CFD in Ship Hydrodynamics - Results of the Gothenburg 2010 Workshop

International Conference on Computational Methods in Marine Engineering

MARINE 2011
L.Eça, E. Oñate, J. García, T. Kvamsdal and P. Bergan (Eds)

CFD IN SHIP HYDRODYNAMICS –

RESULTS OF THE GOTHENBURG 2010 WORKSHOP

LARS LARSSON*, FREDERICK STERN† AND MICHEL VISONNEAU+

*
Chalmers University of Technology
Department of Shipping and Marine Technology
SE-41296 Gothenburg, Sweden
Email: [email protected], web page: http://www.chalmers.se/smt/EN
†
IIHR—Hydroscience & Engineering
The University of Iowa
100 C. Maxwell Stanley Hydraulics Laboratory
Iowa City, Iowa 52242-1585, USA
Email: [email protected], web page: http://www.iihr.uiowa.edu
+
ECN/CNRS
Fluid Mechanics Laboratory - UMR6598
Centrale Nantes BP 92101
44321 Nantes Cedex 3 – France
Email: [email protected], web page: http://www.ec-nantes.fr

Key words: Computational fluid dynamics, hydrodynamics, validation, assessment,

workshop

Summary. The Gothenburg 2010 Workshop on Numerical Hydrodynamics gathered 33

groups with computations for one or more of 18 test cases. All results were collected and
discussed at a meeting in Gothenburg in December 2010. In the present paper some
representative examples from the workshop are presented. The complete results are found in
the workshop Proceedings.

1 INTRODUCTION
In 1980 an international workshop on the numerical prediction of ship viscous flow was
held in Gothenburg1. The purpose was to assess the state-of-the-art and to find directions for
the future developments in the field. Participants in the workshop had been invited long
before and had delivered results for two well specified test cases to the organizers. Detailed
information on the features of each participating method had also been submitted and
compiled in a table. By comparing the computed results on the one hand, and the details of the
methods on the other, the most promising approaches could be sorted out.
Now, more than 30 years have passed since this first workshop and the event has been
repeated a number of times. In 1990 the second workshop was held, again in Gothenburg2.
While practically all methods participating in the 1980 workshop had been of the boundary
layer type, now all but one were of the RANS type. A huge improvement in the prediction of
the flow around the stern was noted. The workshop idea was picked up in Japan in 1994 and

17
 Lars Larsson, Frederick Stern and Michel Visonneau

the third workshop was held in Tokyo in 19943. Notable from this workshop is that free-
surface capabilities had become available in many of the methods. The fourth workshop in the
series was held in Gothenburg in 20004,5. Now, three modern hull forms were introduced as
test cases, and these hulls have been kept ever since. At this time formal verification and
validation (V&V) procedures were introduced. While in the previous workshops the emphasis
had been on the wake and waves for a towed hull, self-propulsion was introduced in 2000.
This was kept in the fifth workshop in Tokyo in 20056, where some sea-keeping and
manoeuvring cases were introduced as well. Even though the same three hulls were used this
increased the number of test cases significantly. A further step in this direction was taken at
the most recent workshop, held in Gothenburg in December of 2010. This workshop is the
topic of the present paper.

2 HULLS
The three hulls used in the workshop were:
1) The KVLCC2, a second variant of a Korean VLCC
2) The KCS, a Korean container ship
3) The DTMB 5415, a US combatant
The KVLCC2 was designed at the Korea Research Institute for Ships and Ocean
Engineering (now MOERI) around 1997 to be used as a test case for CFD predictions.
Extensive towing tank tests were carried out, providing data for resistance, sinkage and trim,
wave pattern and nominal wake at several cross-planes near the stern7,8,9. Mean velocity and
turbulence data were obtained by Postech10 in a wind tunnel. At the CFD Tokyo Workshop in
20056 there was a slight modification of the stern contour of this ship and it was therefore
renamed as KVLCC2M. The modification is explained in Hino6. In the present workshop the
original design was used.
Also the KCS was designed by MOERI for the same purpose as the KVLCC2, and similar
tests were carried out for this hull7,8,9. Self-propulsion tests were carried out at the Ship
Research Institute (now NMRI) in Tokyo and are reported in Hino6. Data for pitch, heave, and
added resistance are also available from Force/DMI measurements11.
Model 5415 was conceived as a preliminary design for a Navy surface combatant around
1980. The hull geometry includes both a sonar dome and a transom stern. Propulsion is
provided through twin open-water propellers driven by shafts supported by struts.
The model test data for the 5415 includes:
 Local flow measurements (Mean velocity and cross flow vectors)12.
 PIV-measured nominal wake in regular head waves (Mean velocity, turbulent
kinetic energy, and Reynolds stresses)13.
 Resistance, sinkage, trim, and wave profiles12.
 Wave diffraction (Waves, 1st harmonic amplitude of mean velocities, turbulent
kinetic energy, and Reynolds stresses)13, 6.
 Roll decay (Motion, free surface, mean velocities)14.
Side views of the three hulls are seen in Figure 1 and the main particulars are given in
Table 1. No full scale ships exist.

18
 Lars Larsson, Frederick Stern and Michel Visonneau

Figure 1: The three ships used in the workshop (upper : KVLCC2; middle : KCS; bottom : 5415)

Main Particulars (Full Scale) KVLCC2 KCS DTMB 5415

Length between perpendiculars LPP (m) 320.0 230.0 142.0
Maximum beam of waterline B (m) 58.0 32.2 19.06
Draft T (m) 20.8 10.8 6.15
Displacement ∆ (m3) 312622 52030 8424.4
Wetted area w/o rudder SW (m2) 27194 9424 2972.6
Wetted surface area of rudder SR (m2) 273.3 115.0 30.8
Block coefficient (CB) ∆/(LPP·B·T) 0.8098 0.6505 0.507
Service Speed U (knots) 15.5 24.0 18.0, 30.0
speed Froude number Fr 0.142 0.26 0.248, 0.413
Table 1 : Main particulars of the three ships

3 TEST CASES
Several types of computations were requested, namely:
1) Local flow at fixed condition, either zero sinkage and trim (denoted FX0) or
dynamic sinkage and trim (FXστ)
2) Resistance, sinkage and trim either at FX0 or at heave- and pitch-free condition
(FRzθ)
3) Self-propulsion at FX0 or FRzθ
4) Heave and pitch in waves either at FRzθ or with free surge (FRxzθ)
5) Forward speed diffraction at FXστ
6) Free roll decay at FXστ and free to roll (FRυ)
Note that maneuvering was not included, since it was the topic of the recently held
SIMMAN workshop in the spring of 200815.
All test cases for the three hulls are listed in Table 2. The measurements were taken at the
organizations within brackets. See the references above. There are altogether 18 cases and the
participants were free to select which cases to compute.

