.

© 2005: Royal Institution of Naval Architects

Marine CFD 2005, Southampton, UK

AUTHOR’S CONTACT DETAILS

Michel Visonneau Richard Underhill

Charge de Recherche CNRS-HDR, France Frazer-Nash Consultancy, UK
michel.visonneau@ec-nantes.fr r.underhill@fnc.co.uk

James Dreyer Ben Moat

Penn State University, USA Southampton Oceanography Centre, UK
jjd@wt.arl.psu.edu ben.moat@soc.soton.ac.uk

Sayyed-Maysam Mousaviraad
Petropars Ltd., Iran
mousaviraad@yahoo.com

Neil Southall
Burness Corlett - Three Quays (IOM) Ltd.,
UK
mousaviraad@yahoo.com

Calogero Falletta
Ship-Yacht Designers & Consultants, Italy
sydac@tn.village.it

Jun Zang
University of Oxford, UK
jun.zang@eng.ox.ac.uk

Qiuxin Gao
University of Strathclyde, UK
gao.q.x@strath.ac.uk

Paolo Becchi
CETENA, Italy
paolo.becchi@cetena.it

Norbert Bulten
Wärtsilä, the Netherlands
Norbert.Bulten@wartsila.com

Sven-Brian Müller
University of Duisburg-Essen, Germany
mueller@nav.uni-duisburg.de

Peter Bull
QinetiQ - Haslar, UK
pwbull@qinetiq.com

Cmdr P R Kulkarni
IIT Delhi, India
prk@am.iitd.ernet.in

© 2005: The Royal Institution of Naval Architects

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

A Step towards the Numerical Simulation of Viscous

Flows around Ships at Full Scale - Recent
Achievements within the European Union Project
EFFORT

M. Visonneau
presented by P. Queutey
Laboratoire de Mécanique des Fluides-CNRS UMR 6598
Ecole Centrale de Nantes, Nantes, FRANCE

Marine CFD 2005, RINA, March 30-31, 2005 - Southampton, UK

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Introduction

The EFFORT EU project

The partners of Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling at full scale
Force coefficients
Conclusions

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Why CFD for ship flows ?
Free-surface full-scale computations
The consortium
Turbulence modelling
Force coefficients
Conclusions

Hydrodynamic aspects play an important role in the quality

of a ship

Hull form design is controlled by resistance (evaluation of global

forces) and powering performance (analysis of local flow field in front
of the propeller)

Viscous flow computations for full-scale ship are inhibited by:

Uncertainty about the proper physical modelling
Difficult accessibility of full-scale experimental data
Numerical difficulties associated with the use of very high aspect
ratio grids

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Why CFD for ship flows ?
Free-surface full-scale computations
The consortium
Turbulence modelling
Force coefficients
Conclusions

Goals of EFFORT

European Full Scale FlOw Research and Technology

Project funded within the 5th Framework GROWTH Program
Refinement and validation of CFD methods for ship flows at full
scale
Evaluation of the free-surface, wake field and hull/propeller
interaction
Development of full-scale experimental measurement campaigns
Validations of full-scale flow field predictions
Computations of additional industrial configurations provided by
the industrial sponsors of EFFORT

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Why CFD for ship flows ?
Free-surface full-scale computations
The consortium
Turbulence modelling
Force coefficients
Conclusions

The partners in EFFORT

The consortium
3 technical centers: MARIN (project manager), HSVA, CTO
5 universities: ECN/CNRS, NTUA, HUT, CTH, Maritime University
of Szczecin
5 industrial partners: BEC, IHC Holland NV, Kvaerner Masa
Yards, Rolls Royce Kamewa, Van Voorden Gieterij BV
1 classification society: Lloyd’s Register

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations Goals of WP3
Turbulence modelling
Force coefficients
Conclusions

Goals of the Workpackage 3 “CFD developments”

Questions addressed in WP3 “CFD developments”

What are the free-surface effects at model and full scale ?
Turbulence modelling plays a decisive role at model scale. What
is the situation at full scale ?
Representation of the propeller action on the flow is needed for
comparisons with total wake measurements
Organisation of an internal workshop to compare and validate the
respective simulation tools developped by each partner

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations Goals of WP3
Turbulence modelling
Force coefficients
Conclusions

Goals of the Workpackage 3 “CFD developments”

Questions addressed in WP3 “CFD developments”

What are the free-surface effects at model and full scale ?
Turbulence modelling plays a decisive role at model scale. What
is the situation at full scale ?
Representation of the propeller action on the flow is needed for
comparisons with total wake measurements
Organisation of an internal workshop to compare and validate the
respective simulation tools developped by each partner

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations Goals of WP3
Turbulence modelling
Force coefficients
Conclusions

Goals of the Workpackage 3 “CFD developments”

Questions addressed in WP3 “CFD developments”

What are the free-surface effects at model and full scale ?
Turbulence modelling plays a decisive role at model scale. What
is the situation at full scale ?
Representation of the propeller action on the flow is needed for
comparisons with total wake measurements
Organisation of an internal workshop to compare and validate the
respective simulation tools developped by each partner

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations Goals of WP3
Turbulence modelling
Force coefficients
Conclusions

Goals of the Workpackage 3 “CFD developments”

Questions addressed in WP3 “CFD developments”

What are the free-surface effects at model and full scale ?
Turbulence modelling plays a decisive role at model scale. What
is the situation at full scale ?
Representation of the propeller action on the flow is needed for
comparisons with total wake measurements
Organisation of an internal workshop to compare and validate the
respective simulation tools developped by each partner

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations Goals of WP3
Turbulence modelling
Force coefficients
Conclusions

Partners in WP3

Contributors and CFD tools

Organization Code Notes
CHALMERS CHAPMAN NS code (no FS)
CNRS ISIS NS code (FS capturing)
HSVA COMETa NS code (FS capturing)
HUT FINFLO NS code (FS fitting)
MARIN RAPID NLP code
MARIN PARNASSOS NS code (fixed FS)
NTUA PARALOS NS code (FS fitting)
a
Commercial code

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

The research vessel Nawigator XXI

Characteristics of model and full scale computations

Type Length Wetted Draft Speed Fr
surface
Model scale 5.5155 m 6.66 m2 0.315 m 1.609 ms−1 0.221
Full scale 55.155 m 666 m2 3.15 m 5.093 ms−1 0.221

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

A wide spectrum of computational strategies !

Marin and CTH use a composite approach :inviscid + viscous wo

coupling
NTUA and HUT use a free-surface fitting methodology
HSVA and ECN/CNRS prefer a free-surface capturing technique.

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Free-surface and wall streamlines (from CNRS)

Model scale Full scale

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Local view of free-surface elevation (from CNRS)

Model scale (top) vs. full scale (bottom)

Bow waves Stern waves
ζ/LPP ζ/LPP
-0.006 -0.004 -0.002 0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.020 0.022 -0.006 -0.004 -0.002 0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.020 0.022

MSC 0.06
0.04
MSC
0.04

0.02
0.02
Y/LPP

Y/LPP
0.00 0.00

-0.02
-0.02

-0.04
FSC
-0.04 FSC -0.06

0.90 0.95 1.00 1.05 -0.20 -0.15 -0.10 -0.05 0.00

X/LPP X/LPP

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Local view of free-surface elevation (from CNRS)

3D view : breaking bow wave

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Comparisons of solutions from WP3 partners

Free-surface capturing strategies

Wave elevations
HSVA Wave elevations
CNRS
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40 40

30 30
Y

Y
20 20

10 10

0 0
-60 -40 -20 0 20 -60 -40 -20 0 20
HSVA : FS : K-w SST : 12 knots X CNRS : FS : K-w SST : 12 knots X

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Comparisons of solutions from WP3 partners

Free-surface fitting strategies

Wave elevations
HUT Wave elevations
NTUA
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40 40

30 30
Y

Y
20 20

10 10

0 0
-60 -40 -20 0 20 -60 -40 -20 0 20
HUT : FS : K-w SST : 12 knots X NTUA : FS : K-ε : 12 knots X

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Comparisons of solutions from WP3 partners

Composite approach : MARIN

Wave elevations
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

40

30
Y

20

10

0
-60 -40 -20 0 20
MARIN : FS : K-w SST : 12 knots X

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” Description of the testcase
Free-surface full-scale computations The various computational strategies
Turbulence modelling Global view of the free-surface
Force coefficients
Conclusions

Comparisons of solutions from WP3 partners

Wave cut at Y = 6m Global comparison of the free-surface elevation

0.6 CNRS
HSVA
MARIN
0.4 HUT
NTUA

0.2
Z

-0.2

-0.4

-0.6

-60 -40 -20 0

12 knots : Wave cut Y=6m X

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Turbulence modelling at full scale

What is known about the role played by turbulence ?

At model scale, the turbulence closure is decisive for the reliability
of the prediction of the wake field.
Isotropic eddy-viscosity based closures provide a far less vortical
flow than non-isotropic closures like EASM or RSTM.
Turbulence anisotropy provides a source of longitudinal vorticity
which leads to a much better agreement of computations with the
flow measurements.
At full scale, the role played by turbulence closures is unclear.

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations at full scale - Nawigator XXI

Full-scale flow at X = 1.6m - HUT (k − ω SST) vs. CNRS (k − ω SST)

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations at full scale - Nawigator XXI

Full-scale flow at X = 1.6m - MARIN (Menter 1eq.) vs. CNRS(k − ω

SST)

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations at full scale - Nawigator XXI

Full-scale flow at X = 1.6m - NTUA (k − ε) vs. CNRS (k − ω SST)

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Full-scale flow at X = 1.6m EASM vs k − ω SST closures -

Nawigator XXI

Large influence of turbulence closures on the wake field even at full

scale ! (computations from CNRS)

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations from CNRS with

measurements at full scale (propeller plane)

San Michaelis - Full-scale flow - k − ω SST closure vs measurements

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations from CNRS with

measurements at full scale (propeller plane)

San Michaelis - Full-scale flow - EASM closure vs measurements

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments” The context
Free-surface full-scale computations Nawigator XXI: wake field
Turbulence modelling San Michaelis: wake field
Force coefficients
Conclusions

Comparison of computations from CNRS with

measurements at full scale (propeller plane)

San Michaelis - Full-scale flow - Measurements vs. RSTM closure

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Force coefficients (x1000) Double model flow

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Force coefficients (x1000) Free surface flow

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Conclusions

Full-scale flows are not dramatically more difficult to compute

than model-scale flows,
Reynolds number influence on the free-surface deformation
appears confined to stern waves,
The role of the turbulence modeling has been reinforced by these
computations, which is in contradiction with the results obtained
during the last Gothenburg 2000 workshop on the KVLCC2 ship,
Even at full-scale, turbulence anisotropy should be accounted for
to get a reliable prediction of the flow in front of the propeller.

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Conclusions

Full-scale flows are not dramatically more difficult to compute

than model-scale flows,
Reynolds number influence on the free-surface deformation
appears confined to stern waves,
The role of the turbulence modeling has been reinforced by these
computations, which is in contradiction with the results obtained
during the last Gothenburg 2000 workshop on the KVLCC2 ship,
Even at full-scale, turbulence anisotropy should be accounted for
to get a reliable prediction of the flow in front of the propeller.

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Conclusions

Full-scale flows are not dramatically more difficult to compute

than model-scale flows,
Reynolds number influence on the free-surface deformation
appears confined to stern waves,
The role of the turbulence modeling has been reinforced by these
computations, which is in contradiction with the results obtained
during the last Gothenburg 2000 workshop on the KVLCC2 ship,
Even at full-scale, turbulence anisotropy should be accounted for
to get a reliable prediction of the flow in front of the propeller.

M. Visonneau Marine CFD 2005

 Introduction
The EFFORT EU project
Workpackage 3 “CFD developments”
Free-surface full-scale computations
Turbulence modelling
Force coefficients
Conclusions

Conclusions

Full-scale flows are not dramatically more difficult to compute

than model-scale flows,
Reynolds number influence on the free-surface deformation
appears confined to stern waves,
The role of the turbulence modeling has been reinforced by these
computations, which is in contradiction with the results obtained
during the last Gothenburg 2000 workshop on the KVLCC2 ship,
Even at full-scale, turbulence anisotropy should be accounted for
to get a reliable prediction of the flow in front of the propeller.

M. Visonneau Marine CFD 2005

 BASIN – Development of a Practical Boundary Element
Code
for Hydrodynamic Analysis

Burness Corlett – Three Quays (IOM) Ltd.

Marine Expert Services
 The Primary Problem

• A software tool for accurate analysis and

prediction of large amplitude ship motions in 6
degrees of freedom
and
• A tool that is suitable for use in a general
marine consultancy environment
 Secondary Requirements

• “Efficient” software running on accessible

equipment
• Flexibility for a variety of applications
• Practicality for use by non-specialists
• Integration with our other software
 The Solution

= BASIN

Boundary-element Analysis for

Seakeeping Investigation
 Business Justification
• Lack of suitable software at an affordable price in the
market place
• Control of the specification and development path
• Control of timing of resource commitment
• Ability to exploit parts of the code for other in-house
applications
• Potential for developing wider knowledge within the
organisation during development
• Potential for development of new consultancy
business areas
• Potential for development as a product
 Methodology

• Potential flow – incompressible, inviscid, irrotational

away from boundaries
• Fluid motion is described by a velocity potential, Ф,
which satisfies Laplace’s equation:

• Following Green’s theorem, construct a solution from

a sum of source (σ) and doublet (µ) distributions
placed on each boundary:
 Methodology

• Euler-Lagrange time stepping – solve the fixed

problem at each timestep, then track points according
to boundary conditions:

• Pressure is found on each panel, then integrated

over the vessel
 Implementation
• Vessel geometry in NURBS form is imported via
IGES/STEP
• Panel method is used to approximate the distribution
of sources and doublets on each surface
• Matrix solution is found via multi-threaded direct
solver or Multipole method
• Free surface is updated using fully-nonlinear
boundary conditions
• Vessel is moved according to calculated forces and
moments
 Dynamic Meshing

• Each vessel, free surface and tank walls are

automatically re-meshed to the dynamic waterline at
every timestep
• Dynamic waterlines found using NURBS intersection
routines
• Meshing gets very complicated for large-amplitude
motions
• Meshing is computationally intensive, and the
scheme used must be as efficient as possible
Dynamic Meshing – the procedure
Dynamic Meshing – the procedure
Dynamic Meshing – the procedure
Dynamic Meshing – the procedure
Dynamic Meshing – the procedure
Dynamic Meshing – the procedure
 Wave Generation and Absorption

• Combined method –source terms on tank walls,

damping zones on free surface
• Damping zones absorb differences between actual
wave height/potential and reference values
• Set reference values to zero for absorbing beaches
• Set reference values to analytical solutions for
absorbing wavemakers
Typical damping zones for a six-degree-of-
freedom seakeeping analysis
 Current Status

• Most functionality is in place for 6DOF motions in

oblique irregular waves – re-meshing, wave
generation/absorption etc.
• Testing of method for ship motions in irregular head
seas is well underway
• Roll damping is modelled using empirical equivalent
linear damping
• Some work is needed to make the method more
‘user-friendly’
 Under development

• Green water effects

• Non-linear roll damping model
• Multihulls
• Ship motions in short-crested seas
• Manoeuvring in waves
• Lifting surfaces
• Bilge vortex shedding
 Problem areas

• Runtime – needs optimisation

• Free surface instability – need to suppress both ‘real’
and numerical instabilities
• Re-meshing – difficult to make this 100% reliable for
very extreme motions
Series 60 Wavemaking
Wave contours for the
Series 60 at Fr. 0.316
Head seas motions for the SL7, Fr. 0.3
Head seas motions for the SL7, Fr. 0.3
Head seas motions for the SL7, Fr. 0.3
Head seas motions for the S175, Fr. 0.25 , λ/L =
1.0
Head seas motions for the S175, Fr. 0.25, λ/L =
1.2
Head seas motions for the S175, Fr. 0.25 , λ/L =
1.4
S175, irregular head seas, 6.1m Hs, Fr 0.25
S175, irregular head seas, 4.2m Hs, Fr 0.25
S175, irregular head seas, 6.1m Hs, Fr 0.25 -
animation
 How are we doing?

