Você está na página 1de 12

Applied Ocean Research 35 (2012) 5667

Contents lists available at SciVerse ScienceDirect

Applied Ocean Research


journal homepage: www.elsevier.com/locate/apor

Seakeeping prediction of KVLCC2 in head waves with RANS


B.J. Guo a, , S. Steen a , G.B. Deng b
a
b

Department of Marine Technology, NTNU, Trondheim, Norway


E.M.N., L.M.F-UMR6598, Ecole Centrale de Nantes, 1, Rue de la Noe, France

a r t i c l e

i n f o

Article history:
Received 16 August 2011
Received in revised form
22 November 2011
Accepted 20 December 2011
Available online 3 February 2012
Keywords:
Added resistance
Ship motions
KVLCC2
Head waves

a b s t r a c t
The present work is devoted to the prediction of added resistance and ship motion of KVLCC2 in head
waves. Systematic validation and verication of the numerical computation demonstrate that reliable
numerical results can be obtained in calm water as well as in head waves. The numerical results are
analyzed in terms of added resistance, ship motions and wake ow. Both free to heave and pitch and
xed model are studied to investigate the contribution to added resistance from ship motion at different
wavelengths, and the results show that ship motion induced added resistance is negligible when the
wavelength  < 0.6 Lpp . The comparison with theoretical calculation based on strip theory and experimental results shows that RANS predicts the added resistance better in all wavelengths. Ship pitch and
heave motion in regular head waves can be estimated accurately by both CFD and strip theory. Finally,
wake ow at the propeller plane in waves is discussed, and the numerical results show that the change
of axial velocity due to one studied incoming head waves is at most over 30% of the ship forward speed.
2011 Elsevier Ltd. All rights reserved.

1. Introduction
CFD approach relying on the resolution of the Reynolds Averaged Navier-Stokes equation (RANS) overcomes the limitation of
the potential ow with respect to effects of water viscosity, wave
dispersion, nonlinearity and wave breaking [1]. Moreover, effects
of turbulence are taken into account with turbulence model. Consequently, the application of CFD in the ship industry is increasing.
In addition to traditional ship motions and resistance information, RANS calculations can easily provide comprehensive ow eld
information, which is helpful for ship design optimization [2].
Prediction of ow around ship hulls using RANS started in the
1980s, and a majority of the studies were initially devoted to ship
resistance and ow eld prediction in steady state. Even in the
Gothenburg CFD workshop 2000, all benchmark test cases concerned steady RANS computation [3]. Before this workshop, a few
seakeeping computations had been performed. Wilson et al. [4] presented an unsteady ow study in head waves, studying the force
and diffraction wave pattern, for the Wigley hull and the DTMB
Model 5415. However, only the diffraction problem with forward
speed at a few wavelengths was discussed. Sato et al. [5] simulated pitch and heave motions in head waves with a modied
Wigley model and Series 60. The agreement between the CFD and

Corresponding author at: Department of Marine Technology, Norwegian University of Science and Technology, Otto Nielsens v10, NO-7491 Trondheim, Norway.
Tel.: +47 73 55 11 98.
E-mail address: bingjie.guo@ntnu.no (B.J. Guo).
0141-1187/$ see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apor.2011.12.003