4 PARTICIPANTS AND METHODS

The workshop participants are listed in Table 3, together with the main features of their
methods. In the first column the acronym of the participating group is given. This is used

19
 Lars Larsson, Frederick Stern and Michel Visonneau

Case
Hull Attitude Measured quantity
Number
1.1a FX0 Mean velocity, Reynolds stresses (Postech WT)
1.1b FX0 Wave pattern (MOERI)
1.2a FX0 Resistance (MOERI)
1.2b KVLCC2 FRzθ Resistance, sinkage and trim (MOERI)
1.4a FRzθ Pitch, heave, added resistance (INSEAN)
1.4b FRzθ Pitch, heave, added resistance (NTNU)
1.4c FRxzθ Surge, Pitch, heave, added resistance (Osaka Univ)
2.1 FX0 Wave pattern, mean velocities (MOERI)
2.2a FX0 Resistance (MOERI)
2.2b FRzθ Resistance, sinkage and trim (MOERI)
KCS Self propulsion at ship point (thrust, torque, force
2.3a FX0
balance or RPM, mean velocity), local flow (NMRI)
Self propulsion at model point (thrust, torque,
2.3b FRzθ
force balance or RPM), sinkage and trim (FORCE)
2.4 FRzθ Pitch, heave, added resistance (FORCE)
3.1a FXστ Mean velocity, resistance, wave pattern (INSEAN)
Mean velocity, resistance, wave pattern,
3.1b FXστ
Reynolds stresses (IIHR)
DTMB 5415
3.2 FRzθ Resistance, sinkage and trim (INSEAN)
3.5 FXστ Wave diffraction, Mean velocity (IIHR)
3.6 FRυ Roll decay (IIHR)

Table 2 : Test cases

in combination with the code name of column three to identify each submission. The cases
computed are given in column two. In the remaining columns the features of each method are
given.
The majority of methods use two-equation turbulence models, k- or k-. There are also some
one-equation models, Spalart-Allmaras or Menter. The anisotropic models are either of the
algebraic stress or Reynolds stress type. Note that there are also some LES/DES methods and
even a DNS method. Most of the participants use no-slip wall boundary conditions, but there
are also several methods with wall functions, both with and without pressure gradient
corrections. The Volume of Fluid (VOF) technique is the most popular one for the free-
surface modeling, followed by the level set methods. There are only three entries with surface
tracking. The propeller is represented either as an actual rotating propeller or through a body
force approximation. Simulations were performed using both finite difference and finite
volume codes, but there was no finite element method. 2nd or 3rd order accurate schemes were
used and limited studies used 4th order schemes. Most methods are pressure based, but there
are also several solving the equations directly or with an artificial compressibility.
The grids used were either single- or multi-block structured (butt-joined or overlapping) or
unstructured. There was a huge variation in grid density, from 0.3M to 300M. Most of the

20
 Table 3(1) : Workshop participants and methods

Cases Turbulence Wall Discretization Grid Velocity

Organization Code Free Surface Propeller
Submitted (incl. non-RANS) Model Type Order Type Pressure
Standard
CD-Adapco 2.2a, 2.2b STAR-CCM+ N VOF - FV 2 U PR
k-ε
CEHINAV PR
3.1a STAR-CCM+ k-ω SST N VOF - FV 1 MU
TU Madrid SIMPLE
k-ω SST
Chalmers 1.1a SHIPFLOW4.3 N - - FV 2 S A
EASM
2.1, 2.2a, 2.2b, k-ω SST PR
CSSRC FLUENT 6.3 N VOF Actual FV 2 MS
2.3a, 3.1a, 3.2 RNG k-ε SIMPLE
N PR
CTO 2.3a STAR-CCM+ k-ε VOF Actual FV 2 U
WO SIMPLE
1.1b, 1.4a, 1.4b, Body PR
ECN/CNRS ISISCFD k-ω, EASM WO VOF FV 2 U
3.6 force SIMPLE
1.1a, 1.1b, 1.4a,
Wilcox k-ω N Nonlin. Body
ECN/HOE 1.4b, 2.1, 3.1a, ICARE FD 2 S DM
k-ω SST WO track force
3.1b, 3.5, 3.6
SHIPFLOW-
FLOWTECH 2.1 k-ω SST N VOF - FV 2 MS A
VOF-4.3
FOI 3.1a OF LES WO VOF - FV 2 U PR
PR
FORCE 2.4 CFDShip-Iowa k-ω SST N Level set - FD 2 OS
PISO
GL&UDE 1.4a, 1.4b, 1.4c, Comet N PR
k-ε VOF - FV Mixed U
Univ. Duisburg 2.2a, 2.2b, 2.4, 3.6 OpenFOAM WO SIMPLE
HSVA 1.1a, 1.2a FreSCo+ 2E k-ω N VOF - FV 3 U PR
WISDAM- Baldwin-Lomax Density PR
IHI/Univ. Tokyo 2.4 N - FV 3 OS
UTokyo and DSGS function ρ MAC
Hybrid k-ε/k-ω
1.1a, 1.4a, 1.4c, Actual PR
CFDShip-Iowa based DES WO
IIHR 2.1, 2.3a, 2.3b, 2.4, Level set Body FD 2~4 S, OS Fractional
V4, V4.5, V6 Hybrid ARS N
3.1a, 3.1b, 3.5, 3.6 force step
based DES
Body
IIHR-SJTU 2.1, 2.3a FLUENT12.0.16 Realizable k-ε N, W VOF FV 3 MS PR
force
IST 1.1a PARNASSOS k-ω SST N - - FD 2 S DM
Kyushu
1.4b RIAM-CMEN DNS N THINC - FD 3 S PR
University