• It has taken longer than we had hoped

– Conflict with other commitments
– Limitations of internal resources
– Problems with integration of third party code
• Opportunities have been missed because of the
delay
but
• Close to having a readily usable consultancy tool
available
• Some use of the code has been used in anger,
albeit on a limited basis
• Significant improvements in development of other
software packages and their integration
• Some other spin-off benefits already starting to be
realised
Burness Corlett – Three Quays (IOM) Ltd.
Marine Expert Services
 BEHAVIOUR OF
SHIP FUNNEL EXHAUST
IN THE
WAKE OF A BLUFF BODY

Cdr. PR Kulkarni
Prof. SN Singh
Prof. V Seshadri
Indian Institute of Technology - Delhi
Dubious Exhaust Emission Performance
Evolution of ship’s superstructure

Lusitania (1907) Aquitania (1913)

Titanic (1911) QE-2 (1969)

Nordic Empress (1979) Viking Serenade (1985)

Evolution of topside of naval ships

(1884) Emden (1901)

Dreadnaught (1904) Lion (1914)

Hood (1924) Clemencau (1934)

King George (1941) Delhi (1999)

Reduced funnel size on modern naval ships
Unfavorable funnel configuration
Manifestation of Smoke Nuisance Problem

Temperature contours IR Signature

Exhaust smoke-superstructure interaction

Flow Visualisation
Exhaust smoke-superstructure interaction

Flow Visualisation

Experiments
Exhaust smoke-superstructure interaction

Flow Visualisation

Experiments

CFD Simulation
Exhaust smoke-superstructure interaction

Flow Visualisation

Experiments

CFD Simulation

Validation
Exhaust smoke-superstructure interaction

Flow Visualisation

Experiments

CFD Simulation

Validation

Parametric
Investigation
Parametric Investigation

Parametric investigation of 140

cases carried out by varying:

- 4 Velocity Ratio’s (Ve/Vw)

- 7 yaw angles

- 5 different superstructure
configurations
Simplified Superstructure Configuration
Scaling Criteria

The phenomenon of interest is near field

dispersion of jet in the disturbed flow
field created by bluff bodies of the
superstructure.

In the near field, the plume path is

inertia dominated rather than buoyancy
dominated.
Scaling Criteria

Three similarities need to be ensured

between the model and the prototype in
order to establish complete
correspondence between laboratory
experiments and full scale plumes.

Geometric Similarity
Scale model

Kinematic Similarity
Velocity Ratio (K) : Ve/Vw

Dynamic Similarity
Reynolds No.
Experimental setup
Experimental setup

Air from
blower Air from
blower
Experimental setup

Air from
blower Air from
blower
 Experimental setup

Pitot Tube

WIND

Air from
blower Air from
blower
 Experimental setup

Pitot Tube

WIND

Air from
blower Air from
blower
Orifice plate
 Experimental setup

Pitot Tube

WIND

Smoke generator Smoke generator

Air from
blower Air from
blower
Orifice plate
 Experimental setup

Illuminating Lamp

Pitot Tube

WIND

Smoke generator Smoke generator

Air from
blower Air from
blower
Orifice plate
CFD Simulation
FLUENT
GAMBIT Select Solver

Set up Select Turb. Model

Geometry
Specify
Operating Conditions
Meshing of Boundary Conditions
Geometry

Specify
Convergence Accuracy
Specify BC

Solve Equations
by Iterations
Computational Domain

Wall

Outlet
Wall

Inlet
(funnel exits)
Wall

Inlet Wall
(Superstructure and funnel))
Grid Adaptation

Tetrahedral cells

Total Pressure Velocity Magnitude

Grid Adaptation

8,11,219 cells 4,61,037 cells

Total Pressure Velocity Magnitude

Grid Adaptation
Observations from Flow Visualisation

K=1
Observations from Flow Visualisation

K=1
Observations from Flow Visualisation

K=2
Observations from Flow Visualisation

K=1

K=2
Comparison of flow visualisation

K=1
Comparison of flow visualisation

K=2
Comparison of flow visualisation

K=1

K=2
Planes chosen for analysis

Horizontal
Plane Q
Vxy
0.9h
0.9 h

Transverse
Planes
Vyz
1 2 3 4

Centerline
Plane
Vxz
Flow Structure - vortices behind bluff body
Horizontal
Plane
Vxy Q 0.9hh
0.9
Flow Structure
Transverse
Planes
Vyz
1 2 3 4

(a) Plane - 1 (b) Plane - 2

Flow Structure
Transverse
Planes
Vyz
1 2 3 4

(a) Plane - 3 (b) Plane - 4

Flow Structure
Centerline
Plane
Vxz
Recirculation zone in wake of bluff body
Effect of momentum

(a) K = 1 (b) K = 2

(c) K = 3 (d) K = 4
Conclusions
• The qualitative comparison between flow
visualisation from wind tunnel studies & CFD
simulation at K = 1 &2 show very good
agreement, which confirm that closure using
standard k-ε turbulence model can predict the
flow and performance characteristics
reasonably well.
Conclusions
• The qualitative comparison between flow
visualisation from wind tunnel studies & CFD
simulation at K = 1 &2 show very good
agreement, which confirm that closure using
standard k-ε turbulence model can predict the
flow and performance characteristics
reasonably well.

• The flow behind the bluff bodies like the

superstructure block/mast is charecterised by
large velocity gradients, recirculation zones and
strong vortex fields.
Conclusions
• The qualitative comparison between flow
visualisation from wind tunnel studies & CFD
simulation at K = 1 &2 show very good
agreement, which confirm that closure using
standard k-ε turbulence model can predict the
flow and performance characteristics
reasonably well.

• The flow behind the bluff bodies like the

superstructure block/mast is charecterised by
large velocity gradients, recirculation zones and
strong vortex fields.

• They generate a vortex trail that depends on the

shape of the bluff body and the degree of its
streamlining.
Conclusions
• The strength of these trailing vortices is a major
contributor of downwash.
Conclusions
• The strength of these trailing vortices is a major
contributor of downwash.

• A velocity ratio of at least 2 should be

maintained to avoid the problem of downwash.
At velocity ratio greater than 2 , the increased
momentum ensures that the smoke stays well
clear of the deck.
Conclusions
• The strength of these trailing vortices is a major
contributor of downwash.

• A velocity ratio of at least 2 should be

maintained to avoid the problem of downwash.
At velocity ratio greater than 2 , the increased
momentum ensures that the smoke stays well
clear of the deck.

• In case of the exhaust smoke– superstructure

interaction on the naval ships, wherein short
funnels are located in the vicinity of taller
structures that are aerodynamically bluff
bodies, it is the momentum of the exhaust that
decides the behavior of the smoke nuisance
problem.
Conclusions
• The study has further shown that CFD is a
powerful tool capable of predicting the larger
scale features of the exhaust smoke-
superstructure interaction on ships, without
making recourse to wind tunnel testing, a
capability that did not exist before.
Conclusions
• CFD shall allow the designer of modern naval
ships with complex topside layouts to factor the
smoke nuisance problem very early into the
design spiral, which shall enable

– Rank the relative merits of different

superstructure configurations

– Assess suitability for air operations

– Provide inputs for flight simulators

– allow the detection and shortfalls in design

and to find efficient means to eliminate
them
 BEHAVIOUR OF
SHIP FUNNEL EXHAUST
IN THE
WAKE OF A BLUFF BODY

Cdr. PR Kulkarni
Prof. SN Singh
Prof. V Seshadri
Indian Institute of Technology - Delhi
ARL
Penn State

Blade Shaping for Off-Design Performance:

Cavitation & Efficiency in Two-Dimensional Cascades

RINA Marine CFD 2005 James J. Dreyer

Southampton, UK Computational Mechanics Div.
ARL/PSU, USA

30 March 2005
ARL
Penn State Acknowledgements

• Office of Naval Research (Tom Calvert and Lynn Petersen)

• Professor Luigi Martinelli of Princeton University

ARL
Penn State Outline

• Background
• Motivation
• Objective
• Approach
• Results
• Summary & Conclusion
ARL
Penn State Background

• 2002-2004: ONR-sponsored, advanced thruster technology

development program

• Primary focus on (in order of interest) motors, acoustics,

materials, and hydrodynamics

• Hydrodynamics:

Develop & demonstrate a shape optimization tool to insert into

our existing design environment to improve the cavitation
performance of candidates in off-design operation
ARL
Penn State Motivation

• Thrusters spend much of their operational life in off-design

conditions

• An advanced thruster is likely to be rim-driven

• A likely form of incipient cavitation for a rim-driven thruster

is leading-edge, incidence-driven surface cavitation
– Cavitation is a significant source of broad-band noise

• Formal optimization of blade sections has the potential to

create thruster designs more resistant to cavitation in
adverse conditions
ARL
Penn State Objective

• Insert a systematic consideration of off-design surface

cavitation performance into an existing design system

Preliminary Streamline Curvature Mean Streamline Blade Stacking Detailed

Design (SCM) (MSM) (STK) Design
[trade-off [WT, TT testing
studies, etc.] 2
CFD modeling]
1.5

1
R

0.5

0
Section Shape

-0.5
Optimization
-1 0 1
X

• Demonstrate the approach in three phases:

– Two-dimensional Cartesian frame-of-reference
– Rotating quasi-three-dimensional frame-of-reference
– Experimental verification
ARL
Penn State Objective

• Pathology of a cavitation “bucket”

At right: “Floor” of bucket (b.)
Computed cavitation surrounded by steep
bucket for a NACA 65410
8.0
sides
blade section 7.0 NACA 65410 (a. & c.)
6.0

5.0
Non-cavitating
σ

4.0

3.0
b. c.
a.
2.0

1.0 Cavitating Cavitating

a. LE pressure-side surface 0.0 c. LE suction-side surface

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11

α (deg.)

Floor is due to benign suction Objective:

face pressure distribution For a given blade section,
increase the width of the
Steep sides are due to the
floor and, if possible,
formation of LE suction peaks decrease the steepness of
b. Suction-side traveling bubble
the sides
ARL
Penn State Approach

• Use a multi-objective approach to delay the formation of

suction peaks at off-design conditions and, if possible,
minimize the impact on section efficiency

• Multi-point Optimization Problem Statement:

Determine the N D design variables, bi ,

NP

that minimize the composite cost function, I = ∑

n =1
cn I n (w n , F )

Design variables:
Normal movement of surface mesh points [ND = O(100)]

Operating points:
In this work, depending on the scenario, NP = 1, 2, or 3
ARL
Penn State Approach

• Use a simple gradient-based approach for the solution of the

optimization problem where the gradient is determined from
the solution of the adjoint problem, i.e.,

~k
δ b = −α k G
k

NP

where G = ∑ cn G (w n , ψ n , F )
k

n =1

and the tilde denotes smoothing and attenuation

ARL
Penn State Approach

• Cavitation Cost Function:

Reduction of Problematic Suction Peaks

I= ∫ Bb
1
2 ( p − p d ) 2 ds

∂ ∂
pd − εˆ pd = p
∂ξ ∂ξ

ε ⋅ ∂ 2 pa ∂ 2 pa
for ξ − ξ min C ≤ δ
 ∂s 2 ∂s 2
εˆ = 
p
min C p stag pt.

 0 for ξ − ξ min C > δ

 p
ARL
Penn State Approach

• Efficiency Cost Function

Minimization of Axial Force

I = ∫ τ xx S x + τ xy S y dξ
Bc

1  ∂u i ∂u j 

τ ij = − pδ ij + 
+
Re  ∂x j ∂x i 


When combined with a fixed transverse force, this is analogous to

maximizing thrust at fixed power for a thruster
ARL
Penn State Approach

• A constant transverse or lift force is maintained at the design

condition throughout the design iterations by a
straightforward stagger angle adjustment

−1
 ∂Cl 
β =β
k k −1
+   δ Cl
 ∂β 
∂Cl
where is continuously updated throughout the design
∂β
ARL
Penn State Approach

• Shape Optimization: Design Cycle Schematic

Candidate Blade Section

Flow
FlowField
FieldSolution
Solution STAGGER ADJUST (n = 1)

Adjoint
AdjointB.C.s
B.C.s
NP
Adjoint
AdjointField
FieldSolution
Solution
NI

Gradient
GradientCalculation
Calculation
SMOOTH & ATTENUATE

Blade
BladeShape
ShapeChange
Change

Domain Re-meshing Final

Domain Re-meshing FinalDesign
Design
ARL
Penn State Results

• CASE 1: Optimize a 30° stagger cascade of NACA 65410

hydrofoils (p/c = 1) for improved cavitation performance over
a +/- 5° incidence range 1.8

BASELINE
1.6

SYN103I 1.4

0.5
1.2

1.0

σ
0.0
0.8

Cp
0.6
-0.5
0.4

0.2
-1.0

0.0

α
28 30 32 34 36 38 40 42 44 46 48
0.0 0.2 0.4 0.6 0.8
x

SYN103I SYN103I
0.5 0.5
SUCTION PEAKS

0.0 0.0
Cp

Cp
-0.5 -0.5

-1.0 -1.0

0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
x x
ARL
Penn State Results

• CASE 1: Optimization History for 50 Design Iterations

35
35
CL
σ
NP = 3:
30 CL C1 = 0.0 , C2 = 1.0 , C3 = 1.0
σ
1.40
30
40 CL
0.00 0.30 • Design condition used only for
40 σ lift constraint
-0.02
1.30 ∆β

• Constant CL at α = 35° is
-0.04 0.25
1.20
-0.06

1.10
maintained by stagger angle
-0.08
0.20 adjustment (∆β)
1.00 -0.10

-0.12
• Off-design spread in σi is

∆β
CL
σ

0.90 0.15

-0.14
reduced significantly in 50
0.80
-0.16
design cycles
0.10
0.70
-0.18 • Slight increase in σi is
0.60 -0.20 observed at the design
0.05

0.50
-0.22 condition
-0.24
0.40 0.00
0 10 20 30 40 50

NDES
ARL
Penn State Results

• CASE 1: Hydrofoil Pressure Distributions & Section Shapes

SECTION DESIGN CYC: 0

C p DESIGN CYC: 0

DESIGN: 0.5
SECTION DESIGN CYC: 50

C p DESIGN CYC: 50
T
C p DESIGN CYC: 50

α = 35° HIGH INCIDENCE:

0.0

α = 40° :

Cp, CTp
C1 = 0.0
Design point used -0.5 C3 = 0.5
only to fix CL
1.225 → σ → 0.977
0.552→ σ → 0.603 -1.0

SYN103I
0.0 0.2 0.4 0.6 0.8 1.0
x

SECTION DESIGN CYC: 0 SECTION DESIGN CYC: 0

C p DESIGN CYC: 0 C p DESIGN CYC: 0

SECTION DESIGN CYC: 50 SECTION DESIGN CYC: 50

0.5 C p DESIGN CYC: 50

T
C DESIGN CYC: 50
p
LOW INCIDENCE: 0.5 C p DESIGN CYC: 50
T
C p DESIGN CYC: 50

0.0
α = 30° : 0.0
Cp, CTp

Cp, CTp
C2 = 0.5
1.105 → σ → 0.639
-0.5 -0.5

-1.0 -1.0

SYN103I SYN103I
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x
ARL
Penn State Results

• CASE 1: Cavitation Bucket

2.4

2.2
NACA 65410 • The 3-operating-point shape
3-POINT OPTIMIZED
2.0 optimization had the desired
1.8 effect of broadening the
1.6 Design Points cavitation bucket
1.4 • At a given σ the optimized
σ

1.2 hydrofoil offers ~3° greater

1.0 incidence range for cavitation
0.8
free operation for the same
transverse force
0.6

0.4
• A slight increase in the level of
the floor of the bucket is
0.2
evident
α
26 28 30 32 34 36 38 40 42 44
ARL
Penn State Results

• CASE 2: Single point optimization of the same 30° stagger

cascade of NACA 65410 hydrofoils for efficiency improvement

NP = 1: C1 = 1.0
1.0
0.610 SECTION DESIGN CYC: 0
1.70 Cp DESIGN CYC: 0
0.609 1.60 SECTION DESIGN CYC:100
-0.07
1.50 Cp DESIGN CYC:100
0.608 CD
η≡ 1.40 0.5

0.607
CL 1.30

-0.08 1.20

0.606 1.10

CL, CD

Cp
1.00 0.0

∆β
η

0.605 0.90
CL -0.09
0.80
0.604 CD
∆β 0.70

0.603
η 0.60
-0.5
-0.10 0.50

0.602 0.40

0.30
0.601
0.20
-0.11 -1.0
0.600
0.10 SYN103I
0 10 20 30 40 50 60 70 80 90 100

NDES 0.0 0.2 0.4 0.6 0.8 1.0

x

• Gain in efficiency at the expense of the generation of a LE suction peak

ARL
Penn State Results

• CASE 3: Combined cavitation/efficiency optimization for the

30° stagger NACA 65410 cascade over a +5° incidence range

NP = 2: C1 = 0.0 , C2 = 1.0 NP = 2: C1 = 5.0 , C2 = 1.0

+5° cavitation only 0° efficiency & +5° cavitation

0.6046 -0.060 1.00 1.25 0.6046 -0.060 1.00 1.25
0.90 1.20 0.90 1.20
0.6044 0.6044
0.80 0.80
-0.080 1.15 -0.080 1.15
0.70 0.70
0.6042 0.6042
0.60 1.10 0.60 1.10