experimental results was good, but some discrepancies were noted,


especially for the heave motion of the Series 60 model. Rhee and
Stern [6] performed a validation work on ship resistance, wave eld
and wake in head waves for a modied Wigley hull. Their calculations restrained ship motion to study the diffraction problem. Hideo
et al. [7] studied added resistance in regular incident waves with
the SR108 container ship. A further study of unsteady surface pressure, added resistance and ship motions of the SR221C model was
performed by Hideo [8]. These works lack a systematic verication
study of the numerical computation.
A forward speed diffraction benchmark test case is proposed
for the Tokyo CFD workshop in 2005. It is the rst time that the
unsteady RANS for ship hydrodynamics was benchmarked [9]. Four
groups were involved in this calculation. The standard deviation for
the 0th-order total resistance and heave force are 4.5% and 6.7%, but
it is about 70% for the pitch moment [10]. Such high level of uncertainty clearly suggested that CFD approach was not yet a reliable
tool for seakeeping application at that time. Since then, more and
more efforts are devoted to improve the accuracy of CFD computation for seakeeping application. Weymouth et al. [11] extended
the work by Wilson et al. [4], and Rhee and Stern [6] to predict ship
pitch and heave motion of a modied Wigley hull in head waves.
Their simulation shows very good agreement with experimental
data, and extensive uncertainty analysis was performed following
the procedure given by Stern et al. [12]. Their studies focus mainly
on long waves. Only a few short wave cases were studied. Deng
et al. [13] studied several short wave cases for a container ship,
but the discrepancy for added resistance estimation in short waves
reached 50%.

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

To prepare for the Gothenburg 2010 CFD workshop, Deng et al.


[10] further studied the added resistance of KVLCC2 in three different wavelengths, but they lacked experimental validation. KVLCC2
are selected as a benchmark model in the Gothenburg 2010 CFD
workshop [14]. 5 groups from 4 countries performed the relevant
calculation, and the seakeeping prediction for six wavelengths was
compared. The shortest wavelength is Lpp / = 1.57. The standard
deviation for the mean resistance is around 11.3%, while the maximum deviation of 22.0% happens at the longest wave. The standard
deviations of the 0th-order heave and pitch motions are 30.9% and
30.6%, and that for the 1st-order cases are 14.1% and 14.6%.
While the number of participants contributing to seakeeping
benchmarks during the Gothenburg workshop conveys a clear message that CFD becomes a feasible tool for seakeeping application,
but the high value of standard deviation and comparison error of
the submitted results do not allow to conclude that CFD is a reliable
tool for such kind of application. One of the main reasons leading to such high level of uncertainty is that nearly all participants
have performed their computation for each test case with a single grid using a single time step. Almost no numerical verication
was performed. In order to assess the reliability of CFD approach
for seakeeping application, this paper extends the work of Deng
et al. [10] about added resistance ship motion analysis for KVLCC2
at different wavelengths, especially in short waves. Due to data
scattering, an uncertainty analysis based on the least squares root
method is followed here, as proposed by Eca and Hoekstra [15]. A
recently revised version can be found in Eca et al. [26]. The shortest wavelength is Lpp / = 2.453. Both free heave and pitch motion
and restrained motion in head waves were performed to study the
contribution from ship motions to added resistance. The ship resistance components in waves are also analyzed to study the viscous
effect on the added resistance coefcient. Another objective is to
study the nominal wake eld at the propeller plane in waves, and
to determine how it changes due to the incident waves and ship
motions. Such a study is generally not possible with any linearized
or inviscid method, and it is quite difcult to measure in a model
test.
2. Ship geometry and test conditions
The KVLCC2 model with a scale 58 is adopted for the numerical calculation. The ship length and main parameters are given in
Table 1. For the sea-keeping test, the designed depth is too small to
avoid water on deck. Thus, the depth of the hull is increased through
the extrapolation of the body plan to a depth of 0.7 m (model scale).
Fig. 1 shows the body plan of model used in both model tests and
numerical calculations. With Fn = 0.142, the ship speed at model
scale is 1.047 m/s.
The waves considered in this study are given in Table 2. The
waves are divided into two groups here. The rst group, containing
the short and intermediate waves, is the waves applied in the experiments performed by the rst two authors in the towing tank at the
Marine Technology Centre in Trondheim, Norway [18]. The second group contains longer waves applied in experiments at Osaka

57

Fig. 1. Body plan (dimensions in mm full scale).