21
 Table 3(2) : Workshop participants and methods

Cases Turb. model Wall Discretization Grid Velocity

Organization Code Free Surface Propeller
Submitted (incl. non-RANS) Model Type Order Type Pressure
1.1a, 1.1b, 1.2a,
Realizable PR
MARIC 2.1, 2.2a, 2.3a, FLUENT N VOF Actual FV 2 MS
k-ε SIMPLE
3.1a, 3.1b, 3.2
Free-surface
MARIN 2.1, 2.3a, 3.1a, 3.2 PARNASSOS 1E Menter N - FD 2 MS DM
fitting
1.1b, 1.2a, 1.2b,
Body PR
MOERI 2.1, 2.2a, 2.2b, WAVIS Realizable k-ε WO Level set FV 3 MS
force SIMPLEC
2.3a, 2.3b
NavyFOAM 1.1a, 3.1a, 3.1b, MS
NavyFOAM Wilcox k-ω WO VOF - FV 2, 3 PR
(NSWC/P S ARL) 3.2 MU
1.1a, 2.1, 2.3a, 1E Modified Body
NMRI SURF N Level set FV 2 S, U A
3.1b, 3.5 Spalart-Allmaras force
NSWC-PC CFDShip-Iowa DES
3.2 N Level set - FD 4 S, OS PR
ARL V4.5 Hybrid k-ε/k-ω
PR
NTNU 1.1a, 1.1b, 1.2a FLUENT k-ω SST N VOF - FV 2 MU
SIMPLE
SNUTT 2.1, 2.3a FLUENT6.3 k-ε W VOF Actual FV 2 MU DM
Southampton Body
2.1, 2.2b, 2.3a CFX 12 k-ω SST W VOF FV 2 MS DM
Univ. QinetiQ force
Body
SSPA 2.3a, 2.3b SHIPFLOW4.3 EASM N - FV 2 OS A
force
SSRC 2.1, 2.3a, 3.1b, 3.5,
FLUENT12.1 k-ω SST N VOF Actual FV 2 MS PR
Univ. Strathclyde 3.6
SVA Potsdam 2.1, 2.2a ANSYS-CFX12 k-ω SST N VOF - FV 2 MU DM
PR
TUHH 1.1b, 2.4 FreSCo+ k-ω SST N VOF - FV 3 U
SIMPLE
Fully
TUHH N coupled w-p,
2.1, 2.2a, 2.3a, CFX12.1 k-ω SST VOF Actual FV 2 MS
ANSYS WO SIMPLER
p-equation
PR
Univ. Genova 3.1a, 3.2 StarCCM+ Realizable k-ε N VOF - FV 2 U
SIMPLE
1.1a, 1.1b, 1.2a, FV MS
VTT FINFLO k-ω SST N Nonlin. track - 3 A, PR
2.1 FD OS
A-Artificial compressibility; DM-Direct method; FD-Finite difference; FV-Finite volume; MS-Multiblock structured; MU-Multiblock unstructured; N-No slip; OS-
Overlapping structure; PR-Pressure correction; S-Single block structured; U-Unstructured; W/WO-Wall functions with/without pressure gradient correction

22
 Lars Larsson, Frederick Stern and Michel Visonneau

simulations were performed with grids in the range 1-10M, but some resistance and self-
propulsion cases had grids up to 24M cells. There were also a few submissions from IIHR
with grids around 300M.

5 SELECTED RESULTS
The Workshop Proceedings16 contain 380 pages of tables and diagrams displaying the
submitted results. Also included in the Proceedings are three evaluation reports written by the
organizers. These reports contain altogether 125 pages of discussions. With the space
limitation of the present paper only a few examples from the Proceedings can be given. The
examples will be presented by subject (c.f. ―Types of computations‖ in Section 3): resistance,
self-propulsion, wave pattern, local flow, heave and pitch, and roll decay. Unfortunately,
sinkage and trim, wave diffraction and the very extensive discussions on Verification and
Verification (V&V) will have to be left out.

5.1 Resistance
In Table 4 a statistical analysis of all computed total resistance coefficients is presented.
While the case and Froude number are presented in columns 1-3, column 4 gives the mean
comparison error Emean in per cent of the measured data value, D. According to the sign
convention of the Workshop Emean is defined as D – Smean, where Smean is the mean of all
simulated values for the particular case. The standard deviation, , is given in column 5 in per
cent of the data value, and in column 6 the estimated data uncertainty is presented. Finally, in
the last column the number of entries for the case is seen. Values within brackets are from the
2005 Workshop6.
Comparing the 2010 and 2005 results a substantial reduction in the standard deviation
(%D) for the towed KVLCC2 and KCS cases is noted, from 6.2 to 1.3 and from 4.2 to 1.2
respectively in the fixed condition. Also, |Emean| for these conditions is well below 2%D,
which indicates that all predictions for this condition are quite accurate, although still not
within the experimental accuracy. There is only one submission for the free KVLCC2
condition and |Emean| for all Froude numbers is of the same size 2.1%D. The free KCS
condition has several submissions and very small comparison errors (0.2%D) and standard
deviations, around 1%D for both.
The self-propelled KCS has standard deviations around 3%D and the comparison error is
very small for the fixed case. However, for the free case |Emean| is quite high: 7.2%D. All three
submissions under predict the resistance significantly. It should be noted that the fixed KCS
in self-propulsion is the only case for which the standard deviation has increased compared to
2005.
5415 with sinkage and trim fixed to the dynamic values have comparison errors below
3%D and standard deviations around 4%D. In view of the fact that the only difference
between 3.1a and 3.1b is the Reynolds number (apart from a very small difference in sinkage
and trim), the difference is large, but the statistical basis is too small for a comparison. For the
free 5415 in 3.2 both the mean error and the standard deviation seem to depend strongly on
the Froude number. The best results are obtained at Fr = 0.28, where the water just clears the
transom. For this condition the mean error is practically zero and the standard deviation

23
 Lars Larsson, Frederick Stern and Michel Visonneau

among the 6 submissions is 2.1%D.

Table 4 shows the statistics for all cases, and indicates the accuracy obtainable for each
case. Even more interesting is however the information found on the last line: the mean error
and the mean standard deviation (weighted by number of entries) for all cases. The mean error
for all computed cases is practically zero; only - 0.1%D, while the mean standard deviation is
2.1%D; a surprisingly small value. In the 2005 Workshop the mean error of all 40
submissions was in fact equally small: 0.1%D, while the mean standard deviation was 4.7%D.
While the distribution between ―simple‖ and ―difficult‖ cases is not the same in the two
Workshops, it seems safe to conclude that the scatter has been reduced considerably. In fact,
even the largest standard deviation in the present computations (Cases 3.1b and 3.2) is smaller
than the mean standard deviation in 2005.

 UD No. of
Hull Case No. Fr Emean %D
%D %D Entries
1.2a (fixed) 0.1423 -1.7 (0.0) 1.3 (6.2) 1.0 (0.7) 5 (13)
KVLCC2
1.2b (free) 0.10~0.15 -2.1 - 1.0 1
2.2a (fixed) 0.26 -1.3 (-1.1) 1.2 (4.2) 1.0 8 (11)
2.2b (free) 0.11~0.28 -0.2 1.2 1.0 27
KCS
2.3a (fixed, prop.) 0.26 -0.3* (-0.9) 3.1 (1.0) - 14* (4)
2.3b (free, prop.) 0.26 7.2 3.3 - 3
3.1a (fixed s&t) 0.28 2.5 (1.6) 3.8 (5.3) 0.6 (2.2) 5 (11)
3.1b (fixed s&t) 0.28 -2.6 4.4 0.6 5
5415 0.138 -2.8 4.4 1.3 5
3.2 (free) 0.28 0.1 (-1.9) 2.1 (-) 0.6 (2.2) 6 (1)
0.41 4.3 1.4 0.6 5
Mean of all cases Emean = -0.1%D mean = 2.1%D 89 (40)

Table 4 : Resistance statistics, all cases (*: Results from MARIC with a hub cap are excluded)