0.6040 -0.100 0.50 1.05 0.6040 -0.100 0.50 1.05

0.40 1.00 0.40 1.00
0.6038 0.6038
0.0/1.0 0.30 0.30
-0.120 0.95 -0.120 0.95
0.20 0.20
0.6036 35 CL 0.6036
0.10 0.90 0.10 0.90
35 CD 5.0/1.0
CL, CD

CL, CD
35 σ
∆β

∆β
η

η
σ

σ
0.6034 0.00 0.85 0.6034 0.00 0.85
-0.140 -0.140
40 σ -0.10 35 CL -0.10
0.6032 35 η 0.80
0.6032 35 CD 0.80
∆β -0.20
0.75 35 σ
-0.20
0.75
0.6030
-0.160 -0.30
0.6030 40 σ -0.160 -0.30
-0.40 0.70 35 η -0.40 0.70
∆β
0.6028 -0.50 0.65 0.6028 -0.50 0.65
-0.180 -0.180
-0.60 0.60 -0.60 0.60
0.6026 0.6026
-0.70 -0.70
0.55 0.55
-0.200 -0.80 -0.200 -0.80
0.6024 0.6024
-0.90 0.50 -0.90 0.50

0.6022 -1.00 0.45 0.6022 -1.00 0.45

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100

NDES NDES

• In general: Direct trade-off between efficiency & cavitation

(at least for a well-designed starting point)
ARL
Penn State Results

• CASE 4: Shape optimization of a 3.5° stagger cascade of NACA 0010

hydrofoils for improved cavitation performance over +/- 5°
Cl NACA 65410
Cd NACA 65410
C l NACA 0010
C d NACA 0010
NP = 3: C1 = 1.0 , C2 = 1.0 , C3 = 1.0
σ NACA 65410 σ NACA 0010
1.5 Cl NACA 0010 1.5
C l 0010 Optimized
Cd NACA 0010 C d 0010 Optimized
σ NACA 0010 7 0.00 σ 0010 Optimized 7 0.00

1.0 6 1.0 6

-0.05 -0.05

5 5

0.5 0.5
-0.10 -0.10
4 4

Cd

Cd
Cl

Cl
σ

σ
0.0 3 0.0 3
-0.15 -0.15

2 2

-0.5 -0.5
-0.20 -0.20

1 1

-1.0 0 -0.25 -1.0 0 -0.25

-20 -15 -10 -5 0 5 10 15 -20 -15 -10 -5 0 5 10 15

α α

• The design flow angle does not lie near the bottom of the cavitation bucket for the
baseline NACA 0010
• For the optimized hydrofoil, the design point lies in the center of the floor of the
cavitation bucket & the extent of the floor is set by the range of conditions
considered in the design
ARL
Penn State Results

• CASE 4: Hydrofoil Loading & Section Shapes

Baseline NACA 0010 vs. 65410 at α = 2° Optimized vs. Baseline NACA 0010 at α = 2°
1.0 1.0
0.7 0.7
NACA 65410 NACA 0010
0.6 NACA 65410 C p 0.6 NACA 0010 Cp
NACA 0010 0.5 Optimal 0.5
NACA 0010 Cp Optimal Cp
0.5 0.5

0.4 0.4
0.0 0.0
y/c

y/c
Cp

Cp
0.3 0.3

0.2 -0.5 0.2 -0.5

0.1 0.1
-1.0 -1.0
0.0 0.0

-0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4
x/c x/c

• All three hydrofoils yield equivalent loading at this condition

•The NACA 65410 and optimized 0010 display no suction peaks
•The optimized 0010 gives the broadest cavitation bucket of the three O(5°-7°)
ARL
Penn State Results

• CASE 5: Shape Optimization of a 0° incidence NACA 65410

hydrofoil for cavitation
Rec = 2.2x106

• The purpose of this exercise is to provide a case for experimental verification

• Note that the optimization was done on a cascade configuration with p/c = 2.5
but the experiment was done on an isolated blade in test section of width 2.5c
ARL
Penn State Results

• CASE 5: Optimization History & Hydrofoil Loading and

Section Shapes
NP = 3: C1 = 0.0 ; C2 = 1.0 ; C3 = 1.0
0.0060
Cd Design
1.6 1.0
σ Design
0.7
1.5 σ High-incidence NACA 65410
σ Low-incidence 0.0059 NACA 65410 Cp
1.4 0.6 Optimized 0.5
Optimized Cp
1.3 0.0058
0.5
1.2
0.0
1.1 0.0057 0.4

y/c
Cd

Cp
σ

1.0
0.3 -0.5
0.9 0.0056

0.8 0.2 High-incidence

0.0055 -1.0
0.7
0.1
0.6
0.0054
0.5 0.0 -1.5

0.4
0 10 20 30 40 50 -0.4 -0.2 0.0 0.2 0.4
NDES x/c

• The design condition is used only to constrain the transverse loading

• Goal was to be as aggressive as possible on cavitation:
σ spread: 1.61/0.45 → 0.92/0.56 ; 10% increase in Cd
ARL
Penn State Results

• CASE 5: Hydrofoil Lift

Baseline NACA 65410 & Optimized Section Performance:

• All simulations were predictions
Comparison of 2D RANS with Measurements
• Some 3D effects from end wall BL were
predicted
Baseline (2D RANS)
1.0
Optimal (2D RANS)
Baseline C p
• Post-mortem analysis showed some
Optimized Cp
Optimized load cell
interaction between blade suction surface
0.5
Baseline-3D corr. and inflow boundary location
Cl

• HOWEVER: Both the predictions and

0.0
measurements show the baseline and
optimized hydrofoils give the same
performance over the range of α
-0.5

• Experimentally, the lift was measure with

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

α
1 2 3 4 5 6 7 8 9 10 11
a load cell (OPTM) and integration of the
measured surface pressure
ARL
Penn State Results

• CASE 5: Cavitation Bucket

• Overall very good correlation between

Baseline NACA 65410 & Optimized Section Performance:
Comparison of 2D RANS with Measurements visual cavitation inception calls and
8.0
predictions using suction peak criterion
Baseline (2D RANS)
7.0
Optimal (2D RANS) • Agreement is excellent for baseline and
Baseline
6.0 Optimized optimized over the floor of the bucket and
5.0
for negative incidence angles
• Simulations somewhat over-predict σ for
σ

4.0

positive incidence angles for both

3.0
hydrofoils
2.0
• There is more scatter in the calls for
1.0 positive angles
0.0
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 • The ∆ between baseline and optimized
α
appears to be fairly well-predicted over
the range of α
ARL
Penn State Results
• CASE 5: Pressure Distributions
NACA 65410
Baseline NACA 65410, In-tunnel: Baseline NACA 65410, In-tunnel: Baseline NACA 65410, In-tunnel:
Comparison of 2D RANS with Measurements Comparison of 2D RANS with Measurements Comparison of 2D RANS with Measurements

1.0 1.0 1.0

0.5 0.5 0.5

0.6 0.6
0.0 0.6 0.0 0.0

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

0.4 0.4
0.4
UPPER

Cp

Cp

Cp
y

y
-1.5 UPPER -1.5 -1.5
LOWER UPPER
LOWER
α = -5.1 deg. UPPER Cp -2.0 UPPER Cp -2.0 LOWER -2.0
LOWER Cp α = 0.4 deg. UPPER Cp
0.2
Measured Cp -2.5
LOWER Cp
-2.5
0.2 α = 5.4 deg. LOWER Cp -2.5
0.2 Measured Cp
Measured Cp
-3.0 -3.0 -3.0

-3.5 -3.5 -3.5

0.0 0.0
0.0
-4.0 -4.0 -4.0

-0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4

x x x

Optimized
Optimized, In-tunnel: Optimized, In-tunnel: Optimized, In-tunnel:
Comparison of 2D RANS with Measurements Comparison of 2D RANS with Measurements Comparison of 2D RANS with Measurements

1.0 1.0 1.0

0.5 0.5 0.5

0.6
0.6 0.6
0.0 0.0 0.0

-0.5 -0.5 -0.5

0.4 -1.0 -1.0 -1.0

0.4 0.4

UPPER
Cp

Cp

Cp
y

UPPER

y
-1.5 -1.5 -1.5
LOWER UPPER
LOWER LOWER
UPPER Cp
α = -5.65 deg. -2.0
α = 0.34 deg. UPPER Cp -2.0
UPPER Cp -2.0
0.2 LOWER Cp
LOWER Cp 0.2
α = 5.84 deg. LOWER Cp
Measured Cp -2.5 0.2 -2.5 -2.5
Measured Cp Measured Cp
-3.0 -3.0 -3.0

0.0 -3.5 -3.5 -3.5

0.0 0.0

-4.0 -4.0 -4.0

-0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4

x x x
ARL
Penn State Summary & Conclusion

• Formulated & demonstrated a robust & computationally

inexpensive approach to shape optimization of blade
sections for cavitation & efficiency improvement over a
range of operating conditions in a cascade configuration

– Formulation based on a steepest descent approach where the

composite gradient is determined from the solution of the
appropriate adjoint problem

– Demonstrated on:
• Multi-point cavitation-only
• Efficiency only
• Multi-point combined cavitation/efficiency
• Off-design starting point

– 100 design iterations with 3 operating points + lift constraint

using RANS modeling requires ~96 min on Intel Xeon (2.8 GHz)
ARL
Penn State Summary & Conclusion

• Experimental verification of the approach on an “isolated”

NACA hydrofoil section
– Very good correlation between observed and predicted
cavitation inception (surface-type)
– The ∆ in cavitation performance between the baseline and
optimized sections was well-predicted

• As a result of this work, the methodology has been

extended to rotating, quasi-three-dimensional coordinates
– Existing thruster:
• Start-up and cross-flow conditions
• Potential for >100 ft improvement in
cavitation inception at Bollard condition
– SYN103I-Q3D is compatible with design
environment
 Consideration on deviations in torque prediction

Norbert Bulten 31.03.2005

Propulsor Technology - Hydrodynamics

Wärtsilä Propulsion Netherlands

© Wärtsilä
 Topics

z Introduction
„ CFD analyses on propulsion systems:
special attention for pressure distributions and
cavitation margins

z CFD analyses of 2D test cases

„ half body
„ NACA profiles

z CFD analysis of waterjet system

„ validation of waterjet inlet CFD model
„ validation of mixed-flow pump CFD model
„ comparison CFD prediction with actual waterjet performance

z Conclusions

© Wärtsilä 2
 CFD analyses of propulsion systems at
WPNL

Tunnel
Thrusters
© Wärtsilä 3
 CFD analyses of propulsion systems at
WPNL

Ducted Propellers
Waterjets

Special attention to pressure distribution and

margins against cavitation
© Wärtsilä 4
 Validation of pressure distribution with
2D test cases

z Test case with analytical solution of pressure distribution

„ geometry based on stream function of source and uniform
flow:
UNIFORM FLOW

q
ψ= θ + vr sinθ
SOURCE

2π

„ analytical solution of pressure distribution:

 sinθ  2  sin(2θ ) 
Cp = −   + 
 π − θ   π − θ 
 
© Wärtsilä 5
 Accuracy of overall pressure prediction
Pressure distribution along half body

1.2
Analytical solution
1
CFD result

0.8

0.6
ERROR 19%

0.4
Cp [-]

0.2
ERROR 3%

0
0 30 60 90 120 150 180
-0.2

-0.4

-0.6
Angle theta [degree]

Error maximum pressure = 19%

Error minimum pressure = 3%
© Wärtsilä 6
 Lift / drag results of 2D NACA0012
profile
Lift and drag comparison NACA0012

1 0.04

Cl_exp
Cl_CFD
Cd_exp

0.75 Cd_CFD 0.03

Cd [-]
Cl [-]

0.5 0.02

0.25 0.01

0 0
0 1 2 3 4 5 6 7 8 9
Section angle of attack [degree]

Comparison of CFD results with literature

© Wärtsilä 8
 Concluding remarks on 2D calculations

z Pressure distribution is calculated quite accurate over most of

chord length.

z Clear deviation of pressure prediction in stagnation point.

z Lift prediction of NACA profile is accurate.

z Drag prediction is rather poor, due to error in stagnation

pressure prediction.

© Wärtsilä 9
 Working principle of waterjet

z Thrust is based on: ∆V m

F = m .a = m . = .∆ V
∆t ∆t
m
T = .(V j − V i ) = ρ .Q .(V j − V i )
∆t

vj

vi
VOLUME FLOW THROUGH SYSTEM IS GOVERNING TOTAL THRUST

© Wärtsilä 10
 Validation of separate CFD models

z Waterjet inlet on model scale

„ pressure distribution along ramp
„ cavitation inception pressure at cutwater
„ velocity distribution upstream of impeller

z Mixed-flow pump on model scale

„ head curve
„ pump efficiency

z Complete waterjet on full scale

„ flow prediction

© Wärtsilä 11
 Waterjet inlet duct CFD analyses

← LOW IVR ↑

← HIGH IVR →

v ship
Inlet Velocity Ratio IVR =
v pump
© Wärtsilä 12
 Comparison of pressure measurements
with CFD results
Comparison CFD and experiments pressure at ramp

0.5

0.4

0.3

Cp_121
0.2 Cp_150
Cp_187
Cp_219
Cp [-]

0.1 Cpm 121

Cpm 150
Cpm 187
0 Cpm 219
Geom
0 200 400 600 800 1000 1200

-0.1

-0.2 INLET ROOF

GEOMETRY

-0.3
Distance along roof [mm]

Pressure along ramp for 4 IVR conditions

© Wärtsilä 13
 Comparison of cavitation inception
experiments with CFD results
Cavitation Inception comparison

4
Sigma_Vtunnel [-]

CFD calculations
1
Measurements

0
1 1.2 1.4 1.6 1.8 2 2.2 2.4
IVR [-]

Cavitation inception line

© Wärtsilä 14
 Comparison of velocity measurements
with CFD results

Velocity distribution (medium IVR)

© Wärtsilä 15
 Comparison of velocity measurements
with CFD results

Velocity distribution (high IVR)

© Wärtsilä 16
 Numerical solution methods for
rotating impellers

z Quasi-steady with multiple frames of reference (MFR)

„ fixed mesh; no rotor-stator interaction
„ source term for body forces due to rotation
„ relative quick method

z Fully transient with moving mesh

„ mesh moves with sliding interface at each time step
„ CPU intensive calculation

© Wärtsilä 17
 Mixed-flow pump CFD analyses

Pressure distribution of mixed-flow pump

© Wärtsilä 18
 Comparison of pump performance
measurements with CFD results
Pump performance

160%

140%

120%
Relative head H/Href [%]

100%

80%

60%

40% CFD_head

EXP_head

20% CFD_efficiency

EXP_efficiency
0%
50% 60% 70% 80% 90% 100% 110% 120%
Relative volume flow Q/Qref [%]

Relative pump head and efficiency

© Wärtsilä 19
 Analysis of complete system

z Combination of inlet and pump CFD models

z Comparison with traditional waterjet performance prediction

software
„ volume flow through system
„ thrust
„ power

© Wärtsilä 20
 CFD model of waterjet

Mesh of inlet + pump

Approx. 1.1 M cells

Calculations are made with

constant pump RPM and varying ship speed

© Wärtsilä 21
 Comparison with performance
prediction software

Volume flow through system:

V_ship Q_predict Q_CFD Difference

3 3
[knots] [m /s] [m /s] [%]
31 12.96 12.93 -0.23 %
35 13.11 13.08 -0.19 %
39 13.26 13.19 -0.54 %
43 13.41 13.31 -0.72 %
47 13.55 13.49 -0.45 %

© Wärtsilä 22
 Determination of thrust

z Thrust based on
„ direct integration of forces on waterjet structure
„ direct integration of forces on complete numerical domain (with
compensation for flat plate drag)

© Wärtsilä 23
 Thrust of waterjet
Thrust comparison

120%
Prediction software
115%
CFD results - waterjet structure
CFD results - numerical domain
110%
Thrust_CFD/T_design [-]

105%

100%

95%

90%

85%

80%
30 32 34 36 38 40 42 44 46 48 50
Ship speed [knots]

© Wärtsilä 24
 Power of waterjet
Power comparison
110.0%
prediction software
107.5% CFD calculation
Power/Power_design_point [%]

105.0%

102.5%

100.0%

97.5%

95.0%

92.5%

90.0%
30 32 34 36 38 40 42 44 46 48 50
Ship speed [knots]

© Wärtsilä 25
 Correlation between lift / drag and
torque
Effect of overprediction of drag on torque

Thrust

V_in

Additional
torque

Addition drag

© Wärtsilä 26
 Conclusions

z CFD methods can be used to analyse propulsion systems

(propellers and waterjets) within acceptable time scales.

z Calculation of pressure distribution shows good results.

z Thrust prediction is quite accurate over a large range of

operating conditions.

z Torque, and consequently power, is over-predicted with a few

percent. This off-set is due to error in stagnation point
pressure distribution.