Table 2
Waves used in calculations and experiments.
Lpp /

 (m)

H (m)

T (s)

2.453
1.571
1.571
1.224
1.090

2.249
3.511
3.511
4.507
5.060

0.05
0.05
0.15
0.15
0.15

1.200
1.499
1.499
1.699
1.800

1.667
0.909
0.625

3.310
6.069
8.827

0.10
0.10
0.10

1.456
1.972
2.378

University. Part of these experimental data can be found in Gothenburg 2010 CFD workshop [14]. The model scale in the experiments
performed by the authors is 58, consistent with the numerical calculation in this study. The model scale is 100 in the experiment by
Osaka University, and those results are scaled to model scale 58 in
Table 2.
In order to analyze the numerical and experimental results, a
right-handed coordinate system (x, y, z) used in the Gothenburg
2010 workshop (Larsson et al. [14]) is adopted here (see Fig. 2).
The numerical and experimental results are all transferred into this
coordinate for comparison and analysis.
3. Computation method
There are three distinct physical problems modeled in this
paper: free trim and sinkage motion in calm water, wave diffraction
with forward speed and the free heave and pitch motion response
due to incident waves. In the forward speed diffraction problem,
the model is xed in the design condition but move with the forward speed. However, a regular wave comes into the uid domain
and introduces an unsteady ow around the ship hull. In the free
motion response problem, the hull, moving forward with constant

Table 1
Main particulars of KVLCC2.
Symbol
Scale
Lpp
B
T
D
Cw
CB

Unit

Ship

Model

m
m
m
m

m3

1
320.0
58.0
20.8
30
0.9077
0.8098
312,622

58
5.5172
1.0000
0.3586
0.5172/0.7
0.9077
0.8098
1.6023

Fig. 2. The coordinate system.

58

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

speed in waves, responds to the force and moment with coupled


pitch and heave motion.
3.1. Computational implementation
The ISIS-CFD ow solver developed by EMN (Equipe Modlisation Numrique) is used for the study here. The turbulent ow is
simulated by solving the incompressible RANS with a nite volume
method. The solver can process multi-phase ows and dynamic
grids. Details refer to FINE/Marine V2.2 [16]. In the present numerical simulation, a second order implicit scheme is applied for the
time discretization. The SIMPLE-like algorithm is used for coupling
the velocity and pressure. An analytic weighting mesh deformation approach is adopted to simulate the ship motion. EASM k
model, which is believed to inherit some advantage of the original RSM model, while maintaining the numerical robustness of
a two-equation model, is used for turbulence modelization. The
free surface is handled with a VOF-type free-surface capturing
approach, which has been proved to be suitable for ow involving
hull shape with section are and breaking waves [17].
3.2. Calculation domain and boundary condition
The calculation domain is 1.75Lpp < x < 2 Lpp , 0 < y < 1.8 Lpp , 1.5
Lpp < z < 0.3 Lpp , where the mid-plane of the ship is located at y = 0,
and ship bottom is at z = 0. The stern is at x = 0 and the bow at
x = Lpp . Only half of the ship hull is used in the calculations, thus
a symmetry boundary condition is adopted at the center plane
boundary.
Prescribed pressure boundary condition is adopted at upper and
bottom boundaries. Far eld is set at the downstream boundary.
On the inlet boundary, the Far eld is given in the calm water calculation, while wave generator is imposed for the wave simulation.
At the wave boundary, velocity components as well as the mass
fraction are prescribed with the analytical solution of 1st order
stokes waves with the given wave period and wave height. The
symmetry boundary condition is set on the outside and middle
boundaries, shown in Fig. 3. The Far eld boundary condition is
applied at the outlet boundary. The ISIS code applies automatically
a free exit boundary condition to the Far eld when ow leaves
the computational domain. Due to the use of upwind discretization
scheme and ship advancing motion, wave reection is not observed
at the exit boundary. No additional waves absorbing treatment is
required.

An unstructured hexahedral mesh is used for the computation.