Figure 2 shows the comparison errors for all submitted cases versus grid size. For each
submission the turbulence model is denoted by a symbol. It is seen that about 90% of all
computations are made with grids smaller than 10M cells. The scatter within this range seems
to be significantly larger than for the larger grids. However, this is mainly caused by the large
scatter of the self-propulsion submissions (represented by filled symbols), so if these are
excluded, and only towed resistance is considered, there is no error decrease above 3M grid
points. All points seem to be within approximately +- 4%D. Not even the very large grid at
300M cells (moved into the figure and marked) shows any significant improvement; it is
slightly below 3%D. However, below 3M cells the maximum errors increase to about 8 %D.
There is a large number of entries for the 2-equation models and the results are generally
good. For the others there are rather few entries. The relatively poor result for the more
advanced methods is a surprise. However, EASM and RS, ARS suffer from one bad point for
a very coarse grid and two bad points computed for the self-propulsion cases, which may be
more difficult than the towed cases. The three very good results for the Menter model were
obtained with the same code and user. It would be interesting to investigate the performance

24
 Lars Larsson, Frederick Stern and Michel Visonneau

of each model in different ranges of grid density, but that has not been done so far.
(% D)
15.0
1E: Spalart Allmaras
1E: Menter
10.0 2E: k-
2E: k-
EASM
5.0
Comparison error E

RS, ARS
DES

0.0
(300M)

-5.0

-10.0

-15.0
0 4 8 12 16 20 24
Grid points (M)
Figure 2: Comparison error of all resistance submissions vs. grid size (turbulence model parameter).
Filled symbols represent propulsion cases

5.2 Self-propulsion
Self-propulsion results were requested only for the KCS hull and only at one Froude
number: 0.26. In Case 2.3a the hull was kept fixed in the zero speed attitude, while in 2.3b the
hull was free to sink and trim. Experimental data are available from NMRI for a 7.3 m hull in
2.3a and from FORCE for a 4.4 m hull in 2.3b. The NMRI hull was without a rudder, while a
rudder was fitted to the FORCE hull.
In 2.3a computations were requested for the model at the ship point, i.e. the hull was towed
to account for the larger skin friction at model scale compared to full scale. This force, the
skin friction correction, SFC, was pre-computed and was the same as in the measurements. In
the experiments the thrust T, was adjusted by varying the rpm, n, such that T = RT(SP)-SFC,
where RT(SP) is the resistance in self-propulsion. Most of the participants did the simulations
in this way, i.e. the force balancing was automatically achieved by the flow code. An
alternative was to avoid the balancing and use the measured rpm in the simulation. In the first
case the achieved n was requested, while in the second case the resulting towing force RT(SP)-
T was to be reported. In 2.3b computations were carried out for the model point, so no towing
force was applied, but the balancing was carried out in the same way as in 2.3a.
The dependence on the grid density of the thrust coefficient, KT, torque coefficient KQ,
achieved n for force balancing and towing force RT(SP)-T for given n are plotted in the
Proceedings, but cannot be presented here due to the space limitation. There is a clear
difference in scatter between the three predictions in the range 10-24 M cells and those below
10M. For KT, KQ and n the maximum scatter in the upper range is around +- 7%, 5% and 2%,
respectively, while in the lower range it is within 19%, 18% and 6%. For the towing force

25
 Lars Larsson, Frederick Stern and Michel Visonneau

RT(SP)-T there are very few entries and the largest error is for an 11.5M grid. All quantities
but n have considerably larger errors than resistance.
Of more interest is perhaps the difference between the actual and modeled propellers and
between the force-balanced and fixed rpm cases. Difficulties of handling the free-to-sink-and-
trim case may be revealed by comparing 2.3a and 2.3b, so the available set of results may be
cut in different directions. To get a quantitative base for these comparisons Table 6 has been
prepared. Here actual propeller results may be compared with those from modeled propellers,
computations with a given SFC with those with a given n, and the fixed attitude results from
2.3a with the free attitude results from 2.3b. The comparisons are made in terms of the mean
error Emean and the mean absolute error |E|mean, both in per cent of the experimental data. A
standard deviation is not meaningful, except in the comparison between 2.3a and 2.3b, since
the other comparisons include two cases.

KT KQ n RT (SP)-T
Items
(No. Entries / Total) E%D |E%D| E%D |E%D| E%D |E%D| E%D |E%D|
mean mean mean mean mean mean mean mean
Actual prop. (9/17) 3.3 4.1 -1.4 2.9 -2.1 2.1 -7.8 7.8
Modeled prop. (8/17) -2.4 6.5 -3.9 8.1 1.6 2.8
Given SFC (12/17) -0.2 6.0 -2.6 6.5 0.4 2.6
Given n (5/17) 2.4 3.2 -2.6 2.6 - - -7.8 7.8
Case2.3a (14/17) -0.6 5,0 -4.6 5.1 0.6 2.3 -7.8 7.8
Case2.3b (3/17) 6.2 6.3 6.7 6.7 -0.3 3.6 - -
Mean (Case 2.3a&b) 0.6 5.2 -2.6 5.4 0.4 2.6 -7.8 7.8
Mean σ (Case 2.3a&b) 7.0 6.0 3.1 8.7
Table 5 : Error statistics, Cases 2.3a and 2.3b

There is a clear trend of smaller scatter for the actual propellers in KT, KQ and n (for
RT(SP)-T there are only actual propellers represented). All three quantities have a smaller
Emean for the actual propellers than for the modeled ones, and the difference is particularly
large KQ. For the mean error Emean there is no clear trend. The actual propeller exhibits a
considerably smaller error in KQ, but for KT and n the errors are slightly larger.
The scatter in the KT and KQ results is quite different between a given towing force and a
given rpm. Emean for given n is only half of that for given towing force, while the mean
signed error Emean is larger for KT. The most surprising result here is the large over prediction
of the towing force for given rpm. If n is given, the towing force is significantly over
predicted, while if SFC is given (and the forces balanced) the rpm is computed very well (see
the relatively small values of Emean and Emean for n). If the propeller is relatively lightly
loaded a small (percentage) change in n may correspond to a relatively large (percentage)
change in trust and a corresponding large change in towing force to acquire force balance.
It is seen in Table 5 that the small number of results for 2.3b (only 3) makes it very
difficult to draw conclusions concerning the differences in accuracy between the fixed and
free cases. Emean and Emean for all quantities have been computed and presented in the table,
but we will refrain from drawing any conclusions.
The last two lines of Table 6 are the most interesting ones. They present the mean values of

26
 Lars Larsson, Frederick Stern and Michel Visonneau

all self-propulsion submissions, i.e. a weighted average of the results in 2.3a and 2.3b. These
numbers may give a general indication of the accuracy obtainable in self-propulsion
predictions. The last line gives the weighted mean of the standard deviations in the two cases.
For KT the mean error is 0.6%D and the mean standard deviation 7%D and the corresponding
values for KQ are -2.6%D and 6%D, respectively. The predicted n for a given SFC has a mean
error of 0.4%D and a standard deviation of 3.1%D, while the numbers are larger for the
towing force for given n: -7.8%D and 8.7%D, respectively.