© Wärtsilä 27
Influence of Scale Effects on the Hydrodynamic
Characteristics of Propellers

Sven-Brian Müller and Moustafa Abdel-Maksoud

Institute of Ship Technology and Transportation Systems (IST),

University Duisburg-Essen

MARINE CFD 2005

RINA – The Royal Institution of Naval Architects

31. März 2005

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 1
Overview

1. Introduction
2. Problem definition
3. Numerical investigation
4. Calculation method
5. Results
6. Summary

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 2
1. Introduction

• Increase of the applied scale factor because of the limited dimensions of test
facilities in the towing tanks

• Extrapolation problem of the thrust and torque coefficient from model test
results to the full-scale

• Availability of extensive information about the characteristics of the propeller

flow by applying CFD methods
Example: Wall shear stress
distribution on the propeller blade

Propeller diameter of current container ship projects

Reference: Mewis and Klug

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 3
2. Problem Definition

Industrial project

• Experimental determination of the thrust and torque coefficients of a propulsion

system at model scale

• Optimisation of the test procedure

• Use of modern measuring technique

• Extrapolation of the measurement results to the full-scale

• Extrapolation according to the ITTC procedure 1978: the prediction includes a

correction for the thrust and for the torque coefficients

• These corrections consider the influence of: Reynolds’ number, Ratio

thickness/cord length and pitch ratio

• Thus the ITTC procedure is hardly able to consider the local flow conditions

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 4
2. Problem Definition

Final stage of a project

Competition between different designs of the propeller manufacturers

Risk
Because of extrapolation inaccuracies from model to full-scale a design may be
selected, which does not represent the best solution for the full-scale

Result
High scientific and economic interest about the extrapolation method for propeller
manufacturers and shipyards

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 5
3. Numerical Investigation

Investigated propeller geometry

Z=4; AE/A0 = 0.55; P/D = 0.9 ; dh/D = 0.25;

(t/c)0.7 = 0.03867; c0.7/D = 0.2948; Fresh water 25 C°
Dmodel = 0.25 m; Dfull-scale = 4, 6, 8, 10, 12 m

Computation properties
• Open water test condition, i.e. with parallel inflow
J = 0.6, 0.7, 0.8

• Computations of the flow around one blade

• Consideration of the interaction effect with other

blades by application of a periodic
boundary condition in space

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 6
3. Numerical Investigation

Dimensions of the calculated area

Diameter: 5.5 D
Length: 20 D

Area of propeller blades

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 7
3. Numerical investigation

Numerical grid: approx. 9x105 grid points per blade

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 8
4. Calculation method

CFX-5.7

• Finite volume method

• Turbulence model SST
• Rotating and stationary coordinate systems
• Discretisation scheme of second order
• Multi-grid technology
• Structured and unstructured computing grid
• Local refinement and multi-block technology

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 9
5. Results

Open water test diagramm

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 10
5. Results

Ratio of the full-scale KQ coefficient relative to the model

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Universität Duisburg-Essen]
© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 11
5. Results

Ratio of the full-scale KT coefficient relative to the model

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Universität Duisburg-Essen]
© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 12
5. Results

Ratio of the full-scale efficiency coefficient relative to the model

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Universität Duisburg-Essen]
© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 13
5. Results

Pressure distribution - model

Rn = 3.33E+05

J = 0.8

suction side pressure side

Rn = 2.52E+05

J = 0.6

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 14
5. Results

Pressure differences relative to Rn = 3.33E+05; J = 0.8

Rn = 1.24E+07

suction side pressure side

Rn = 1.11E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 15
5. Results

Pressure differences relative to Rn = 2.52E+05; J = 0.6

Rn = 1.20E+07

suction side pressure side

Rn = 1.08E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 16
5. Results

Pressure differences relative to Rn = 3.33E+05; J = 0.8

Rn = 1.24E+07

leading edge trailing edge

Rn = 1.11E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 17
5. Results

Pressure differences relative to Rn = 2.52E+05; J = 0.6

Rn = 1.20E+07

leading edge trailing edge

Rn = 1.08E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 18
5. Results

Wall shear stress - model

Rn = 3.33E+05

J = 0.8

suction side pressure side

Rn = 2.52E+05

J = 0.6

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 19
5. Results

Wall shear stress differences relative to Rn = 3.33E+05; J = 0.8

Rn = 1.24E+07

suction side pressure side

Rn = 1.11E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 20
5. Results

Wall shear stress differences relative to Rn = 2.52E+05; J = 0.6

Rn = 1.20E+07

suction side pressure side

Rn = 1.08E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 21
5. Results

Wall shear stress differences relative to Rn = 3.33E+05; J = 0.8

Rn = 1.24E+07

leading edge trailing edge

Rn = 1.11E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 22
5. Results

Wall shear stress differences relative to Rn = 2.52E+05; J = 0.6

Rn = 1.20E+07

leading edge trailing edge

Rn = 1.08E+08

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 23
6. Summary

• Investigation of the viscous flow around a propeller geometry at 6

Reynolds‘ numbers and 3 advance ratios
• Computation of the full-scale data using CFD results and the ITTC
procedure

With increase of the Reynolds‘ number:

• Dependence of the thrust coefficient:

- CFD results: high dependence, clear increase
- ITTC results: limited dependence
• Reduction of the pressure coefficient on the suction side
• General high reduction of the dimensionless wall shear stress,
especially on the leading edge and tip regions

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 24
Thank you for your attention!

If you have questions, please contact!

Sven-Brian Müller
Tel: 0049-203-379-1167
Mueller@nav.uni-duisburg.de

Moustafa Abdel-Maksoud
Tel: 0049-203-379-2539
Maksoud@nav.uni-duisburg.de

Influence of Scale Effects on the Hydrodynamic Characteristics of Propellers

© Copyright [IST, Sven-Brian Müller and Moustafa Abdel-Maksoud] 25
 MARINE
MARINE CFD
CFD 2005
2005
30 -31 March
30-31 March2005,
2005,Southampton
Southampton(UK)
(UK)

CETENA
CETENA
Italian
Italian Ship
Ship Research
Research Centre
Centre
HYDRODYNAMIC
HYDRODYNAMICDepartment
Department
Via
ViaIppolito
Ippolitod’Aste,
d’Aste,55- -16121
16121Genova
Genova

Comparison
Comparison between
between RANSE
RANSE calculations
calculations and
and panel
panel
method
method results
results for
for the
the hydrodynamic
hydrodynamic analysis
analysis of
of
marine
marine propellers
propellers
Ing. Paolo BECCHI Ing. Chiara PITTALUGA
paolo.becchi@cetena.it chiara.pittaluga@cetena.it
Tel: 0039 010 599 5480 Tel: 0039 010 599 5478
 List
List of
of Contents
Contents
PROPELLER GEOMETRY
CFD tools and computation settings
PANEL tool : PROPACE
z Panel Method Sensitivity Analysis
z Panel Grid settings
z Viscous Correction Formulation
z Kutta-Joukowsky Condition
RANSE tool : CFX5
z Settings
z Domain Decomposition
z Propeller mesh
z Solver settings
VALIDATION
z Comparison of Experimental Data and Numerical Results
z Viscous Correction Formulation results
z CFX5 – PROPACE Comparison
CONCLUSIONS
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) 2
 European Project :: LEADING
European Project LEADING EDGE
EDGE
Prediction of leading edge and tip flow for the
design of quiet and efficient screw propellers
8 MARIN (The Netherlands)
8 SSPA (Sweden)
8 DTU, Technical University of Denmark (Denmark)
8 CHALMERS University of Technology (Sweden)
8 FLOWTECH International AB (Sweden)
8 IZAR Construccciones Navales S.A. (Spain)
8 VTT, Technical Research Centre of Finland, (Finland)
8 WARTSILA Propulsion Netherlands BV (The Netherlands)
8 HSVA, Hamburgische Schiffbau-Versuchsanstalt GmbH (Germany)
8 NLR Stichting Nationaal Lucht- en Ruimtevaartlaboratorium (The Netherlands)
8 SINTEF Applied Mathematics (Norway)
8 CETENA S.p.A (Italia)
8 HUT, Helsinki University of Technology (Finland)
8 FINCANTIERI –Cantieri Navali Italiani S.p.A. (Italy)
8 CTO , Ship Design and Research Centre (Poland)

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) 3

 PROPELLER Geometry
PROPELLER Geometry
Main
Main propeller
propeller geometry
geometry data
data
in model scale
in model scale
•• Propeller diameter: 0.233
Propeller diameter: 0.233 m
m
•• Boss/diameter:
Boss/diameter: 0.321
0.321
•• Number
Number of
of blades:
blades: 44
•• Rotational
Rotational Speed:
Speed: 14
14 HzHz
•• Maximum
Maximum Skew:Skew: 27
27°°
•• AE/AO:
AE/AO: 0.729
0.729
The
The propeller
propeller was
was analysed
analysed at
at an
an
off - design pitch.
off-design pitch.

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

4
 CFD tools and computation settings

Panel
Panel Method
Method PROPACE
PROPACE
PROPACE
PROPACE is is aa software
software code
code for
for the
the
hydrodynamic
hydrodynamic analysis
analysis ofof marine
marine
propellers,
propellers, developed
developed by
by CETENA
CETENA since
since
90’s.
90’s.

MAIN
MAIN CHARACTERISTICS
CHARACTERISTICS
•• Program
Program based
based onon aa potential
potential flow
flow
model
model
•• Viscous
Viscous effects
effects evaluated
evaluated by by an an
empirical
empirical formulation
formulation
•• All
All the
the blades
blades are
are considered
considered in
in the
the
computation
computation (no(no key-blade
key-blade approach)
approach)
•• Steady
Steady version
version available
available

Future
Future implementations
implementations
•• Unsteady
Unsteady version
version (to
(to be
be tested)
tested)
•• Sheet
Sheet cavitation
cavitation model
model (to
(to be
be tested)
tested)
•• Induced
Induced pressure
pressure analysis
analysis (in
(in progress)
progress)
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
5
 CFD tools and computation settings

Panel
Panel Method
Method Sensitivity
Sensitivity Analysis
Analysis Sensitivity
Sensitivity Analysis
Analysis -- KT
KT results
results
MKC
MKC -- 30
MKC
30 SPANWISE
SPANWISE points
points KT
KT
MKC -- 35
35 SPANWISE
SPANWISE points
pointsKT
KT
MKC
MKC -- 40
40 SPANWISE
SPANWISE points
pointsKT
KT
MKC - 45 SPANWISE points
MKC - 45 SPANWISE points KTKT
MKC
MKC -- 50
50 SPANWISE
SPANWISE points
points KT
KT
KT
KT exp
exp

Experimental
Experimental KT
KT value
value

MKC
KT -- MKC
KT
30
30 32
32 34
34 36
36 38
38 40
40 42
42 44
44 46
46 48
48 50
50
Chordwise
Chordwise Direction
Direction Panel
Panel Number
Number

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY 6

 CFD tools and computation settings

Panel
Panel Grid
Grid settings
settings
Based on the Sensitivity Analysis, The panel grid configuration was set in order
Based on the Sensitivity Analysis, The panel grid configuration was set in order
to
to provide
provide the
the higher
higher defined
defined geometrical
geometrical discretization:
discretization:
HUB
HUB CONFIGURATION
CONFIGURATION BLADE
BLADE PANEL
PANEL GRID
GRID
Hub
Hub zone 1:
zone 1: Number
Number of Points in:
of Points in:
Length
Length [m]:
[m]: 0.1166
0.1166 Spanwise
Spanwise direction
direction 35
35
Number
Number of
of sections:
sections: 25
25 Chordwise
Chordwise direction
direction 50
50
Hub
Hub zone
zone 2:
2: Points
Points distribution
distribution law:
law:
Length [m]:
Length [m]: 0.0583
0.0583 Spanwise direction
Spanwise direction COSIN
COSIN
Number of sections: 25
Number of sections: 25 Chordwise
Chordwise direction
direction COSIN
COSIN
Hub
Hub zone
zone 3:
3:
Length
Length [m]:
[m]: depending
depending onon the
the blade
blade WAKE
WAKE CONFIGURATION
CONFIGURATION
root
root section
section length
length in
in axial
axial Number
Number of of sections
sections in
in
direction
direction axial
axial direction:
direction: 200
200
Number of sections: 25
Number of sections: 25 Wake
Wake Pitch
Pitch COSTANT
COSTANT
Hub
Hub zone
zone 4:
4: Wake
Wake contraction
contraction model
model No
No Contraction
Contraction
Length
Length [m]:
[m]: 0.1166
0.1166
Number
Number of
of sections:
sections: 25
25

Hub
Hub zone
zone 5:
5:
Length [m]:
Length [m]: 0.0583
0.0583
Number
Number of
of sections:
sections: 25
25

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

7
 CFD tools and computation settings

Viscous
Viscous Correction
Correction Formulation
Formulation
The
The thrust
thrust and
and torque
torque coefficient
coefficient are
are estimated
estimated by
by the
the potential
potential and
and
friction
friction contributions:
contributions: K == K −− K
K K K
TT TOT
TOT TT POT
POT TT FRIC
FRIC
K TOT =
KQQTOT =K POT +
KQQ POT +KKQQFRIC
FRIC
TTFRIC Q
QFRIC
FRIC =
=ZZ⋅⋅ FRIC =
=ZZ⋅⋅
K FRIC K FRIC
KTT FRIC KQQFRIC
ρρ⋅⋅nn22 ⋅⋅D
D4
4
ρρ⋅⋅nn22 ⋅⋅D
D5
5

The
The total
total friction
friction contributions
contributions are
are evaluated
evaluated by
by the
the sum
sum of
of the
the single
single
friction
friction contribution
contribution of
of each
each panel
panel on
on the
the blade:
blade:
11   PP(ζζ) 
TFF == ∑ ⋅⋅ρρ⋅⋅V VRR 2 ⋅⋅C SinArc
CDDii,,jj ⋅⋅SSii,,jj ⋅⋅Sin 
2
T tan
Arctan
2
ii,,jj 2  π ⋅ ζ ⋅
 π ⋅ ζ ⋅ D 
D
11   PP(ζζ)  rr
QFF == ∑ ⋅⋅ρρ⋅⋅V VRR 2 ⋅⋅C CosArc
CDDii,,jj ⋅⋅SSii,,jj ⋅⋅Cos ⋅⋅(ζζ ⋅⋅R
R) ζζ ==
2
Q tan
Arctan
2
ii,,jj 2  π ⋅ ζ ⋅
 π ⋅ ζ ⋅ D 
D R
R
where
where S
Si,j means the area of each blade panel, C Di,j is
i,j means the area of each blade panel, CDi,j is the
the drag
drag coefficient
coefficient
evaluated
evaluated by
by aa semi-empirical
semi-empirical formulation
formulation as
as follows…
follows…

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

8
 CFD tools and computation settings

Kutta -Joukowsky Condition

Kutta-Joukowsky Condition
The
The PROPACE
PROPACE solver
solver can
can be
be set
set with
with two
two different
different Kutta-Joukowsky
Kutta-Joukowsky condition
condition models:
models:
the
the Morino
Morino Kutta
Kutta Condition
Condition (MKC)
(MKC) andand the
the Iterative
Iterative Kutta
Kutta Condition
Condition (IKC).
(IKC). The
The
difference
difference concerns
concerns the
the value
value of
of the
the doublets
doublets that
that aa panel
panel method
method needs
needs to
to consider
consider on
on
the
the blade
blade wake.
wake.

(∆φ)ww = ( )
Wake
Wake doublet
doublet formulations:
formulations: r r
MKC : φuu − φll
φ − φ + U ∞∞ ⋅ rtete
(∆φ)ww = ( )
r r
IKC : φ − φ + U ∞∞ ⋅ rtete + (∆W )ww
uu ll
Linear
Linear equation
equation system:
system:
N N
 ∂φ 
Npp N
NWW
Npp
∑ ijij jj ∑ imim mm ∑ ijij ∂n  , i = 1,2,K, N pp
D φ + W ( ∆ φ) = S
jj==11 m==11
m jj==11   jj
The
The IKC
IKC introduces correction ∆W
introduces aa correction ∆W aimed
aimed to to provide
provide the
the minimum
minimum pressure
pressure
difference
difference at
at the
the trailing
trailing edge,
edge, but
but itit has
has aa lower
lower stability
stability due
due to
to the
the fact
fact that,
that, in
in this
this
case,
case, the
the solver
solver tries
tries to
to minimize
minimize aa numerical
numerical problem
problem andand then
then the
the solution
solution can
can be be
affected
affected by
by aa non
non physical
physical optimisation.
optimisation. ForFor this
this work,
work, the
the calculation
calculation was
was performed
performed
with
with the
the MKC.
MKC.