The mesh near the free surface and around the ship hull is rened
locally in order to well capture the free surface and solve the complex ow around the hull, while the mesh becomes coarse toward
the top and bottom of the tank. The mesh is illustrated in Fig. 4.
Wall function is employed for the computation. The value of y+ is
around 3542 except near the bow and the stern, indicating that
the rst grid points are located within the range of validation for
wall function.
In the numerical calculation, the ship is free in pitch and heave,
while ship motion in other freedoms is xed. The ship model starts
at rest with a progressive acceleration to reach Fn = 0.142 at t = 10 s
following a sinusoidal ramp prole to avoid too much excitation
of the free surface.
4. Uncertainty analysis of numerical calculation
Due to the complexity of the simulated model, the difculty to
obtain a well converged mean ow solution for an unsteady ow,
as well as the difculty to ensure grid similarity using an unstructured grid generator, it is not always possible to obtain a monotonic
convergent solution when performing numerical uncertainty study
using a grid convergence index approach. Sophisticated methods
are needed to assess the uncertainty of the numerical results [11].
For the verication and validation of the numerical calculation,
several methods based on Richardson Extrapolation have been proposed (Roache et al. [19], Stern et al. [12], Eca and Hoekstra [15]).
Eca and Hoekstra [15] introduced a Least Squares Root version of
the Grid Convergence Index approach to evaluate discretization
uncertainty, which takes the numerical scatter into account. The
procedure proposed by Stern et al. [12] was implemented in Guo
and Steen [20]. Considering the scatter in the numerical result, the
verication and validation of CFD results here follows the least
squares root approach by Eca and Hoekstra [15].
5. Discussion of results
Grid similarity is required when performing numerical uncertainty estimation based on Richardson extrapolation. Compared
with simple geometries, grid similarity around a ship is much more
difcult to ensure. Care has been taken when generating the grid
to keep grid similarity as much as possible. Renement ratio used
for uncertainty estimation is determined from the number of grid
cells.
Added resistance in waves is the difference between the mean
resistance in waves and the calm water resistance. To obtain the
added resistance, two different computations were performed: a
calm water resistance computation and a sea-keeping computation in head waves. Both results will be presented separately in the
following sections.
5.1. Steady ow

Fig. 3. Boundary conditions.

Steady computation is adopted for the calculation in calm water.


The accuracy of this calculation is examined by the comparison
of CFD results with experimental results. The uncertainty analysis
then provides the spatial discretization error. Table 3 shows the
numerical resistance Rt (N), ship sinkage 3 (mm) and trim 5 ( )
obtained with four meshes. e is the difference between numerical
and experimental results.
Table 3 shows that the comparison error of numerical ship resistances is around 2.6% for the ne mesh. The ship trim has good
agreement with experimental result, while the sinkage is underestimated. One of the reasons for the large relative error of sinkage is
its small value. A small magnitude will give a large relative error. In

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

59

Fig. 4. Mesh used for numerical simulation.

our measurement, the sinkage is measured with respect to a reference position at the carriage. The variation in elevation of towing
tank rails above water is estimated to be around 1 mm. Although the
difference between numerical prediction and measurement data is
smaller than 1 mm, the comparison error is still as high as 12%.
5.1.1. Verication of numerical calculation
The numerical resistance and ship motions versus renement
ratio hi /h are shown in Figs. 5 and 6. The tted curves based on
the least squares root approach as well as tted curves based on
theoretical accuracy Ptheory = 2 are also plotted, where hi is a mesh
size calculated with cell number.
The observed order of accuracy PG = 2.02 for the resistance is
very close to the theoretical value. The PG of sinkage and trim are
also close to the theoretical value Ptheory = 2. The observed order of
accuracy PG , extrapolated solution 0 , standard deviation US % D,
and uncertainty UG are all listed in Table 4. The standard deviation
US % D is less than 1% for each physical parameter, which shows that
the numerical results distribute smoothly along the tted curve.

This could also be seen in Figs. 5 and 6. The uncertainty is less than
6% for all the predicted parameters.
5.1.2. Validation of numerical calculation
The validation of the numerical model in calm water is given in
Table 5. Steady ow is adopted for the numerical simulation in calm
water, so time step uncertainty (UT ) does not exist in the validation.
Since iterative errors are much smaller than the relative change of

Table 3
Ship resistance and motions in calm water.
Coarse

Medium

Fine

Finest

Mesh

351,018

775,031

1,365,729

2,284,161

Exp.