5.3 Wave pattern

Wave pattern predictions were reported for Cases 1.1b (KVLCC2), 2.1 (KCS) and 3.1a, b
(5415). The hulls represent completely different ship types and Froude numbers, so the
capability of the codes to predict the free surface was tested over a wide range of possibilities.
Several different graphs were used to evaluate the performance of the codes. A general
overview was provided in the wave contour plots, where the wave height was given in a
region surrounding the hull. Wave cuts at three distances from the center plane were also
presented for all hulls. These cuts enabled a very detailed comparison between computed and
measured waves, since the measured data were presented in every plot.
Since the predicted waves are strongly dependent on the grid density near the surface every
participant was requested to provide the following grid information: number of grid points per
fundamental wave length along the waterline, number of grid points in the transverse
direction on the surface at midship and step size in the vertical direction near the hull at
midship. This information was plotted in a graph that is presented after the wave figures for
each case in the Proceedings.
Due to the space limitation only one example will be given here, namely the most
challenging one: Case 1.1b. The Froude number for KVLCC2 is quite low, 0.142, which
means that the fundamental wave length 2Fr2 is only 1/8 LPP, so a large number of cells are
required to get a sufficient number of cells per wave length. Small cells are also required in
the vertical direction, since the maximum wave height is less than 1% LPP.
Wave contours for KVLCC2 are presented in Figure 3. The measurements by MOERI
reveal a complex wave pattern with very short waves essentially located at the edge of the
Kelvin wedge. For a hull of this type with pronounced shoulders four wave systems should be
expected: one from the high pressure regions at each end of the hull and one for each
shoulder. However, the speed is so low in this case that no waves seem to be generated near
the stern. The dominating wave system is that from the bow, but close inspection also reveals
a more weak system originating at the forward shoulder and merging with the bow system
after a short distance.
The best results were obtained by ECN/CNRS-ISISCFD, which is an unstructured grid
solver with surface capturing and the VOF technique. The results, as seen in Figure 3, reveal
all the details of the wave pattern. In fact, the computations display the generated shoulder
wave system more clearly than the experiments. It is seen inside the main system from the
bow and merges with the latter around x/LPP=0.75. ISISCFD has a newly implemented grid
adaptation technique where the original grid is refined in several steps and concentrated in
regions where a large grid density is required. The total number of grid points was 5.5M, but

27
 Lars Larsson, Frederick Stern and Michel Visonneau

grid convergence was reported even with 2.7M grid points. The grid density plot (not shown
here) shows that along the hull the step size has two peaks, like in most grid plots, one at the
bow and one at the stern. In this case the maximum grid density at the bow is around 200
points per wave length (ppwl), while at the stern the density is somewhat lower, around 150
ppwl. Along the main part of the hull the density is around 70 ppwl. There is an interesting
variation in the grid density in the transverse direction. Like in all methods the transverse grid
density is very large close to the hull, in order to resolve the boundary layer, so there is first a
rapid drop in ppwl, moving outwards. In most methods this drop is gradually reduced and a
relatively smooth curve is obtained from the hull towards the outer edge of the free surface
domain. However, in this case the adaptivity created a peak in the region y/LPP= 0.15-0.25.
This is where the bow wave system passes x=0, where the grid density is reported. In the
vertical direction the step size was reduced to about 3×10-4LPP close to the surface, which
corresponds to about 20 cells per maximum wave height.

Figure 3: Wave pattern around the KVLCC2 at Fr=0.142.

Top : measurements from MOERI, bottom : computations by ECN/CNRS-ISISCFD

5.4 Local flow (steady case)

A large number of local flow results were requested at the Workshop. Contours of axial
velocity and cross-flow vectors were to be provided for all three hulls both under steady (all
hulls) and unsteady (5415) conditions, even with an operating propeller (KCS). This
information was requested for one or more cross-planes. Velocity profiles in the propeller

28
 Lars Larsson, Frederick Stern and Michel Visonneau

plane were also asked for. Turbulence data at the propeller plane were requested for KVLCC2
and 5415. For KVLCC2, wake fraction contours and limiting streamlines were to be reported
and surface pressure distributions should be given for KVLCC2 and KCS.
Here only one example can be given, and to be able to compare with previous workshops
we select the flow at the propeller plane of the KVLCC2. The flow at model scale around this
hull is characterized by the gradual development of an intense stern bilge vortex which creates
a strong distortion of the axial velocity iso-contours at the propeller plane. See Figure 4. This
distortion is due to the transport of low momentum fluid from the vicinity of the hull to the
center of the flow field under the action of the longitudinal vortex. Under the main vortex, one
can guess the existence of a secondary counter-rotating vortex close to the vertical plane of
symmetry. This leads to the so-called hook-shape of the iso-wakes which is clearly visible
both in the towing tank experiments from MOERI and wind tunnel experiments from Postech.
In Figure 4 the Postech results are presented. There is however a difference between the two
results particularly in the vicinity of the vertical plane of symmetry (the level U=0.4). These
local differences may be attributed to blockage effects, the tunnel blockage being more than
6% while the towing tank blockage is only 0.3%. On the other hand, it seems easier to control
the quality of the measurements (in terms of flow symmetry for instance) in a wind tunnel
than in a towing tank where small free-surface deformations may create perturbations. These
various sources of experimental errors will have to be considered during the comparisons with
computations which were performed without any blockage effect and free-surface
deformation. It should also be mentioned that the hull geometry provided for the G2010
workshop is slightly different from the original KVLCC2 since a semi-hemispherical cap
(dummy hub) has been placed at the end of the hub to make the flow around the shaft center
line smooth.

(a) (b)
Figure 4: Cross-flow vectors, 2D ‗streamlines‘and axial velocity contours at the propeller plane of the KVLCC2.
(a) Measurements at Postech9 ; (b) Computations by IIHR-CFDShip Iowa-V.4.5 (ARS)

At the propeller plane the bilge vortex has developed and its impact on the iso-wakes is
very large. A first group of results is in very good agreement with experiments, namely
IIHR/CFDShip-Iowa-V4.5 (ARS, DES), NTNU/FLUENT, NMRI/SURF,
Chalmers/SHIPFLOW4.3, MARIC/FLUENT6.3 and NavyFOAM/NavyFOAM. The
IIHR/CFDShip-Iowa-V4.5 (ARS) results are displayed in Figure 5. Except
NavyFoam/NavyFoam which uses a k-ω model (original version of 1998 according to their
paper), all other results are based on various anisotropic turbulence models. IIHR/CFDShip-
Iowa-V4.5 used both an algebraic Reynolds Stress model (ARS) and an ARS based DES