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

9
 CFD tools and computation settings

RANSE
RANSE Method
Method :: CFX
CFX 5.7
5.7
SOLVER:
SOLVER: CFX
CFX v5.7
v5.7
finite
finite volume
volume RANSE
RANSE solver
solver
GRIDDER:
GRIDDER: ICEM
ICEM CFD-hexa
CFD-hexa
structured
structured multiblock
multiblock mesh
mesh generator
generator

Open water => steady calculation

1 blade
Rotating reference frame
B.C.
• Inflow at 2.0D undisturbed velocity
• Outflow at 4.0D zero of 1st pressure derivative
• Far-field at 5.0D undisturbed velocity
• Periodic boundary condition (non 1:1matching)
• Domain skew 270°
• Blade section employed for the periodic surface generation up to 0.9

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

10
 CFD tools and computation settings

RANSE
RANSE Method
Method
Multiple Frame of Reference System (MFR):

8 the Rotating Reference Frame is applied to the fluid domain close to the propeller
blade in order to add additional terms compared to those in the inertial system

8 A Frozen Rotor algorithm with a GGI interface guarantees the conservation of the fluid
properties at the domain connection
OUTER domain

Domain Decomposition :
¾ structured multiblock
¾ 2 independent sub-domain
¾ 1.130.000 nodes

INNER domain

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

11
 CFD tools and computation settings

Domain
Domain Decomposition
Decomposition
OUTER domain

INNER domain

INNER sub-domain : 830.000 nodes

H Grid ( 6 main block)
O Grid (5 main block) around the blade

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

12
 CFD tools and computation settings

Domain
Domain Decomposition
Decomposition
OUTER domain

INNER domain

OUTER sub-domain : 200.000 nodes

H Grid ( 5 main block)
O Grid (3 main block around the interface)

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

13
 CFD tools and computation settings

Solver
Solver settings
settings
NUMERICAL METHOD
¾ Velocity formulation Absolute velocity dependent
¾ MFR Multi Frame Ref. System
¾ Spatial discretisation Finite-volume colloc.
¾ Convection Terms Discr Upwind stnd
¾ Order of acc. : Second
¾ Diffusion Terms Discr Upwind
¾ Order of acc. : Second
¾ Pressure-velocity CouplingFully coupled
¾ Turbulence Model Two-equations, K-e
¾ Wall function Wall function 3000 iteration 2°
order scheme
¾ without press.grad.
¾ y+ (5 ¸ 55)
¾ Convergence iterations 3100,
¾ Residual RMS < 10-6
¾ multi-grid algorithm AMGW
¾ Initial Value undisturbed velocity ( Va, 0, 0) [m s-1]
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
15
 Validation
Validation

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) 16

 VALIDATION

Comparison
Comparison of
of experimental
experimental data
data and
and
RANSE
RANSE results
results
KT
J Cet / Exp 1
Exp 2 / Exp 1 Cet / Exp2

0.2 0.15% 2.12% 2.28%

0.5 1.00% 3.35% 2.32%

0.736 4.14% 0.18% 3.80%

10KQ
J
Exp 2 / Exp 1 Cet / Exp 1 Cet / Exp2

0.2 0.14% 4.03% 3.88%

0.5 1.36% 6.15% 4.73%

0.736 1.81% 6.40% 4.50%

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

18
 VALIDATION

Viscous
Viscous Correction
Correction Formulation
Formulation
For
For confidentiality
confidentiality reasons,
reasons, the
the detailed
detailed formulations
formulations cannot
cannot be be published,
published,
but
but they
they are
are indicated
indicated by
by the
the following
following notations:
notations:
•• “ACTUAL”:
“ACTUAL”: concerns
concerns the
the formulation
formulation actually
actually implemented
implemented in in the
the
PROPACE
PROPACE code;
code;
•• “LAMINAR”:
“LAMINAR”: concerns concerns aa formulation
formulation involving
involving aa friction
friction coefficient
coefficient
(CF)
(CF) depending
depending on on the
the Reynolds
Reynolds number
number inin laminar
laminar condition;
condition;
•• “TURBULENT”:
“TURBULENT”: concerns
concerns aa formulation
formulation involving
involving aa friction
friction coefficient
coefficient
(CF)
(CF) depending
depending on on the
the Reynolds
Reynolds number
number inin turbulent
turbulent condition;
condition;
•• “ENVELOPE”:
“ENVELOPE”: concernsconcerns aa formulation
formulation involving
involving where
where thethe friction
friction
coefficient
coefficient (CF)
(CF) is
is the
the maximum
maximum between
between the
the laminar
laminar and
and the
the turbulent
turbulent
condition.
condition.

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

19
 VALIDATION

Propace
Propace MODEL
MODEL scale
scale results
results

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

20
 VALIDATION

Propace
Propace MODEL
MODEL scale
scale results
results

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

21
 VALIDATION

Propace
Propace FULL
FULL scale
scale results
results

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

22
 VALIDATION

Propace
Propace FULL
FULL scale
scale results
results

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

23
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
1.00
J=0.734
J=0.734 -- r/R=0.600
r/R=0.600
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
27
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.734
J=0.734 -- r/R=0.700
r/R=0.700
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
28
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.734
J=0.734 -- r/R=0.800
r/R=0.800
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
29
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.734
J=0.734 -- r/R=0.900
r/R=0.900
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
30
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
1.00
J=0.500
J=0.500 -- r/R=0.600
r/R=0.600
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
34
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
1.00
J=0.500
J=0.500 -- r/R=0.700
r/R=0.700
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
35
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.500
J=0.500 -- r/R=0.800
r/R=0.800
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
36
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.500
J=0.500 -- r/R=0.900
r/R=0.900
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
37
 CFX5 – PROPACE COMPARISON

Pressure
Pressure distribution
distribution over
over the
the blade
blade

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY

39
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.200
J=0.200 -- r/R=0.600
r/R=0.600
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
41
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.200
J=0.200 -- r/R=0.700
r/R=0.700
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
42
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
1.00
J=0.200
J=0.200 -- r/R=0.800
r/R=0.800
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
1
-0.70
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
43
 CFX5 – PROPACE COMPARISON

CFX5
CFX5
PROPACE
PROPACE (BACK)
PROPACE
(BACK)
PROPACE (FACE)
(FACE)
1.00
J=0.200
J=0.200 -- r/R=0.900
r/R=0.900
1.00
0.90
0.90
0.80
0.80
0.70
0.70
0.60
0.60
0.50
0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
-CP
-CP

0.00
0.00
-0.10
-0.10
-0.20
-0.20
-0.30
-0.30
-0.40
-0.40
-0.50
-0.50 p − p0
-0.60
-0.60 CP =
-0.70 1
-0.70 ⋅ ρ ⋅ VR 2
-0.80
-0.80 2
-0.90
-0.90
-1.00
-1.00
0.00
0.00 0.10
0.10 0.20
0.20 0.30
0.30 0.40
0.40 0.50
0.50 0.60
0.60 0.70
0.70 0.80
0.80 0.90
0.90 1.00
1.00
x/C
Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) SUMMARY
44
 CONCLUSIONS
CONCLUSIONS
The two codes have pointed out a similar behavior for the trust
coefficient KT, a difference in a range 4-5% have been found with respect
to the available experimental data. Differently the torque coefficient KQ
shows different accuracy: a lost of 4-5% between RANSE and
experimental, whilst the accuracy decrease to 12-15% for the panel –
experimental comparison.
To this aim, the investigation carried out on the viscous correction
formula for the panel method allowed to point out that it is possible to
reduce this accuracy lost with different formulation to take into account the
viscous effects.

Monday, 18 April 2005 MARINE CFD 2005 - Southampton (UK) 45

Thank you for your attention

The End
INDEX
 MARINE CFD 2005
30-31 March 2005, Southampton University, UK

NUMERICAL SIMULATION OF
FREE SURFACE WAVE INDUCED
SEPARATION
S. H. Sadathosseini,
S. M. Mousaviraad,
M. H. Sadr
(Petropars Ltd., IRAN)

1
 Outline
„ Introduction
„ Previous Studies
Petropars „ Computational Method
Ltd.,
IRAN
„ Numerical Modeling
„ Results
„ Concluding remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

2
 Introduction
Hydrodynamics of surface piercing bodies involves wave
effects, because of wave making, wave breaking, and/or
incident waves.
Petropars
Ltd.,
IRAN “Free surface wave induced separation” is separation
solely due to wave induced effects.

The problem is very important in ocean and marine

engineering with regard to ship performance, wake
Marine CFD
2005, 30-31 signatures, and platform stability.
March 2005,
Southampton
University,
UK

3
 Introduction
Complexities: free surface deformations, vorticity,
turbulence, along with the already formidable subject of
three dimensional boundary layer separation.
Petropars
Ltd.,
IRAN Numerical modeling: lower cost and higher level of
producible data.

Present study: numerical simulation of free surface wave

induced separation, VOF method for free surface
Marine CFD
2005, 30-31 modeling. Also, shape effects are studied.
March 2005,
Southampton
University,
UK

4
 Outline
„ Introduction
„ Previous Studies
Petropars
Ltd., „ Computational Method
IRAN
„ Numerical Modeling
„ Results
„ Concluding remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

5
 Previous Studies (1)
¾The wave induced separation was first identified by
Chow (1967) who performed experimental study using
vertical (surface piercing) and horizontal foils.
Petropars
Ltd.,
IRAN

¾It was also studied by Stern et al. (1989) using a

surface piercing flat plate with attached wave
generating upstream horizontal submerged foil (foil-
Marine CFD plate model).
2005, 30-31
March 2005,
Southampton
University,
UK

6
 Previous Studies (2)
¾Choi and Stern (1993) performed laminar and
turbulent CFD calculations. In comparison to the
experimental data, the extent of separation region was
Petropars
Ltd.,
grossly over/under predicted for the laminar/turbulent
IRAN
solutions.

¾Zhang and Stern (1996) studied the problem through

RANS simulation with exact nonlinear kinematic and
Marine CFD
approximate dynamic free surface boundary conditions.
2005, 30-31
March 2005,
Southampton
University,
UK

7
 Previous Studies (3)
¾Pogozelski et al. (1997) performed experimental study
of free surface wave induced separation, but with
different foil geometry.
Petropars
Ltd.,
IRAN

¾Metcalf et al. (2001) provided detailed experimental

data documentation of the wave elevations and surface
pressures for surface piercing NACA0024 hydrofoil.
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

8
 Outline
„ Introduction
„ Previous Studies
Petropars
Ltd., „ Computational Method
IRAN
„ Numerical Modeling
„ Results
„ Concluding remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

9
 Computational Method
RANS (Reynolds Averaged Navier Stokes) simulation
using RSM (Reynolds Stress Model) for turbulent
modeling.
Petropars
Ltd.,
IRAN
Free surface modeling: an interface capturing method,
volume of fluid (VOF) which takes the effect of outer
air into consideration and solves RANS equations
simultaneously for both water and air.

Marine CFD As a result, the present study shows better agreement

2005, 30-31
March 2005, with the experimental results, since all previous
Southampton
University, numerical calculations used surface tracking methods.
UK

10
 Computational Method
In VOF method, an additional transport equation is
solved for the volume fraction of water in each cell.

Petropars It is accomplished by the solution of a continuity

Ltd.,
IRAN equation for the volume fraction of water. This
equation has the following form:

∂α w r
+ v .∇α w = 0
Marine CFD
2005, 30-31 ∂t
March 2005,
Southampton
University,
UK

11
 Computational Method
The volume fraction equation will not be solved for air;
the volume fraction of air will be computed based on the
following constraint:
Petropars
αw + αa = 1
Ltd.,
IRAN

The properties appearing in the transport equations are

determined by the presence of the component phases in
each control volume. The density in each cell, for
example, is given by:
ρ = α w ρw + α a ρa
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK All other properties (e.g., viscosity) are also computed in
this manner.
12
 Outline
„ Introduction
„ Previous Studies
Petropars „ Computational Method
Ltd.,
IRAN
„ Numerical Modeling
„ Results
„ Concluding remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

13
 Numerical Modeling
The first model is a NACA0024 foil having a chord
length of 1.2 m, a span of 1.5 m (in water), and a
maximum thickness of 29 cm.
Petropars
Ltd.,
IRAN This is a simplified geometry that has insignificant
separation at large depths, thus making an ideal
geometry by isolating the wave induced separation.

Three conditions are simulated, with reference to the

Marine CFD
2005, 30-31 experimental data, i.e., Fr= (0.19, 0.37, 0.55) and the
March 2005,
Southampton corresponding Re= (0.822, 1.52, 2.26) × 10 6.
University,
UK

14
 Numerical Modeling
The second test case is a circular cylinder. Having a
diameter of 1.2 m (equal to the foil chord length), it is
intended to evaluate the shape effects on the wave
induced separation.
Petropars
Ltd.,
IRAN Since the geometries are symmetrical, only half domains
which consist of 215000 hexahedral structured cells are
solved. The cells near free surface in both air and water
fields are designed to be very small (2 mm height) to
Marine CFD
catch accurate water deformation results.
2005, 30-31
March 2005,
Southampton
University,
UK

15
 Outline
„ Introduction
„ Previous Studies
Petropars „ Computational Method
Ltd.,
IRAN
„ Numerical Modeling
„ Results
„ Concluding remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

16
 Results

Petropars
Ltd.,
X-wall shear stress contours
IRAN on the foil for Fr=0.19

Marine CFD
2005, 30-31
March 2005,
At this Froude number, the separation region is very small,
Southampton
University,
and the wave effects are limited to depths very close to the
UK free surface. Therefore, the flow recovers to 2D at about
Z=-30cm.
17
 Results

Petropars
Ltd.,
X-wall shear stress contours
IRAN on the foil for Fr=0.37

Marine CFD
2005, 30-31
March 2005,
The wave effects become strong, extend to about Z=-1 m,
Southampton
University,
and the separation in free surface area starts at about
UK X/L=0.32.

18
 Results

Petropars
Ltd.,
X-wall shear stress contours
IRAN on the foil for Fr=0.55

Marine CFD
2005, 30-31
The wave effects become even stronger and extend to
March 2005,
Southampton
about Z=-1.15 m.
University,
UK
The separation region is smaller than that of Fr=0.37, and
the separation in free surface area occurs at about
X/L=0.56.
19
 Results

Petropars
Ltd.,
X-wall shear stress contours
IRAN on the circular cylinder
for Fr=0.19

Marine CFD
2005, 30-31
Separation point at free surface: X/L=0.86
March 2005,
Southampton
Separation point at large depths: X/L=0.67
University,
UK
The separation pattern is dominated by the shape effects,
and the free surface wave effects only delay the separation.
20
 Results
X-wall shear stress
contours
on the circular cylinder
for Fr=0.37 and Fr=0.55
Petropars
Ltd.,
IRAN

Separation points at free surface: X/L=0.4 and X/L=0.64,

respectively, which are very different from those of large
depths, but almost the same as those of the hydrofoil.
Marine CFD At these high Froude numbers, the separation pattern is
2005, 30-31
March 2005, Froude dependent, and free surface wave induced
Southampton
University, separation is dominant. The shape effects only slightly
UK
delay the separation point.

21
 Results
0.15
Wave profile 0.1
Experiment [Zhang &Stern] Numerical [Zhang &Stern]
Present Simulation
along the foil 0.05

and comparison

Z(m)
0

Petropars with experimental -0.05

Ltd., and numerical -0.1

IRAN
results -0.15
0 0.2 0.4 0.6 0.8 1
for Fr=0.19
X/L

The wave profile is similar to that of a typical ship.

The results are nearly as accurate as those of Zhang &
Marine CFD
2005, 30-31 Stern, indicating that the effects of air are not significant
March 2005,
Southampton at low Froude numbers.
University,
UK
The bow wave peak is about 1.6 percent of L.

22
 Results

0.1

Wave profile 0.05

along the

z(m)
0
Petropars circular cylinder -0.05
Ltd.,
IRAN for Fr=0.19 -0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L

The free surface waves are dominated by the strong

Marine CFD
pressure distribution of bluff body.
2005, 30-31
March 2005,
Southampton
University,
The bow wave peak is about 6.6 percent of L.
UK

23
 0.15

0.1 Experiment [Zhang &Stern] Numerical [Zhang &Stern]

Present Simulation

Results 0.05

Z(m)
0

-0.05
Wave profile -0.1

along the foil -0.15

0 0.2 0.4 0.6 0.8 1
for Fr=0.37 X/L

0.15

0.1 Experiment [Zhang &Stern] Numerical [Zhang &Stern]

Petropars Present Simulation
Ltd., 0.05
Wave profile
Z(m)

IRAN 0

-0.05 along the foil

-0.1 for Fr=0.55
-0.15
0 0.2 0.4 0.6 0.8 1

X/L

Marine CFD
The present modeling agrees better with the experimental
2005, 30-31
March 2005,
data, which suggests the significance of the air effects at
Southampton
University,
higher Froude numbers.
UK
The bow wave peak for Fr=0.37 and Fr=0.55 are 6 and
12 percent of L, respectively.
24
 Results 0.1

0.05
Wave profile

z(m)
along the circular 0

cylinder -0.05

for Fr=0.37 -0.1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L
Petropars
Ltd.,
IRAN
0.3

0.2
0.1 Wave profile
z(m)

0 along the circular

-0.1
-0.2
cylinder
-0.3 for Fr=0.55
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Marine CFD X/L

2005, 30-31
March 2005,
Southampton
University, The bow wave peak for Fr=0.37 and Fr=0.55 are 9 and
UK
18 percent of L, respectively, which are much greater
than those of the foil.
25
 Results
0.15

0.1 Fr=0.19 Fr=0.37 Fr=0.55

Comparison of the 0.05

wave profiles

Z(m)
0

of the hydrofoil -0.05

Petropars
at different -0.1

Ltd., Froude numbers -0.15

0 0.2 0.4 0.6 0.8 1
IRAN
X/L

For Fr=0.19, the wave pattern is similar to that of ships.