Rt
e

19.886
9.26%

19.166
5.31%

18.942
4.08%

18.674
2.60%

18.200

3
e
5
e

5.938
7.49%
0.1328
5.40%

5.753
10.38%
0.1297
2.94%

5.671
11.65%
0.1282
1.75%

5.642
12.10%
0.1277
1.35%

6.419
0.126
Fig. 5. Grid convergence of total resistance in calm water.

60

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

Fig. 6. Grid convergence of ship motions in calm water.

Table 4
Uncertainty of ship resistance and motions.

Rt
3
5

Table 6
Ship resistance and motions at  = 0.9171 Lpp predicted with T = Te /140.

PG

0

US % D

UG

UG % D

2.02
2.78
2.57

18.232
5.5073
0.1263

0.36%
0.13%
0.15%

0.219
0.37
0.0064

3.39%
5.76%
5.07%

Table 5
Validation of numerical model in calm water.

Rt
3
5

UG % D

UD % D

UVal % D

2.60%
12.10%
1.35%

3.39%
5.76%
5.07%

1.0%
1.26%
0.88%

3.53%
5.90%
5.15%

calculation parameter with mesh size, iterative uncertainty (UI ) is


also neglected. Then numerical simulation uncertainty USN in calm
water is equal to grid uncertainty (UG ) [12]. The validation uncer2 = U 2 + U 2 , where U is the
tainty is calculated according to UVal
D
D
G
uncertainty of model test. The validation comparison error E is the
difference between the numerical solution from nest case (S) and
experimental data (D).
Table 5 shows that the validation comparison error E of the
resistance and trim is smaller than the validation uncertainty.
Beside, both the comparison error and the validation uncertainty are relatively small, from which it can be concluded that
calculated resistance and trim are in good agreement with experimental results. For sinkage (3 ) this is not the case the relative
error between calculation result and experimental result is large
(12.1%), however, the absolute error is relatively small (0.78 mm),
since the sinkage is very small at the low speed studied. The low
value of the sinkage means that even small bias errors might have
large inuence on the relative error. We believe that a variation of
the elevation of the rails in the towing tank is a major contributor
to the deviation in sinkage. The validation uncertainty of sinkage
would be 15.5% in this case, if a variation of rail elevation of 1 mm
is included in the experimental uncertainty.

Coarse

Medium

Fine

Finest

Mesh

748,748

1,346,253

2,147,731

3,626,036

Caw
e

5.035
9.43%

4.827
4.91%

4.664
1.37%

4.610
0.20%

4.601

0th 3
e
0th 5
e
1st 3
e
1st 5
e

6.135
5.84%
0.1301
5.03%
12.005
3.22%
1.373
1.16%

6.258
3.96%
0.1327
3.14%
11.873
2.08%
1.360
0.199%

6.369
2.25%
0.1331
2.85%
10.963
5.74%
1.295
4.58%

6.385
2.01%
0.1335
2.56%
11.170
3.96%
1.298
4.38%

6.516
0.137
11.631
1.357

from different meshes are compared with experimental data in


Table 6. Added resistance coefcient Caw is calculated and nondimensionalized according to:
Caw =

Raw
ga2 B2 /Lpp

(1)

where Raw = R wave Rcalm is added resistance, R wave is mean resistance in waves and Rcalm is resistance in calm water.  is the water
density,  a is the wave amplitude. B and Lpp are given in Table 1.

5.2. Unsteady ow
5.2.1. Uncertainty estimation for spatial discretization
Considering the large discrepancy near the ship motion peak
area (22.0%) reported at the Gothenburg 2010 workshop, wavelength /Lpp = 0.9171 is selected here for the uncertainty analysis.
To evaluate the spatial discretization error, 4 different meshes are
used. For those computations, the same time step with 140 time
steps per wave encounter period is used. Numerical results in waves

Exp

Fig. 7. Grid convergence of added resistance in waves.