29
 Lars Larsson, Frederick Stern and Michel Visonneau

version while both Chalmers/SHIPFLOW4.3 and NMRI/SURF use the Explicit Algebraic
Stress Model developed by ECN-CNRS some years ago. Computations performed with
FLUENT (NTNU/FLUENT and MARIC/FLUENT6.3) are based on a more complex
Reynolds Stress Transport Model which solves additional transport equations for the
Reynolds Stress components. It is interesting to notice that all these results agree better with
the towing tank than with the wind-tunnel experiments. The main stern bilge vortex is very
accurately captured and the hook-shape of the iso axial velocity contours is very well
reproduced. A second counter-rotating vortex, hardly visible in the experiments, is present in
all these computations. In that region, the agreement between the best solutions and the
experiments is less good. Let us recall that this is a region where the flow is probably
influenced by the shape of the hub which is slightly different because of the presence of the
hub cap in the computations. One can also notice that the ARS based DES solution contains
more intense longitudinal vortices, a characteristic already observed in other test cases (Case
3.1a for instance).
On the other hand, linear eddy viscosity models without ad-hoc rotation correction under-
estimate the intensity of the bilge vortex (ECN-BEC/Icare, HSVA/FreSCo+, IST-
MARIN/PARNASSOS-SST, VTT/FINFLO). A noticeable exception is NavyFoam, based on
the Wilcox's k-ω 1998 model, which gives a prediction similar to that obtained with algebraic
Reynolds stress model. As this model is not as widely used as the SST model for example,
and this peculiar performance needs to be further validated by other flow solvers.
IST/MARIN presents some good predictions for the nominal velocity obtained with linear
eddy-viscosity model with rotation correction or with a linear turbulence model. The
improvement obtained by those ad-hoc modifications seems to be limited only to the mean
velocity field at propeller plane. In particular, the recirculation region seems to be extended
more upstream.
It is well known that turbulence anisotropy is an additional source of longitudinal vorticity
production. In fact the turbulence anisotropy acts as a direct source term in the transport
equation of the longitudinal vorticity. Having the normal Reynolds stresses available at the
propeller plane makes possible a detailed verification of the amount of measured anisotropy in
the plane. If one compares the relative values of the normal turbulent stresses, one can notice
a strong anisotropy inside the characteristic hook shape found in the iso-axial velocity
contours. For instance, max(uu)=0.016, max(vv)=0.007 and max(ww)=0.008 while
max(k)=0.016. Most of the codes using explicit anisotropic turbulence models are able to
predict with a reasonable agreement the turbulence structure at this cross-section. For
instance, NMRI/SURF finds max(uu)=0.014, max(vv)=0.008 and max(ww)=0.010. The
relative weights of the respective normal turbulent stresses are correctly predicted by DES,
RSTM and EASM turbulence models while the linear isotropic models fail to reproduce the
measured turbulence characteristics. Therefore, instead of using the right mechanism to
enhance the longitudinal vorticity production, correction factors are used to limit the
production of turbulence and consequently, to reduce locally the level of turbulent viscosity.
This is illustrated by comparing the normal turbulent stresses computed by IST-
MARIN/PARNASSOS(SST) to IST-MARIN/PARNASSOS(RCSST), for instance. One
cannot see any significant difference on the normal turbulent stresses while the iso axial
velocity contours differ. If one compares also the turbulent normal stresses predicted by the

30
 Lars Larsson, Frederick Stern and Michel Visonneau

Reynolds Stress Transport model implemented in Fluent, one can notice a remarkable
agreement between NTNU/FLUENT and MARIC/FLUENT6.3. Both organizations found
max(uu) around 0.01, max(vv) around 0.004 and max(ww) around 0.005, results which are
consistently smaller than the experiments. This is again an illustration of a consistent trend
associated to a specific turbulence modeling, independently from the grid (which has to be
fine enough) and from the user of the solver (who has to be experienced enough...).
This is also the first time that DES results are available for the KVLCC2. The spatial
distribution of uu is organized around two peaks of uu with maximum values around 0.02,
greater than the measurements supposed to be around 0.016. The same structure in two peaks
is observed for vv and ww with maximum values around 0.003 and 0.004, respectively. The
turbulence anisotropy is therefore more pronounced than what is observed with the
anisotropic non-linear turbulence closures. This is probably related with the more pronounced
longitudinal vorticity which is found in the DES computations. Although the boundary
delimitating full RANSE and full LES formulations is not known, one can suspect that this
effect is due to the LES formulations by comparing with the normal turbulent stress
distribution obtained with the ARS model.
The turbulent shear stresses uv and uw were also measured and can be used to perform a
detailed assessment of the computations. The agreement of all the computations is reasonable
for uv except for the DES computations which again do not reproduce the measured spatial
distribution. For uw, most of the contributors find a zone of uw > 0.002 which is consistently
smaller than what is observed in the experiments, except the DES closure implemented in
CFDShip-Iowa which is in good agreement with the experiments.
Compared to the situations in 2000 and 2005, one can observe that much progress has been
made towards consistent and more reliable computations of afterbody flows for U-shaped
hulls. The intense bilge vortex and its related action on the velocity field is accurately
reproduced by a majority of contributors employing very similar turbulence models
implemented in different solvers and on different grids. The debate on the relative importance
of discretization vs. modeling errors opened in the mid-nineties should now be closed by the
acknowledged prominent role played by turbulence anisotropy as long as a reasonable grid is
used. From that point of view, around 3 million points are enough to assess the turbulence
closures without any significant pollution from discretization errors. The turbulence data
confirm that the turbulence anisotropy is large in the propeller disk and more specifically in
the core of the bilge vortex. Explicit Algebraic Stress and Algebraic Reynolds Stress Models
reproduce satisfactorily the measured structure of the turbulence and appear to be, up to now,
the best answer in terms of robustness and computational cost for this specific flow field,
compared to RSTM or DES strategies. Having recourse to Delayed DES will probably
strongly improve the present DES results. A more detailed analysis of the turbulence
characteristics remains to be done through the use of the Reynolds stress anisotropy tensor,
anisotropy invariant maps and the analysis of the turbulent kinetic energy budget.