For Fr=0.37, the bow wave becomes more significant; the
wave steepness is larger; and the wave pattern is
Marine CFD
2005, 30-31 relatively flat in the separation region. For Fr=0.55, the
March 2005,
Southampton free surface has an even more complicated wave system,
University,
UK with increase in bow wave peak, wave steepness, and
distortion of the free surface in the separation region.
26
 Results
0.3

0.2 Fr=0.19 Fr=0.37 Fr=0.55

Comparison of the 0.1

wave profiles of the

z(m)
0
Petropars circular cylinder at -0.1
Ltd.,
IRAN different Froude -0.2

numbers -0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

For Fr=0.19, the free surface elevations are affected by

Marine CFD the strong pressure distribution of the body. For Fr=0.37
2005, 30-31
March 2005, and Fr=0.55, the wave patterns are Froude dependent.
Southampton
University, The wave height, the wave steepness, and the distortion
UK
in the separation region become larger with Froude.

27
 Results (Shape Effects Investigation)
Fr=0.19
Comparison 0.1

between the wave 0.05

Circular Cylinder NACA0024

profiles of the

z(m)
0
Petropars foil and the circular
Ltd., -0.05
IRAN cylinder
-0.1
at Fr=0.19 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L

The shape effects are dominant; the wave pattern of the

Marine CFD foil is similar to that of ships, mainly affected by the high
2005, 30-31
March 2005, pressure stagnation point; and the wave pattern of the
Southampton
University, circular cylinder is affected by the strong pressure
UK
distribution of the blunt shaped body.

28
 Results (Shape Effects Investigation)
Fr=0.37
0.15
Comparison Circular Cylinder NACA0024
0.1
between the wave
0.05
profiles of the

z(m)
Petropars foil and the circular 0

Ltd.,
IRAN cylinder -0.05

at Fr=0.37 -0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X/L

The wave pattern, i.e. the wave steepness and the trend
Marine CFD of the wave elevations, is Froude dependent. However,
2005, 30-31
March 2005, the bow wave peak and the distortions in the separation
Southampton
University, region are larger for the circular cylinder, because of its
UK
blunt shape.

29
 Results (Shape Effects Investigation)
Fr=0.55
Comparison 0.3

Circular Cylinder NACA0024

0.2
between the wave 0.1

profiles of the

z(m)
0

Petropars foil and the circular -0.1

Ltd., -0.2
IRAN cylinder -0.3
at Fr=0.55 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X/L

Although the primary pattern of the wave remains Froude

Marine CFD dependent, the shape effects seem to become stronger;
2005, 30-31
March 2005, the wave steepness, the wave height, and the distortions
Southampton
University, in the separation region are more significantly affected
UK
by the shape effects.

30
 Circular Cylinder NACA0024

Frictional Pressure Total Frictional Pressure Total

0.015
0.3

Variations of Drag Coefficients

Variations of Drag Coefficients
0.25
0.01

Results 0.2

0.15

0.1

0.05
0.005

0
0

-0.05 -0.005
0.19 0.37 0.55 0.19 0.37 0.55
Fr Fr

Variations of drag coefficients Variations of drag coefficients

versus Fr for the versus Fr for the
circular cylinder hydrofoil
Petropars
Ltd.,
IRAN All values are subtracted from the corresponding values
for Fr=0.19.
As Fr increases, the pressure drag coefficient increases
due to the effects of the bow wave. The frictional drag
Marine CFD coefficient, on the other hand, decreases as Fr increases
2005, 30-31
March 2005, from 0.19 to 0.37 and then increases a little as Fr
Southampton
University, increases further to 0.55, which is consistent with the size
UK
of the separation region.

31
 Results (Wave Breaking)

Petropars
Ltd.,
IRAN

Marine CFD
2005, 30-31 The phenomenon is extremely complicated due to the
March 2005,
Southampton effects of unsteadiness, turbulence, and air trapping. The
University,
UK present simulation was able to cope with these difficulties;
nevertheless further developments are needed to explain
the details of the problem. 32
 Outline
„ Introduction
„ Previous Studies
Petropars
Ltd.,
„ Computational Method
IRAN
„ Numerical Modeling
„ Results
„ Concluding Remarks
Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

33
 Concluding Remarks
9The drag coefficients, the free surface waves, and the
separation patterns are all Froude dependent.

Petropars 9The bow wave peak increases with Fr.

Ltd.,
IRAN
9The separation region increases as Fr increases from
small (0.19) to medium (0.37), and then decreases as Fr
increases further to high (0.55).
Marine CFD
2005, 30-31 9Associated with the wave pattern, the pressure drag
March 2005,
Southampton coefficient increases with Fr. The frictional drag
University,
UK coefficient decreases and then increases with Fr in
agreement with the separation region.
34
 Concluding Remarks

9The effects of air on the accuracy of the numerical

modeling are very significant, especially for medium and
Petropars high Fr.
Ltd.,
IRAN
9At very low Fr, wave induced separation is not strong,
and shape effects dominate the flow and separation regime.

9At higher Fr, the wave effects become very significant,

Marine CFD
2005, 30-31 and govern the primary pattern of the separation.
March 2005,
Southampton
University,
UK

35
 Concluding Remarks
9The depthwise extent of the separation, despite of its
streamwise extent, is highly affected by the shape effects.

Petropars 9The distortions of the wave pattern in the separation and

Ltd.,
IRAN wake regions are shape dependent.

9At very high Fr, the flow becomes unsteady, and the
waves arise and break down periodically. Numerical
simulation is possible only with robust free surface and
Marine CFD
2005, 30-31 turbulence modeling techniques and fine grids.
March 2005,
Southampton
University,
UK

36
 Concluding Remarks
9The present numerical model is useful both in taking
insight into the complicated problem of free surface wave
induced separation, and in providing an implement of
Petropars
Ltd.,
design and optimization for ocean engineering applications.
IRAN

9 The next steps are to:

9Evaluate the performance of turbulence models.

Marine CFD
9Study the effects of Reynolds number.
2005, 30-31
March 2005,
9Assess the critical Fr, at which the wave breaking
Southampton
University,
flow begins.
UK

37
 Questions?

Petropars
Ltd.,
IRAN

Thank You for Your Attention

Marine CFD
2005, 30-31
March 2005,
Southampton
University,
UK

38
CFD’2005 • Southampton • March 30, 2005

Numerical simulation of yaw effect

Quixin Gao
Vladimir Shigunov
Dracos Vassalos

Department of Naval Architecture

and Marine Engineering
Universities of Strathclyde and Glasgow
 Presentation outline
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

• Introduction

• Mathematical model

• Test cases

• Validation

• Attitude effects

• Yaw effects

• Conclusions
 Introduction
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

The traditional method currently used by towing tanks

for ship manoeuvrability predictions is based on

• calculation of hydrodynamic derivatives of forces and

moments with respect to the individual degrees of freedom

• These derivatives are then used in manoeuvrability

simulation programs to evaluate ship performance in
real manoeuvres for design purposes
 Introduction
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

The hydrodynamic derivatives are calculated

• from experiments,
• using potential flow simulation methods
• or regression formula

Potential methods can only be used in manoeuvrability

calculations with additional care because of strong
viscous effects

Experimentally based predictions are expensive and

can be applied only to the type of hulls they were obtained for

In this study, numerical solution of RANS equations was

used to predict force and flow field around Series 60 ship
model under steady oblique motion
 Objectives
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

• To validate the accuracy of RANSE approach for steady

sway motion

• To evaluate the attitude effects

• To study yaw effects

 Previous work
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

• Longo and Stern (1997)

towing tank experiments of yaw effect on ship flow
for a 3.048 m Series-60 model
• Patel, Ju and Lew (1990)
one of first numerical studies into asymmetric effect
* neglected free surface effect
* only stern part of model was included in the simulation
• Cura Hochbaum (1998)
numerical simulation for steady drift motion
* free surface and ship attitude effects were not
considered in the calculation
• Alessandrini and Delhommeau (1998)
calculated viscous free surface flows past Series-60
in steady sway motion and steady circulation
* ship attitude effects were excluded
 Mathematical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Governing Equations

• continuity equation:
r
∇ ⋅V = 0

• momentum equations:
r rr r
∂ ( ρV ) / ∂t + ∇ ⋅ ( ρVV ) = ρ g − ∇P + ∇ ⋅τ

• turbulence model:
r
∂ ( ρ k ) / ∂t + ∇ ⋅ ( ρVk ) = P − ρε + ∇ ⋅ [( µ + µt / σ k )∇k ]
r
∂ ( ρε ) / ∂t + ∇ ⋅ ( ρV ε ) = C1 Pε / k − C2 ρε 2 / k + ∇ ⋅ [( µ + µt / σ ε )∇ε ]

• mass conservation – transport of water volume fraction

r
∂rw / ∂t + ∇ ⋅ (rwV ) = 0
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Test case:

• Series-60 hull form

• numerical study
* of free surface effect
* attitude change effect
* and yaw angle effect
on ship hydrodynamics

• in all test cases, Froude number was 0.316

• The computational cases are listed below

 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Froude yaw angle sinkage trim heel

number
0.316 10 yes yes yes

0.316 10 no no no

0.316 10 double model

0.316 7.5 yes yes yes

0.316 5 yes yes yes

0.316 2.5 yes yes yes

 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Grid

• several three-dimensional grids were generated

• the total cell numbers about 400000
• based on the uncertainty study of resistance calculations

0
Z

-2
2

Y
-2
0
2 -2
4
6
X 8
10
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Wave profile (restrained condition, no attitude change):

ζ /L
0.04

measured, port
measured, star
0.03
computed, port
computed, star

0.02

0.01

-0.01
0 0.2 0.4 0.6 0.8 1 1.2

x/L
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Wave Pattern (restrained condition, no attitude change):

1 1

Level z Level z
19 0.0265 19 0.0265
17 0.0224 17 0.0224
15 0.0183 15 0.0183
13 0.0143 13 0.0143
11 0.0102 11 0.0102
0.5 0.5
9 0.0062 9 0.0062
7 0.0021 7 0.0021
Y/L

Y/L
5 -0.0020 5 -0.0020
3 -0.0060 3 -0.0060
1 -0.0101 1 -0.0101

0 0

0 0.5 1 1.5 0 0.5 1 1.5

X/L X/L

measurement calculation
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
1.0
Numerical simulation of yaw effect
0

-0.01

Velocity field at x/L = 0.1 -0.02

-0.03
U
1.00
0.96
0.91
0.87
-0.04

z
0.83
0.79

z
• measurement -0.05 0.74
0.70
-0.06
(hydrostatic attitude) -0.07
me 003  01 Feb 2005  title
-0.08
-0.05 0 0.05
y
1.0
0

-0.01

• calculation -0.02 U
1.00
-0.03 0.96
(hydrostatic attitude) 0.91
0.87
-0.04

z
0.83
0.79
-0.05 0.74
0.70
-0.06

-0.07

-0.08
-0.05 0 0.05
y
1.0
0

-0.01
0
6 -0.02 U
1 1.00
7 -0.03 0.96
3 0.91
9 0.87
• calculation 4
0
-0.04

z
0.83
0.79
-0.05 0.74
(trim, heel and sinkage corresponding -0.06
0.70

to the running attitude) -0.07

-0.08
-0.05 0 0.05
y
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Velocity field at x/L = 0.9

0 1.0
-0.02 U
0.99
0.92
-0.04 0.85

z
0.78
• measurement -0.06
0.72
0.65

(hydrostatic attitude) -0.08

0.58
0.51

Frame 003  01 Feb 2005  title

-0.05 0 0.05
y

0 1.0
-0.02 U
0.99
0.92
• calculation -0.04 0.85

z
0.78
0.72
(hydrostatic attitude) -0.06 0.65
0.58
0.51
-0.08

-0.05 0 0.05
y

0 1.0

• calculation -0.02 U
0.99
0.92
(trim, heel and sinkage corresponding -0.04 0.85

z
0.78
0.72
to the running attitude) -0.06 0.65
0.58
0.51
-0.08

-0.05 0 0.05
y
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Forces and moment coefficients (×1000)

case Ct Cs Cm
experiment 7.12 26.78 -1.83
changed 7.01 25.26 -1.81
attitude -1.54% -5.68% -1.09%
hydrostatic 6.91 24.98 -1.45
attitude -2.95% -6.72% -20.8%
double model 3.85 18.3 -1.08
(no free surface) -45.9% -31.7% -41.0%
 Validation of Numerical Model
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Forces and moments

8
7
Calculations: 6
5
4 Ct
• case 1: running attitude 3
2
• case 2: hydrostatic attitude 1
0
• case 3: no free surface Exp. [4] Case 3 Case 2 Case 1

30 -2

25
-1.5
20

15 Cs -1 Cm

10
-0.5
5

0 0
Exp. [4] Case 3 Case 2 Case 1 Exp. [4] Case 3 Case 2 Case 1
 Numerical simulation of yaw effect
Quixin Gao • Vladimir Shigunov • Dracos Vassalos

Ct

5
6
7
8

0
2
4
Resistance coefficient ×1000

6
measured
calculated

8
Yaw effect

10
12
yaw, degree
 Yaw effect
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Side force coefficient ×1000

Cs 30
measured
25
calculated
20
15
10
5 yaw, degree
0
0 2 4 6 8 10 12
 Yaw effect
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Coefficient of yaw moment ×1000

0 2 4 6 8 10 12
0
yaw, degree
-0.5

-1

-1.5
measured
Cm

-2 calculated
 Conclusion
Quixin Gao • Vladimir Shigunov • Dracos Vassalos
Numerical simulation of yaw effect

Based on comparison of computed and measured results,

the following conclusion could be drawn:

a. The general agreement between computational results and

measurement are satisfactory. However, increasing mesh
resolution and enhanced turbulence models are
recommended in future study to improve
numerical accuracy

b. The effects of attitude change on hydrodynamics are

significant in large yaw angles and should be taken
into account in the numerical prediction
of manoeuvring motion

c. Free surface effect cannot be neglected

for high Froude number as well
 Numerical simulation of yaw effect
Quixin Gao • Vladimir Shigunov • Dracos Vassalos

Thank you!
Questions
 Second-order wave forces and free-surface
elevation around a moored ship in steep
uni-directional and spread waves

J. Zang, K. Wang, R. Eatock Taylor and P. H. Taylor

Department of Engineering Science

University of Oxford

This work is supported under EU FP5 project REBASDO

 Outline

„ Motivation
„ Wave diffraction theory and numerical approach
„ Main features of numerical model
„ New wave
„ Validation
„ Results and discussions
„ Unidirectional waves – comparison with experiments

„ Spread sea (forces)

„ Moored vessel

„ Conclusions
„ Acknowledgement
 Motivation
This work is supported under EU FP5 project REBASDO
‘Reliability based design of FPSO system’

„ Diffraction theory
For an FPSO, with a ship-shaped geometry,
is first order wave diffraction theory accurate
enough?

„ Wave Conditions
• Uni-directional wave
• Directional spreading sea
 Objective

• Second order wave diffraction analysis

- sum frequency component
- difference frequency component
• Non-linear hydrodynamic loads on fixed or free floating vessel
• Non-linear free surface run up around the bow of the ship
• Effect of directional spreading in approaching waves on wave-
structure interaction
 Non-linear Wave Diffraction Theory

Stokes perturbation method

Φ = ε Φ (1) + ε 2 Φ ( 2 ) + ε 3 Φ ( 3) + LL

where ε is of order of wave steepness.