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

61

Fig. 8. Grid convergence of ship motions in waves.

Table 6 indicates that the numerical added resistance


approaches the experimental data as the mesh is rened. The added
resistance is almost constant and the error is reduced to 1.37%,
when the mesh number is increased to ne case. It is observed
that the comparison error for added resistance in waves is smaller
than that for the calm water resistance. The possible reason for the
higher accuracy of added resistance is that the error in both experimental and numerical results is canceled by subtracting resistance
in calm water from that in waves.
The numerical resistance and ship motions in waves versus
renement ratio hi /h, together with tted curves based on the least
squares root approach as well as tted curves based on theoretical
accuracy Ptheory = 2 are shown in Figs. 7 and 8. Fig. 7 shows the grid
convergence of added resistance. That for ship motions is illustrated
in Fig. 8.
Fig. 7 shows that the PG = 2.43 of added resistance is very close
to the theoretical value. The points are distributed around the tted
curve with standard deviation US less than 1% of the experimental
data, which clearly indicates that the numerical added resistance
is very reliable.
Fig. 8 indicates that the observed order of accuracy (PG = 2.74,
PG = 1.44 for the 0th and 1st order separately) of heave motion is
close to 2, while the PG of the pitch motion is a bit farther from
the theoretical value. The 1st order of pitch converges very slowly.
The possible reason for the poor convergence behavior is that other

sources of numerical errors, such as nonlinear iteration error and


data sampling error, can not be neglected compared with spatial
discretization error when the mesh becomes very ne. In addition,
time discretization error with the same time step may not be the
same for different meshes. The PG for the 0th order pitch is very
large. The 0th order pitch from different meshes is almost constant after the mesh is increased to the medium size. The 0th order
amplitude for motion is a second order quantity in a seakeeping
application. It is quite normal that poor convergence behavior is
observed for those quantities.
The observed order of accuracy PG , the extrapolated solution
0 , standard deviation US % D, and uncertainty UG % D are all listed
in Table 7.
The uncertainty of numerical results in waves is larger than that
in calm water. The large uncertainty of added resistance and the

Table 7
Spatial uncertainty of ship resistance and motions in waves.

Caw
0th 3
0th 5
1st 3
1st 5

PG

0

US % D

UG % D

2.43
2.74
6.97
1.44
0.70

4.4304
6.5443
0.1357
10.141
1.096

0.72%
0.49%
1.11%
3.59%
2.15%

7.62%
4.86%
3.14%
13.63%
6.60%

62

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

Table 8
Ship resistance and motions at  = 0.9171 Lpp predicted with ne mesh and different
time step sizes.