5.5 Heave and pitch

Several seakeeping cases were included in the workshop. Test cases 1.4a,b and 2.4 are for
pitch and heave in regular head waves for KVLCC2 and KCS, restrained in surge. In test case

31
 Lars Larsson, Frederick Stern and Michel Visonneau

1.4c for KVLCC2 the surge is also released. Test case 3.5 is forward speed diffraction for
5415 fixed at dynamic sinkage and trim, while Case 3.6 is free roll decay for 5415 at the same
attitude. These test cases have not been included in previous CFD Workshops, except test
case 3.5, which was included at CFD Workshop Tokyo 2005.
New model tests were conducted for KVLCC2, without appendages, rudders, and
propellers. The tests were carried out for all three 1.4 cases (a,b,c). The free surge tests were
conducted in the Osaka University‘s towing tank for a 1/100 scaled model (LPP=3.2 m) using
a spring to attach the model to the towing carriage. Fixed surge tests were conducted with the
same model in the INSEAN 220× 9× 3.5 m towing tank and for a larger model (LPP =5.5172
m) at NTNU.
For Case 1.4a-c calm water, four experimental data sets are available which show different
values for resistance, sinkage, and trim. Unfortunately, uncertainty of the data was not studied
so the origin of the differences cannot be assessed. For the heaving and pitching motions the
surge, neglected in 1.4a and b, may be important, since data for surge motion shows that its
amplitude increases with wavelength, reaching up to 50% of the wave amplitude for the range
of wavelengths tested. For Case 1.4a, there was no phase reference recorded for the data, so
phases are not used for CFD error studies. Also, for test case 1.4b the surge motion was
partially constrained by a spring system, and as a result the measured resistance amplitude is
not used for comparison.
The CFD simulations of cases1.4a-c included 12 submissions from 5 institutions with the
number of grid points of 0.3-4.7 M. Verification studies were performed by one submission
for Case 1.4a in calm water and waves. The quantities with no higher order effects including
steady resistance, 0th harmonic of resistance, and 1st harmonic of motions had minimum
prediction errors for all cases. CFD achieved the prediction of 1st order and higher order
quantities in the average level of 16%D and 59%D, respectively. For steady and 0th harmonic
of resistance, the average error was 17.5%D while the 1st harmonic amplitude and phase were
41%D and 6.5%D, respectively. For motions, the prediction error was 66%D for steady and
0th harmonic while it was around 11%D for 1th harmonic amplitude and 28%D for 1st
harmonic phase. Therefore, for resistance, the largest error values were observed for the 1st
amplitude, followed by 0th amplitude and then 1st phase. For both heave and pitch motions,
the largest error values were observed for the 0th amplitudes followed by 1st phase and 1st
amplitude.
The number of grid points seems to have an obvious effect on both motions and resistance
results. For most conditions, the smallest errors were for the submission with largest number
of points. The other submissions were usually with higher errors based on how coarse their
grids are. It may be that all codes would reach the small level of error, if using the finest grid.
The smallest error averaged over amplitudes and phase for resistance was 11.19%D for Case
1.4c with λ/LPP =1.1 for CFDShip-Iowa with 4.7M grid points. Also the smallest error
averaged over amplitudes and phase for motions was 12.88%D for Case 1.4c with λ/LPP =1.1
for CFDShip-Iowa with 4.7M grid points.

5.6 Roll decay at forward speed

The model-scale test for 1/46.6 scale 5415 bare hull with bilge keels free to roll-decay

32
 Lars Larsson, Frederick Stern and Michel Visonneau

advancing in calm water was performed in the IIHR towing tank13.The flow conditions were
Re= 2.56×106, Fr=0.138, sinkage = 2.93×10-4 LPP, trim = -3.47×10-2 degrees and initial roll
angle = 10 degrees. Data were procured for the forces and moments using a strain gage load
cell, the unsteady ship roll motion using a Krypton Motion Tracker, the unsteady wave
elevation on the starboard side using four servo wave probes, and unsteady velocities at x/LPP
= 0.675 in a region near bilge keels using a 2D PIV system.
Four organizations (ECN-ICARE, ECN/CNRS-ISISCFD, GL&UDE-OpenFOAM, and
SSRC/Univ of Strathclyde-Fluent 12.2) contributed for this test case, all using URANS
methods. All the submissions showed non-linear oscillations for the total resistance as
observed in the experiments. The mean resistance was predicted within 10%D of the
experimental data. The amplitude and period of the roll motions were predicted very well
within 0.85%D as shown in Figure 5.

Figure 5: Time history of roll angle

(open circle: experiments; solid black line: CFD (GL&UDE-OpenFOAM))

(a) (b)
Figure 6: Contours of U velocity at x/LPP = 0.675 during second cycle of roll decay (t/Te = 0)
(a) Experiment (IIHR, Irvine et al. 2004); (b) Computations by ECN/CNRS-ISISCFD

The ECN-ICARE simulation on a 0.8M grid failed to predict the Kelvin wave pattern and
the development of the wave troughs and crests due to the roll motion. Likewise, the
SSRC/Univ of Strathclyde-Fluent 12.2 predictions on a 3M grid also showed poor Kelvin

33
 Lars Larsson, Frederick Stern and Michel Visonneau

wave predictions, and this was claimed to be due to grid resolution issues.
On the other hand, the ECN/CNRS-ISISCFD predictions on 4.9M grid showed overall
good agreement for the Kelvin wave pattern and the development of wave troughs and crest at
the shoulder, but the waves were closer to the hull compared to the data and dissipated away
from the hull. Overall, the wave elevation predictions improved with grid resolution, but for
such low Fr (which exhibits a small Kelvin wavelength) even larger grids are required to
accurately predict the wave elevation pattern.
The coarse grid ECN-ICARE simulation over-predicted the boundary layer thickness and
under-predicted the cross flow velocities by 30%D. However the general trends in the cross-
flow pattern at the different roll phases were well predicted. This suggests that the generation
of the bilge keel vortices was predicted well, but the vortex strengths were under-predicted
and significantly diffused compared to the experimental data. ECN/CNRS-ISISCFD
predictions on a finer grid compared very well with the experiments, where the minimum and
maximum velocities compared within 6% D, see Figure 6. The cross-flow predictions were
slightly better than the axial velocity, where latter showed over-prediction of the transport of
low momentum fluid away from the bilge keel. This suggests that the vortex advection due to
roll motion was predicted well, but the bilge keel vortex inception was not predicted
accurately. Even finer grids near the bilge keel are required to capture the vortex inception
accurately.
Overall, the results showed that the force and roll motion predictions are not significantly
dependent on a large resolution, but it is more important for the wave elevation and flow
predictions. A further point to note is that the anisotropic turbulence models do not show
significant improvement over the isotropic models for the global variable predictions. This is
because the vortex generation is imposed by the geometry of the bilge keels, and thus the
turbulence models do not influence the flow predictions. This is different from the KVLCC2
test case 1.1a or 5415 test cases 3.1 and 3.5, where the vortices are advected and the
anisotropic turbulence models show improved predictions.

6 CONCLUSIONS
The Gothenburg 2010 Workshop on numerical Ship Hydrodynamics was huge effort by a
large number of people. 33 groups participated and computed one or more of the 18 test cases
for the three hulls. The results represent the state-of-the-art in computational hydrodynamics
at present. For the areas covered in the present paper the main conclusions are:
- The mean error (data(D) – simulation) for all computed resistance cases was
practically zero; only - 0.1%D, and the mean standard deviation was 2.1%D. The
latter represents a considerable improvement since 2005, where the mean standard
deviation was 4.7%D.
- Of the reported self-propulsion predictions 9 used an actual rotating propeller,
while 8 used a hybrid approach with a potential flow propeller model linked to the
viscous method. The scatter between the former predictions was roughly half of
that of the latter, while the mean error was about the same. The total mean error for
KT and KQ was 0.6%D and -2.6%D, respectively, while the mean standard
deviation was 7.0%D and 6.0%D, respectively.