To second order, the non-linear diffracted potential is

−iωi t −i (ωi +ω j ) t
Φ( x, y, z, t ) = φ ( x, y, z )e
∑
i
(1)
+ ∑ ∑ φ ( x, y, z )e
+
ij
i i j
−i (ωi −ω j ) t
+ ∑ ∑ φij− ( x, y, z )e + ...
i j

Note sum and difference terms in freq

 Second order boundary value problem

∂2φs± ∂2φs± ∂2φs±

+ 2 + 2 =0 in fluid domain
∂x 2
∂y ∂z

∂φ7± ∂φi± on the surface of structure SB

=− + B±
∂n ∂n

∂φ j± for j=1,…,6. on SB
= nj
∂n
∂φs± on the sea bed
=0
∂z
∂φs± ω ± 2
− φs = F ± on the free surface
∂z g
 Extension to Directional Spreading Sea

Directional spreading incoming wave velocity potential can be expressed by

−iωi t −iωij+ t −iωij− t

φ ( x, y, z)e [φ ( x, y, z)e + φ ( x, y, z)e
N N N + −
Φ( x, y, z, t ) = ∑
i
(1)
+ ∑ ∑
ij ij ] + ...
i =1 i =1 j =1

Where φ is a space variable, can be given by

φi φ im( 1 ) ( x , y , z ) e i ( k i ( x cos( θ m ) + y sin( θ m )))

M
( x, y, z) =
(1 )
∑
m =1

± i ( ki ( x cos(θ m )+ y sin(θ m )) ± k j ( x cos(θ n )+ y sin(θ n )))

φij ( x, y, z ) = φ
M M ±
∑ ∑
ijmn ( x, y, z )e
m=1 n=1
 Computational demand

The computational demand for 2nd-order diffraction under directional

spreading sea will be
N × N × M × M pairs of calculation

Where N --- total components of frequency

M --- total components of wave direction

If we consider each input wave is the sum of all the waves from different
directions, but with same frequency.

We get N×N pairs

With new non-plane wave linear and 2nd-order calculations.

 The Main Features of Numerical Model
(DIFFRACT)

• Quadratic Boundary Element Method

• Two plane-symmetry used to reduce computational time
• Mesh can be automatically created by mesh generators
• Finite water depth
• Single or multiple fixed or floating structures
• Output 1st-order and 2nd-order hydrodynamic wave forces on structure
• Output 1st-order and 2nd-order free surface on and around structure
• Focussed wave group input for both unidirectional waves and
directional spreading sea
 New Wave – a wave group for modelling,
experiments and design
In a linear, random Gaussian sea-state, 5

the average shape of a extreme crest is 4

η (t ) = A ρ (t ) 3

∑ S (ω i ) cos(ω i t )
= A
1

∑ S (ω i ) 0
0 0.5 1 1.5 2 2.5 3

where ρ(t) is the unit auto-correlation 1.5

function and A is the amplitude of the 1

extreme event. 0.5

Ref Lindgren (1970), Boccotti (1983) -0.5

(also Tromans et al. 1991)

-1
-20 -10 0 10 20
 Validation

„ The free surface elevations have been very carefully compared

and agreed with semi-analytical solutions (Buldakov 2003) for a
bottom mounted circular cylinder

„ The forces and free surface elevations have also been validated
against the published papers (Kim & Yue 1990)

„ Some further validations have been made by comparing the forces

with those by Huang and Eatock Taylor (1997), Moubayed and
Williams (1995).

Ref Zang (Oxford report, 2003)

 FPSO Model Tests

Wave conditions: uni-directional wave group

spectrum T = 0.8 ~ 1.2s. A = 0.062m. h = 1.2m. (Ak = 0.25)

Structure: FPSO-like L = 1.124m. W = 0.325m. b = 0.125m.

waves head-on

FPSO experimental model FPSO Numerical mesh at

at Imperial College University of Oxford
Experimental data analysis
– based on accurate crest and trough focus alignment
Linear approximation to incoming wave
linear = (crest – trough)/2 + O(3rd)

0.12
Simulated for DIFFRACT (Num.)
0.1 Symmetric linear input (Expt.)

0.08

0.06

0.04

0.02

-0.02

-0.04

-0.06

-0.08
-6 -4 -2 0 2 4 6
Time t (s)

For 2nd order diffraction : 13 linear freq “ 91 pairs for QTFs

0.15

Surface elevation at Bow

0.1

0.05 Numerical results:

linear incoming wave
0 1st-order surface with diffraction
1st + 2nd order free surface
-0.05

-0.1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

0.15

Num. vs. expt

0.1

1st + 2nd order free surface

experimental non-linear free surface 0.05

-0.05

-0.1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Inclined uni-directional waves interaction with the ship
four different wave headings ( β = 0o, 15o, 30o, 45o)

Maximal and minimal non-linear free surface spatial profile for weather side
0.2
β =0
β =15
0.15 β =30
β =45
0.1
elevation (m)

0.05

-0.05

-0.1

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

x (m)
 Maximal and minimal non-linear free surface spatial profile for lee side
0.2
β =0
β =15
0.15 β =30
β =45
0.1
elevation (m)

0.05

-0.05

-0.1

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

x (m)
 Directional spread sea

A Gaussian directional spreading function

(θ −θ ) 2
−
1 2σ θ 2
D(θ ) = e (σθ = 15o and 30o)
2π σ θ

Surge QTF β 0=0o Pitch QTF β0=0o

0.4
uni-directional uni-directional
o Spread sea σθ=15o focusing at -0.4
0.35 Spread sea σ =15 focusing at -0.4
θ o 0.1 Spread sea σ =30o focusing at -0.4
Spread sea σ =30 focusing at -0.4 θ
θ o Spread sea σ =15o focusing at 0.0
0.3 Spread sea σ =15 focusing at 0.0 θ
θ o Spread sea σ =30o focusing at 0.0
Spread sea σ =30 focusing at 0.0 0.05 θ
θ
0.25
0
0.2

-0.05
0.15

-0.1
0.1

0.05 -0.15

0 -0.2
4 5 6 7 8 9 10 11 12 13 4 5 6 7 8 9 10 11 12 13
ω ω
 QTFs for main drift forces (b = 15o)
( The forces are normalised by πρgbA1A2 )

o
Surge QTF β 0=15 o Sway QTF β 0=15
0.4 0.2
uni-directional
Spread sea σθ=15 o uni-directional
o
0.35 Spread sea σ =15
Spread sea σ =30 o 0.18 θ o
θ Spread sea σ =30
θ

0.3
0.16

0.25
0.14
0.2
0.12
0.15

0.1
0.1

0.08
0.05

0 0.06
4 5 6 7 8 9 10 11 12 13 4 5 6 7 8 9 10 11 12 13
ω ω
Moored vessel analysis
 Quadratic transfer functions

-4 -3
x 10 Surge QTF x 10 Sway QTF
6 1.8
o
β=0 o
o 1.6 β=15
β=15 o
5 o β=30
β=30 β=45
o
o 1.4
β=45

4 1.2

1
3
0.8

2 0.6

0.4
1
0.2

0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 0.2 0.4 0.6 0.8 1 1.2 1.4
ω ω
 Conclusions

• 2nd-order wave diffraction is very significant locally

• wave-structure interaction gives strong local enhancement and radiation
of free waves at 2nd order sum freqs.
• long wave 2nd order difference - large set-up around bow

• Isolated wave group tests reveal physics

• no significant 3rd order effects for wave close to breaking
• 1st + 2nd order theory for free surface elevation compares remarkably
well to the experiments and analytical solution
• Uni-directional waves do not always lead to the maximum wave
forces on a ship-shaped FPSO
 Acknowledgement

The authors are very grateful to

„ Prof. C. Swan and Dr. R. Gibson

of Imperial College for access to their high quality experiments.

„ Dr. E. Buldakov
(currently UCL) for providing analytical solutions for the comparison.
( Ref. AOR (25), 301-320, 2003)

The authors would also like to thank the EU for the financial support
and the useful discussions with the partners during the review
meetings. We would also like to thank the Oxford Super Computing
Centre, University of Oxford for providing continuous computing
support for the calculation
 The effect of ship shape
and anemometer location
on wind speed
measurements obtained
from ships
B I Moat1, M J Yelland1, A F Molland2 and R W Pascal1
1) Southampton Oceanography Centre, UK
2) School of Engineering Sciences, Ship Science,
University of Southampton, UK

4th International Conference on Marine CFD, University of

Southampton, 30-31 March 2005.

NOTE: as of 1st May 2005 Southampton Oceanography Centre

becomes National Oceanography Centre, Southampton
• Wind speed measurements can be
severely biased by the presence of the
ship
• CFD can be used to predict/correct wind
speed measurements
 OUTLINE
• Background
• Description of the CFD code
• CFD code validation
• Results
– research ships (individual ships)
– tankers/bulk carriers/general cargo ships (generic
modelling approach)
– Container ships
• Research ship design
• Recommendations
• Conclusions
 Background
• Research ships limited coverage, but
measurements of high quality.
• Merchant ships routinely report
meteorological parameters at sea
surface (wind speed and direction)
• Data used in satellite validation, ocean
atmosphere modelling forcing and
climate research
 Background: impact of flow
distortion on climate studies
• 10 % error in mean wind speed
– 27 % bias in the momentum exchange
– 10 % bias in the heat exchange
 CFD code description
• Commercial RANS solver VECTIS
• Mesh generation
– Non-uniform Cartesian mesh
– (generate 500,000 cells/hour)
• 3-dimensional and isothermal
• MEAN FLOW ONLY (STEADY STATE)
• RNG turbulence model
• Simulations based on up to 600,000 cells
• All results normalised by the wind speed
profile at the measurement site
 VALIDATION

• Comparison to 2 previous wind tunnel

studies
– Martinuzzi and Tropea (1993)
– Minson et al. (1995)
• Comparison to in situ wind speed
measurements made from a ship
– Moat et al. (2005)
 Validation: channel flow over a
surface mounted cube
tunnel roof
accelerated
flow

decelerated flow H = cube

z/H height

Re=105

cube top

normalised wind speed

• Good comparison with RNG

 Validation: boundary layer flow
over a surface mounted cube

decelerated flow

H = cube
z/H height

accelerated
Re=4x104
flow

normalised wind speed

• Good comparison with RNG

 Validation: In situ wind speed
measurements from RRS Charles
Darwin
Measurements were
made using 6
anemometers.

Instruments were
located on a 6 m mast.

Only beam-on wind

speed data used.

Wind speed profile measured above a ‘block like’ ship.

 Validation: comparison with in
situ wind speed measurements

decelerated flow

accelerated flow
H = bridge to
sea level height
z/H
Re=1.3x107

normalised wind speed

• Agreement to within 4%
 RESULTS: research ships
• Project running since 1994
• Over 11 ships have been studied
– American, British, Canadian, French and
German
• Present results from well exposed
anemometers in the bow of 2 UK ships
– RRS Discovery
– RRS Charles Darwin
 Results: RRS Discovery

Typical
anemometer
location

length overall = 90 m

• Wind speed measurements are biased by

about 5 %
 Results: RRS Charles Darwin

Typical
anemometer
location

length overall = 70 m

• Wind speed measurements are biased by

about 10%
 Results: research ships
Wind speed bias (%)

bow RRS Discovery

port starboard RRS Charles Darwin

Relative wind direction

Locate anemometers as high as possible above

the platform, not in front
 RESULTS: tankers, bulk
carriers and general cargo
ships
Typical
anemometer
location

www.shipphotos.co.uk

Large number of ships. Cannot be studied individually.

The ships are large complex shapes
Results: A generic ship model

bow stern

• Ship dimensions from RINA publication

Significant ships (1990-93)
• Tankers/bulk carriers/general cargo ships can
be represented by a simple shape.
• Good agreement with WMO Publication 47
Results: A generic ship model
bridge
anemometers

bow stern

• Perform CFD simulations over the simple

geometry
• Bridge anemometers
• Flows directly over the bow
Wind tunnel: flow visulisation
mean flow direction

Standing vortex
in front of the
deck house
 Wind tunnel: flow visulisation
mean flow direction

Vortices produced
above the bridge top
Standing vortex
in front of the
deck house
• Decelerated region increases with distance
from the leading edge
 Wind tunnel: flow visulisation
mean flow direction

Less disturbance
with increase in
height
Vortices produced
above the bridge top
Standing vortex
in front of the
deck house

• Complex flow pattern

CFD: Airflow above the bridge
accelerated flow
3D simulation of the
airflow over the tanker.
(RNG turbulence closure)

decelerated flow
Tanker with recirculation.

Flow direction

Qualitatively, the numerical model reproduces the

general flow pattern quite well.
CFD: Airflow above the bridge
accelerated flow.
3D simulation of the
airflow over the tanker.
(RNG turbulence closure)

decelerated flow
Tanker with recirculation.

Flow direction

Qualitatively, the numerical model reproduces the

general flow pattern quite well.
 Normalised wind speed profile

deceleration and
z/H recirculation
H

bow stern

Normalised wind speed

• Wind speed accelerated by about 10 %

• Decelerated by up to 100 %
 Normalised wind speed profile

deceleration and
z/H recirculation
H

bow stern

Normalised wind speed

Region of high
velocity gradients
Accuracy of CFD generic ship
simulations
• Comparisons of simulations show
variations of:
– Mesh density (1 %)
– Turbulence model (2 %)
– Scaling the geometry (3 %)
– Wind speed profile (4 %)
• VECTIS agrees to 4 % or better with in
situ wind speed data
 Container ships
Anemometer
locations

www.shipphotos.co.uk

• More complex shape than a typical tanker

• Irregular container loading ???
 CFD: General flow pattern
decelerated

1.0 accelerated 1.0 accelerated

container ship
1.0

bow bridge

accelerated
1.0 1.0

accelerated decelerated
decelerated
(Moat et al. 2005)
 CFD: General flow pattern
decelerated

1.0 accelerated 1.0 accelerated

container ship
1.0

bow bridge

accelerated
1.0 1.0
typical tanker
accelerated decelerated
decelerated
(Moat et al. 2005)

• Bow influences the bridge flow

• Complex flow and the subject of future work
 Research ship design:
RRS James Cook

Anemometer location

First steel cut 26th January 2005

• CFD will be used to examine the flow

distortion at the bow mast site
 Recommendations: merchant
ships
Anemometer position
height, z (m)

Bridge
Depth of the
Bow recirculation region

Distance from leading edge, x (m)

• Anemometers will be less distorted in the bow

• Locate anemometers as high above the deck
as possible and above the leading edge
 FUTURE APPLICATION OF
RESULTS: MERCHANT SHIPS
• To predict the wind speed bias
– Ship type
– Ship length
– Anemometer position
• Parameters are now available (WMO-
47)
 CONCLUSIONS

• CFD is a valid research tool to examine

the mean airflow over ships
• Research ships: anemometers biased
by about 10% (highly dependent on
position)
• Streamlined superstructure for accurate
wind speed measurements
 CONCLUSIONS

• Tankers/bulk carriers/general cargo

ships: anemometers biased high by
10% and low by 100%
• Position anemometers as high as
possible above the deck
• Locate anemometers in the bows of the
ship
 FUTURE WORK

time = 3 sec

• How does the turbulence structure

change with ship shape ?
 FUTURE WORK

LES code GERRIS

Iso-surface of
wind speed
at 90% of the
inflow velocity
time = 3 sec

• Good representation of atmospheric

turbulence in the wake region of a ship
 Acknowledgements
Funding from Meteorological Service
of Canada and the Woods Hole
Oceanographic
Institution, USA.