T/t
Caw
e
0th 3
e
0th 5
e
1st 3
e
1st 5
e

Large

Middle

Small

140

200

280

4.664
1.37%
6.369
0.77%
0.1331
2.85%
10.963
5.74%
1.295
4.58%

4.7250
2.70%

4.7451
3.13%

6.3299
1.39%
0.1334
2.63%
12.043
3.54%
1.399
3.10%

6.3131
1.65%
0.1333
2.70%
12.259
5.40%
1.377
1.47%

Extrapolated

4.756/3.034
3.37%
6.298/2.261
1.89%
0.1320/
3.65%
12.320/4.4
5.92%
1.396/
2.87%

Exp

4.601

0.137
11.631
1.357

5.2.2. Uncertainty estimation for time discretization


In Table 6, only a very small change of added resistance is
observed between the ne and the nest mesh. In order to save
CPU time, the ne mesh is used for the time convergence study.
The time convergence study is given in Table 8.
The extrapolated values in Table 8 are estimated with theoretical order of convergence PG = 2. The reason for the use of theoretical
order of convergence is that it is not always possible to obtain
monotonic convergent for all cases in the time convergence study.
The observed order of convergence is also given in Table 8, for
resistance, as well as for 0th-order and 1st-order heave motion.
Monotonic convergence is not observed for other quantities; hence,
order of convergence cannot be given. For those quantities, the
change of numerical results with time step is very small. Thus, it
might be concluded that time discretization error for those quantities is small compared with other sources of error, such as data
sampling error and non-linear iteration error.
5.2.3. Validation of numerical model
For this unsteady simulation, the simulation time is chosen as
60 s, which includes about 45 wave encounter periods (Te ). In order
to reduce the effect of sampling error on the numerical added resistance and ship motions in waves, the data in the last 16 s (about
12 Te ) is used for analysis. In order to show the effect of sampling
error, the numerical results at three different time windows are
given in Table 9. These three different time windows are all chosen
in the last 16 s of simulation. All of them have the same start point.
Table 9 shows that the result from each time window is very similar,
which indicates that the effect of sampling error can be neglected.
The validation of the numerical model in waves is given in
Table 10. The validation uncertainty UVal is calculated according
2 = U 2 + U 2 + U 2 , and U is time step uncertainty.
to UVal
T
D
T
G
Table 9
Numerical results at three different time windows.
1

10

12

Exp

t/T

Caw

4.7493

4.7453

4.7451

4.601

6.3110
0.1330
12.2629
1.377

6.3092
0.1327
12.2523
1.377

6.3131
0.1333
12.259
1.377

6.419
0.137
11.631
1.357

0th 3
0th 5
1st 3
1st 5

Caw
0th 3
0th 5
1st 3
1st 5

E%D

UD % D

UT % D

UG % D

UVal % D

3.13%
1.65%
2.70%
5.40%
3.38%

0.48%
1.26%
0.88%
1.26%
0.88%

0.89%
0.40%
0.17%
6.99%
2.86%

7.62%
4.86%
3.14%
13.63%
6.60%

7.69%
5.04%
3.27%
15.37%
7.25%

6.419

1st heave motion is due to the discrepancy of the result from the
coarse mesh. We believe that numerical results from the present
numerical model are accurate enough. The high level of uncertainty
is due to the design of uncertainty estimation procedure using data
range based on very coarse grid solution.

Time window

Table 10
Validation of numerical model in waves (/Lpp = 0.9171).

It is seen from Table 10 that the validation comparison error


E is smaller than the validation uncertainty Uval for all studied
variables, from which it can be concluded that the numerical added
resistance and ship motions in waves are in agreement with the
experimental data. The results presented in Table 10 indicate that
comparison errors for all quantities are relatively small. Four of
them are smaller than 4%, which can be considered as accurate
enough for engineering application. However, the validation uncertainty is relatively high, ranging from 3% to 15% with three of them
higher than 7%. We believe that such a high level of uncertainty does
not imply that the numerical prediction is unreliable. It indicates
that reliable uncertainty estimation cannot be obtained due to the
limitation of computational resources for such kind of application.
A typical run with a grid containing 2 M cells over one encounter
periods using 200 time steps per encounter periods and 10 nonlinear iterations per time step takes about 400 min with 16 cores,
corresponding to about one-third of a clam water resistance computation. More reliable uncertainty estimation certainly requires
ner mesh, smaller time step, better nonlinear convergence, and
longer simulation time.
With the validation of the numerical model, additional seakeeping simulations are performed with different wavelengths. In order
to reduce the numerical error, the nest mesh case is rened
near free surface for simulation in short waves  < 0.9171 Lpp , with
time step t = Te /140. For the simulations of the longer waves
( > 0.9171 Lpp ), the ne mesh is changed a bit near the free surface
for simulation with time step t = Te /200. The number of grid cells
increase vertically and decrease longitudinally, and the total number of cells is almost the same. All these mesh cases ensure there
are at least 70 cells per wavelength and 11 cells per wave height
near the free surface for the all waves studied here.
5.3. Ship motion in waves
Comparison of numerically predicted ship motions with experimental data are given in Fig. 9. Ship heave and pitch are
non-dimensionalized according to
Y3 =