34
 Lars Larsson, Frederick Stern and Michel Visonneau

- Very accurate predictions of the wave pattern over the entire discretized free
surface were obtained in the most accurate solutions for all three hulls. Even for the
most challenging case, at Fr=0.142, all details of the wave pattern were captured by
a method with adaptive gridding.
- The details of the nominal wake of the tanker hull could be predicted very well
with methods using anisotropic turbulence models. This was true both for mean
velocity and turbulence. The mean velocity distribution could also be captured with
isotropic models with rotation correction, but with an erroneous distribution
between the normal Reynolds stresses. DES seemed to over-predict the anisotropy,
thereby exaggerating the bilge vortex strength and the ―hooks‖ in the wake
contours.
- Three new sets of data were provided for the workshop in the seakeeping area. The
tanker was tested in head waves free in pitch and heave at two organizations and
free also in surge at a third laboratory.
- A Fourier analysis was carried out of the unsteady motions for the heave and pitch.
For the tanker cases the average accuracy in resistance was 17.5% in the 0 th
harmonic, while the first harmonic amplitude and phase had errors of 41%D and
6.5%D, respectively. The 0th harmonic of the heave and pitch had an error of 66%,
while the first harmonic amplitude and phase differed from the data by 11% and
28% respectively
- For the roll decay the mean error in resistance was around 10%, while the
amplitude and period were predicted very well, 0.85% on the average. Waves were
predicted rather well by a method using adaptive grids and 4.9M cells, but other
methods with coarser grids failed. This complicated case at low Froude number
(0.142) calls for larger grids, at least if they are not adapted. This is the case also
for the mean velocities in the boundary layer.

7 ACKNOWLEDGEMENTS
The workshop was organized by a committee with six members: the authors and Dr.
Emilio Campana, Dr. Suak Ho Van and Prof. Yasuyuki Toda, whose contributions are
gratefully acknowledged. Very important contributions have also been made by Lu Zou, who
compiled all results and prepared the Proceedings and by Dr. Alessandro Iafrati, who
developed and maintained the web site. Prof. Rickard Bensow and Andreas Feymark prepared
and compiled the questionnaire and Shanti Bhushan, Hamid Sadat-Hosseini and Maysam
Mousaviraad contributed to the specification and analysis of the seakeeping cases. Finally, the
great efforts by all workshop participants in the preparation and delivery of all computed
results shall not be forgotten.

REFERENCES
[1] L. Larsson, (Ed.) ―SSPA-ITTC Workshop on Ship Boundary Layers‖, SSPA Report 90,
Gothenburg, Sweden (1981).
[2] L. Larsson, V.C. Patel, and G. Dyne, (Eds.) ―SSPA-CTH-IIHR Workshop on Viscous
Flow‖, Flowtech Research Report 2, Flowtech Int. AB, Gothenburg, Sweden (1991).

35
 Lars Larsson, Frederick Stern and Michel Visonneau

[3] Y. Kodama, H. Takeshi, M. Hinatsu, T. Hino, S. Uto, N. Hirata, and S. Murashige,

Proceedings, CFD Workshop, Ship Research Institute, Tokyo, Japan (1994).
[4] L. Larsson, F. Stern, and V. Bertram, (Eds.) Gothenburg 2000-A Workshop on
Numerical Hydrodynamics, Department of Naval Architecture and Ocean Engineering,
Chalmers University of Technology, Gothenburg, Sweden (2002).
[5] L. Larsson, F. Stern and V. Bertram, ―Benchmarking of computational fluid dynamics
for ship flow: the Gothenburg 2000 Workshop‖, J. Ship Research, Vol. 47: 63-81 (2003)
[6] T. Hino, (Ed.) ―2005 Proceedings of CFD Workshop Tokyo 2005‖, NMRI report (2005).
[7] S.H. Van, W.J. Kim, D.H. Yim, G.T. Kim, C.J. Lee, and J.Y. Eom, ―Flow Measurement
around a 300K VLCC Model‖, Proceedings of the Annual Spring Meeting, SNAK,
Ulsan, pp. 185-188 (1998a).
[8] S.H. Van, W.J. Kim, G.T. Yim, D.H. Kim, and C.J. Lee, ―Experimental Investigation of
the Flow Characteristics around Practical Hull Forms‖, Proceedings 3rd Osaka
Colloquium on Advanced CFD Applications to Ship Flow and Hull Form Design,
Osaka, Japan (1998b).
[9] W.J., Kim, D.H., Van, and D.H., Kim, ―Measurement of Flows around Modern
Commercial Ship Models‖, Exp. in Fluids, Vol. 31, pp. 567-578 (2001).
[10] S.J. Lee, H.R. Kim, W.J. Kim, and S.H. Van, ―Wind tunnel tests on flow characteristics
of the KRISO 3600 TEU Containership and 300K VLCC Double-deck Ship Models‖, J.
Ship Res., Vol. 47, No 1, pp. 24-38 (2003).
[11] C. Simonsen, J. Otzen, and F. Stern, ―EFD and CFD for KCS Heaving and Pitching in
Regular Head Waves‖, Proc. 27th Symp. Naval Hydrodynamics, Seoul, Korea (2008).
[12] A. Olivieri, F. Pistani. A. Avanaini, F. Stern, and R. Penna, ―Towing Tank Experiments
of Resistance, Sinkage and Trim, Boundary Layer, Wake, and Free Surface Flow around
a Naval Combatant INSEAN 2340 Model‖, Iowa Institute of Hydraulic Research, The
University of Iowa, IIHR Report No. 421, pp. 56 (2001).
[13] J. Longo, J. Shao, M. Irvine, and F. Stern, ―Phase-Averaged PIV for the Nominal Wake
of a Surface Ship in Regular Head Waves‖, ASME J. Fluids Eng, Vol. 129, pp. 524-540
(2007).
[14] M. Irvine, J. Longo, F. Stern, ―Towing Tank Tests for Surface Combatant for Free Roll
Decay and Coupled Pitch and Heave Motions”, Proc. 25th ONR Symposium on Naval
Hydrodynamics, St Johns, Canada (2004).
[15] F. Stern, K. Agdrup, S.Y. Kim, A.C. Hochbaum, K.P. Rhee, F. Quadvlieg, P. Perdon, T.
Hino, R. Broglia, and J. Gorski, ―Experience from SIMMAN 2008: The First Workshop
on Verification and Validation of Ship Maneuvering Simulation Methods‖, Journal of
Ship Research, in press.
[16] L. Larsson, F. Stern, and M. Visonneau, ―Gothenburg 2010-A Workshop on Numerical
Ship Hydrodynamics‖, Proceedings Vol. 2, Department of Naval Architecture and
Ocean Engineering, Chalmers University of Technology, Gothenburg, Sweden (2010).

36