Contact
ben.moat@soc.soton.ac.uk
www.soc.soton.ac.uk/JRD/MET/cfd_shipflow.php
Thrust and Torque Performance
Predictions using CFD
Mr Karl Randle, Mr Peter Bull, QinetiQ Ltd
A presentation to: RINA Marine CFD 2005
31 March 2005
 3
Introduction
• Overview of
hydrodynamics at
Haslar
• CFD methods for
propellers
• Standard propeller
blade
– Grid resolution study

• Skewed propeller blade

– Advance ratio study

• Conclusions
Test and Evaluation 4

• Physical model testing

– hull optimisation
– propulsor interaction
– validation of numerical models
– design risk management
– safety case validation

• Computational modelling
– hull optimisation codes

• Computational fluid dynamics

– panel, RANS, LES, DES methods
– propulsor design and flow analysis
 5

Concepts
• Cost reduction
– hull resistance reduced
• transom flaps and wedges
• novel hull forms e.g. trimaran
• partial air cushion support
catamaran (PACSCAT)

– propeller efficiency increased

• podded propulsion

– reduced propeller weight

• composite propeller
 6

Development
• Environment
– wave wake
• hull optimisation

– pollution
• cross-flow filtration
• waste management systems

– noise
• design of low noise propellers
• internal and external noise predictions
 7
Major Facilities
• Ocean Basin
– 122m x 61m x 5.5m
– Waves to 0.45m significant height
– Rotating arm

• Ship Tank
– 270m x 12m x 5.5m
– Carriage speeds up to 12.25m/s
 8

Major Facilities
• Quiet Water Tunnel
– Working section 0.76 m dia, 3.8 m long
– Working pressure up to 300 kPa
– Speed up to 15 m/s

• Cavitation Tunnel
– Working section 2.4 x 1.2 x 5.35 m
– Working pressure up to 160 kPa
– Speed up to 7.9 m/s

• Circulating Water Channel

– Working section 1.4 x 0.84 x 5 m
– Working pressure up to 100 kPa
– Speed up to 5.5 m/s
 9
Requirements for Propulsor Design
• Steady and unsteady forces
• Steady and unsteady flow
characteristics
– Tip vortices
– Blade wakes

• Interaction between propulsion

system and ship
– Shafts, brackets, rudder and hull

• Cavitation
– Onset and inception speed
– Extent, thickness and time variation
of cavitation

• Near and far field noise

 10

Computational Fluid Dynamics Methods

• Two distinct CFD methods
• In-house developed CFD package for propellers (MACH0)
– Fast and reliable
– Propeller and propulsion characteristics
– Extensively validated at model scale for range of propellers
– Used for full scale propeller design
– Limited range of models

• Commercial CFD package for ships and submarines

(CFX5)
– Flexible and robust
– Forces and moments acting on hull and appendages
– Validated over a range of Reynolds numbers for ship and submarine hulls
– Used for prediction of inflow conditions to the propulsion system
– Wide range of models
 11

Comparison Between CFD Methods

• Two different
propellers Standard
– Standard DTRC 4119 propeller
blade
– Skewed blade

• Open water thrust and

torque
• Pressure and skin
friction on the blade Skewed
blade
surfaces propeller

• Computational
requirements
 Grid Generation Method 12

• Based on in-house design

method
– Geometry definition based on Radial
standard design techniques grid
surface
– Pressure and suction sides of each
blade defined with upstream and down
stream extensions for the grid
– Skewed ‘H’ single block structured
grid

• Method can be extended to

Axial
include more advanced grid grid
generation capabilities surface

– Gridgen
– ICEM CFD
Grid Resolution Study 13

• Standard DTRC 4119

KT
blade J 0.5 0.833 1.1
MACH0 0.3120 0.1470 0.0280
• Comparison between CFX 500k 0.3080 0.1531 0.0311
thrust and torque CFX 1M - 0.1511 -
CFX 2M 0.3016 0.1512 0.0311
– MACH0, CFX5 and measured
Measured 0.285 0.146 0.034
values
Thrust
• Successive grid KQ
refinement in the axial, J 0.5 0.833 1.1
MACH0 0.0531 0.0287 0.0107
girthwise and radial CFX 500k 0.0507 0.0295 0.0112
directions for CFX5 CFX 1M - 0.0288 -
CFX 2M 0.0481 0.0288 0.0106
results Measured 0.0477 0.0280 0.0106
– 500K, 1M and 2M cells
Torque
– High aspect ratio cells near
blade tip
Pressure on the Blade Suction Surface 14

1M
cell
grid

500K 2M
cell cell
grid grid
Advance Ratio Study 15

• Skewed blade KT
• Comparison J 0.8 1.0 1.2 1.4
between thrust and MACH0 0.3753 0.2730 0.1683 0.0616
torque CFX 0.3714 0.2695 0.1665 0.0561
– MACH0, CFX5 and Measured 0.3534 0.2774 0.1812 0.0754
measured values
Thrust
• Range of advance KQ
ratios J 0.8 1.0 1.2 1.4
– 0.6 to1.5 in 0.1 MACH0 0.0802 0.0635 0.0450 0.0245
intervals CFX 0.0781 0.0630 0.0450 0.0227
– 1M cells Measured 0.0781 0.0660 0.0482 0.0263
– High aspect ratio cells
near blade tip Torque
– Highly skewed cells
near the blade tip
Comparison between Thrust and Torque 16

Coefficients
1

0.8

0.6
KT, 10KQ, Eta

0.4

0.2

0
Kt Ship Tank Kt CFX 10 Kq CFX
10 Kq Ship Tank Efficiency Ship Tank Efficiency CFX
10 Kq MACHO Efficiency MACHO Kt MACHO

-0.2
0.5 0.7 0.9 1.1 1.3 1.5
Advance ratio (J =Va/nD)
 17
Pressure on the Blade Surfaces

J=0.8 J=1.2

J=1.0
J=1.4
 Conclusions from CFD Comparison 18

• Preliminary results show

promising correlation
– Comparison between CFD
methods
– Comparison with measured data

• Detailed benchmark
required to obtain more
information
– Consistent turbulence model
– Computational requirements
– Ease of use
Stator/rotor interaction
• Range of different
propulsion systems
Integration with Propulsor Inflow 19

Predictions
• Complex inflow conditions from
shaft and brackets
– High Reynolds number flows at
full scale

• Transient flow
– Interaction with the bracket
wakes
– Interaction with the rudder and
hull

• Free surface gravity waves

• Cavitation Fully appended ship

• Ship motion
 20

Conclusions
• Integrated flow prediction
methods
– Viscous flows
– Free surface flows
– Propulsor flows

• Validated at model scale

– Range of relevant geometries
– Provides detailed information of the
flow characteristics

• Increasingly applied in the Complex propulsion system

design cycle
VOF-DYNAMIC MESH SIMULATIONS OF
UNSTEADY SHIP HYDRODYNAMICS

M. Visone, M. Eid
BLUE GROUP, Engineering & Design, Italy
P. Bertetti, R. Gandolfi
AZIMUT, Italy
C. Falletta, P.L. Ausonio
SYDAC Srl, Ship-Yacht Designers & Consultants, Italy
D. Paterna, R. Savino
DISIS, University of Naples, Italy

MARINE CFD 2005

AVENUE CAMPUS, SOUTHAMPTON UNIVERSITY, UK
30,31 March 2005
 Introduction

The computation of the complex hydrodynamic and aerodynamic flows around Planing
Hulls is extremely challenging and requires state-of-the-art numerical techniques and
computer technology.

Prediction of Hull performances with the aid of computational tools is particularly important
for two main reasons:
1) Avoid the uncertainties associated with the use of empirical equations, which are
only applicable to similar hull shapes (Savistky Method with some literature or
home-made correction procedure).
2) Reduce the prohibitive cost and time associated with extensive model testing,
which reliability is not assured, especially if the tests are performed on small
models, due to the Reynolds effect.

This paper presents a CFD approach for the calculation of Lift, Drag, Dynamic Trim and
Sinkage of Planing Hulls moving at steady speed through calm water, using a commercial
RANS CFD code.
The results of the computational simulation have been compared with the results of
different tests performed on two large models at the Brodarski Institute in Zagreb.
 Background

We started our work drawing from the interesting paper presented by Thornhill and others
in 2002 (24th Symposium on Naval Hydrodynamics - Japan), where the results of
calculations performed using the FLUENT CFD code on a planing hull, were compared with
the results obtained by the Savitsky approach and by experiments using a relatively small
model.

In the numerical calculation they didn’t model turbulence and they used a relatively low
resolution grid (of the order of 150,000 elements), calculating the frictional components
separately.

In general they found an over prediction of the net pressure on the hull surface, leading to
higher lift values, with respect to the experimental values. They indicated as possible
causes of that discrepancy an insufficient grid resolution, the lack of turbulence modeling
and the treatment of spray.

The approach here presented is an attempt to analyse these problems more in detail.
 CFD Computation

Computations have been performed taking into account both the water-air interface and the
flow turbulence.

‰ The air-water interface has been explicitly captured during computations, using a
Volume-of-Fluid algorithm.

‰ Turbulence has been modelled using the Reynolds Average Navier Stokes (RANS)
equations and the k-ω model, with suitable wall functions for near wall treatment.

‰ An hybrid tetrahedral/hexahedral mesh has been set up to calculate the flow field around
the hull in a mixed air-water environment.

‰ Two degrees of freedom have been taken into account: a vertical movement, and a
rotation around the axis normal to the symmetry plane.

‰ An “ad hoc” iterative computational procedure has been developed, based on a dynamic
mesh algorithm to compute, at steady speed, equilibrium position and orientation of the
hull.
 CFD Computation (cont’d)

Computational Grid
The hybrid grid has been coupled with “matching surface” techniques to join the different
grid blocks. This procedure allows smaller mesh sizes, faster modelling of complex
geometries and a faster dynamical adaption of the mesh, with a computational time cost
reduction.

Matching Surface

This procedure allows dynamical adaption of the

mesh without define a new mesh of Fluid Domain.

Approximately one million computational cells have been used for the numerical simulations.
 CFD Computation (cont’d)

Boundary Conditions and Degree of Freedom

Interface

Outlflow
The model and flow field are symmetrical Inlet Air

about the XZ plane at the model’s

centerline, so only half the width of full
domain is meshed. Surface Hull V

Inlet Water

z
y Symmetry

Z
ƒ Z : Vertical movement, corresponding to
the hull Sinkage
ϑ
G z
ƒ ϑ : Rotation (Trim angle) around the axis
normal to the symmetry plane, and passing
x
through the hull center of mass, that
determines the hull attitude.
 CFD Computation (cont’d)

Numerical Computation

All computations have been performed using the Fluent 6.1 CFD code

ƒ The unsteady RANS and VOF equations have been solved using the sequential algorithm
available in Fluent, based on the SIMPLE method by Patankar.

ƒ A time step of 0.01 s has been chosen in the time implicit algorithm, to avoid numerical
instabilities associated with the highly nonlinear equations.

ƒ Each run of the code has been performed on a four Pentium processors Linux cluster and
required some days to reach a steady state solution.
 Test Case

Two different hulls have been considered (Hull-A and Hull-B).

Hull-A : 15 (m) overall length model with full appendages an spray rails,
Hull-B : 18 (m) overall length with spray rails only.

The towing tests were performed on two models with

scale 1 : 3.8 (Model A) and 1 : 6.0 (Model B).
Both models were towed through the shaft line thrust
bearing position, with the same inclination of the
shaft.

One run for each model, has

been considered.

Hull-A : 15 (m) overall length. Hull-B : 18 (m) overall length.

V = 15 (m/s) V = 17.5 (m/s)
 Hull Equilibrium Position Procedure

An iterative methodology has been developed to determine the hull equilibrium position,
corresponding to constant values of the hull speed.

‰ The first step consists of solving the flowfield in an initial position (obtained using the
Savitsky method) of the hull, and then to evaluate the corresponding forces and moments.

‰ Two other steps follow, where either the hull sinkage or the hull trim angle are arbitrarily
modified. In these new positions, new CFD computations are performed.

‰ Three sets of forces data are available, that can be used to predict the equilibrium
position, assuming a linear dependence of forces on position.

‰ The new computations do not require an externally generated grid. A user defined
function has been developed to change the hull position and orientation, which acts in
combination with the Fluent capability to dynamically adjust meshes when boundaries are
modified.
 Hull Equilibrium Position Procedure (cont’d)

Static Equilibrium Conditions

Taylor Series Expansion (at first term) of Lift, Drag
and Pitching Moment around the (unknown)
equilibrium values. Leq = W – Teq sin(α)

 ∆L   ∆L  
Deq = Teq cos(α) (2)
∆L =L − Leq =   ⋅( ϑ −ϑeq ) +   ⋅(Z −Zeq ) 
 ∆ ϑ z  ∆Z ϑ  M=Teq sin(α) XTG –Teq cos(α) ZTG
 ∆D   ∆D   L
= − =
∆D D Deq  ⋅
 ( ϑ −ϑ +
eq ) 
⋅ −
 (Z Zeq )  (1) XTG

 ∆ ϑ z  ∆Z ϑ 
z
M D
 ∆M   ∆M   v G ϑ
= −
∆M M Meq  =  ⋅( ϑ −ϑeq ) +   ⋅(Z −Zeq ) Teq ZTG
x O

ϑ
 ∆ z  ∆Z ϑ  α

W
• XTG : Distance, in X direction, between T and CG (m)
• ZTG : Distance, in Z direction, between T and CG (m)
• α : Thrust angle respect to X axis (deg)

Equations (1), together with (2), constitute a linear algebraic system of three equations in
the three unknowns: Z (Sinkage), ϑ (Trim) and Teq (Thrust).

The derivatives appearing in the system may be computed by finite differences, using the
results of the previous CFD calculation.
 Hull Equilibrium Position Procedure (cont’d)

Flow Diagram Describing the Procedure.

In order to reduce the iterations number, the
(i ) (i )
Savitsky Method
θ ,Z initial position is that obtained using the
approximate Savitsky Method.

CFD

L(i ), M (i ) , D(i ) , L(zi ), M z( i ) , Dz( i ) , L(ϑi ), Mϑ( i ) ,Dϑ( i )

Linear Algebric System

θ (i +1) , Z (i +1) ,T ( i +1 )
eq

CFD

L(i +1) , M (i +1) , D (i +1)

Equilibrium?

NO Yes
End
 Results

Computational Results for Hulls in Equilibrium Position : Static Pressure

(Pa)
(Pa)
Hull-A : 15 (m) overall length. Hull-B : 18 (m) overall length.
V = 15 (m/s) V = 17.5 (m/s)
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Velocity Magnitude

(m/s)
Hull-A : 15 (m) overall length. V = 15 (m/s) Hull-B : 18 (m) overall length. V = 17.5 (m/s)
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Path Lines

water
air

Hull-B : 18 (m) overall length. V = 17.5 (m/s)

Proposed Method allows to study, at the same time, hydrodynamic and

aerodynamic flows around a motor boats.
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Path Lines

Hull-A : 15 (m) overall length. water

V = 15 (m/s)
air

Hull-B : 18 (m) overall length. V = 17.5 (m/s)

Proposed Method allows to study, at the same time, the complex

hydrodynamic and aerodynamic flows around a motor boats.
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Free Surface

Hull-A : 15 (m) overall length.

V = 15 (m/s)

Hull-B : 18 (m) overall length.

V = 17.5 (m/s)
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Free Surface

Hull-A : 15 (m) overall length.

V = 15 (m/s)

Z (m)

Hull-B : 18 (m) overall length.

V = 17.5 (m/s)

Z (m)
 Results (cont’d)

Computational Results for Hulls in Equilibrium Position : Free Surface

Z (m)

Hull-A : 15 (m) overall length. V = 15 (m/s)

 Results (cont’d)

Computational Results : Free Surface (Animation)

Hull-B : 18 (m) overall length.

V = 17.5 (m/s)

Z (m)

Hull-A : 15 (m) overall length.

V = 15 (m/s)

Free Surface Developments during the Iterative Calculation Process.

 Results (cont’d)

Computational and Experimental Results Comparison: Free Surface

(m)

Hull-A : 15 (m) overall length. V = 15 (m/s)

 Results (cont’d)

Computational and Experimental Results Comparison: Free Surface

(m)

Hull-A : 15 (m) overall length. V = 15 (m/s)

 Results (cont’d)

Computational and Experimental Results Comparison: Sinkage, Trim, and Drag (Thrust)

θ θexp ZG ZGexp Both the results present a conformable behaviour

Model
(deg) (deg) (m) (m) between experimental and numerical data.
Hull-A 4.5 4.2 2.44 2.16
‰ The computed Drag values seem to
Hull-B 3.1 3.8 2.03 1.97 underestimate the experimental ones in both
cases.
Model (θ – θexp) / θexp (ZG – ZGexp) / ZGexp
‰ A lower computed Drag is, in turn,
Hull-A +7% + 13 % associated with the lower values of the
Hull-B - 18 % +3% computed sinkage Z indipendently from the
values of the trim angle ϑ.
Relative Error

‰ About the Hull-A, the effect of the higher trim

angle (compared with the experimental value)
Model L (N) W (N) D (N) Dexp (N) on the computed drag is partially compensated
Hull-A 106 550 101 500 16 620 17 150 by the effect of the higher CoG rise, so that the
overall effect is low (differences on drag are
Hull-B 121 000 126 050 17 200 19 340
about 3%).
Model (L – W)/W (D – Dexp)/Dexp ‰ The computations on the Hull-B show that the
Hull-A +5% -3% effects of the trim angle and the boat CoG rise
act in the same direction to reduce the drag
Hull-B -4% - 11 %
with respect to the experimental data.
Relative Error
 Future Developments

Further analyses are foreseen to try to explain the observed discrepancies.

‰ From an experimental point of view:

ƒ Further model and/or full scale tests with pressure distribution measurements to obtain a
more detailed comparison with the computed values.

‰ From a numerical point of view:

ƒ Sensitivity study of
• Mesh size and quality.
• Turbulence models.

An additional area of action should be the attempt to reduce computational time and costs.
 Conclusions

‰ CFD computations have been performed to determine the forces and moments acting on
the surface of a planing boat, using a numerical procedure that combines the RANS and
VOF approaches.

‰ A global iteration procedure has been developed to determine the Hull Sinkage and Attitude
for a given speed condition.

‰ The numerical results have been compared with the available experimental data, and a fair
agreement has been found.

‰ It is necessary to find a suitable trade off about mesh resolution and surface
capturing methodology to reduce the computational time and costs to acceptable
levels.

‰ Although further investigation seems necessary from both the numerical and experimental
point of view, the combined RANS-VOF CFD numerical simulation appears to be very
promising in the planing hull field.