3
,
a

Y5 =

5
ka

(2)

where  a is wave amplitude, k is wave number. Experimental


result by the authors are marked with triangle ( ) in Fig. 9,
whereas experimental results from Osaka University are marked
with square ( ) [14].
Strip theory proposed by Salvesen et al. [21] has been proved
to be able to predict ship motions with satisfactory accuracy. The
numerical results from CFD are also compared with the results
given by strip theory in Fig. 9. The calculation with strip theory
refers to Guo and Steen [18].
Fig. 9(a) compares heave motion amplitude at the center of
gravity predicted by CFD and strip theory with the experimental data measured by ourselves and Osaka University. The heave
motion amplitude is almost zero at high wave frequency, where
wavelength is less than 0.4 Lpp . At low wave frequency, where
wavelength is longer than 1.5 Lpp , the heave amplitude is close to
the incident wave amplitude. The heave motion changes very fast,
when wavelength is between Lpp and 1.5 Lpp . Both CFD simulation

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

63

Fig. 9. Comparison of RAO of ship motions amongst experiment, strip theory and CFD calculation.

and strip theory could estimate heave motion of KVLCC2 with high
accuracy. Pitch motion is depicted in Fig. 9(b), which shows that
both experiment and CFD result falls on the pitch curve given by
strip theory. In details, the CFD results match experimental data
well except at low frequency. The discrepancy between CFD and
experiment may be due to a slight difference in pitch radius of
gyration (r55 ) between CFD simulation and experiment.
The comparison shows that strip theory also can predict the vertical motion of KVLCC2 with high degree of accuracy. The reason is
that large hydrostatic coefcients C33 , C55 and linear exciting force
give high accuracy of strip theory in the present study. Ship motions
from CFD simulation show good agreement with experimental data
in relatively short waves. In long waves, numerically and experimentally predicted heave also agree well, but the numerical pitch
motion does not match the experimental result well. It should be
noticed that surge motion is free in the measurement performed in
Osaka University, while it is xed in the CFD computation.
The 0th order harmonic pitch and heave predicted with CFD
simulation are compared with measured data in Table 11. The comparison shows that numerical results match the experimental data
well, and that the ship sinkage and trim in waves are independent
of wavelength.
5.4. Added resistance

Added resistance comparison amongst CFD calculation, experimental measurement and theoretical calculation REM is shown
in Fig. 10. Comparison shows that the CFD results have better
agreement with experimental data than that from REM, especially in short waves. REM could predict added resistance in long
waves with acceptable accuracy. However, it underestimates the
ship added resistance in short waves. The reason for small added
resistance in short waves is that the wave diffraction effect is
not well modeled by REM, and the diffraction dominates the
added resistance in short waves. The largest added resistance is
observed around  = 1.1 L, and CFD results near this wavelength
shows good agreement with measurement. REM underestimates
added resistance near this region. The discrepancy of REM prediction is probably due to the nonlinear effects with large ship motion,
which cannot be modeled by REM.
The results show that CFD calculation predicts the added resistance with higher accuracy than REM. However, the cost of CFD
simulation in small waves is very high, due to the small mesh
size and time step (/x = 70, Te /t = 140) to capture the waves
well. Fujii and Takahashi [23] divided added resistance into two
parts: one part is due to ship motions and the other part is due
to wave reection. Since the ship motions are very small in short
waves, it will be economical to x the hull in predicting added
resistance, if the contribution of ship motions to added resistance
is also negligible. The contribution from ship motions to added

Reliable evaluation of added resistance is important for ship


design aiming at sustaining ship service speed and saving propulsion power. The added resistance calculated by CFD simulation
is compared with that from the radiated energy method (REM)
proposed by Gerritsma and Beukelman [22]. Knowledge of ship
motions and hydrodynamic characteristics required to calculate
added resistance with the REM, is provided by strip theory in this
study. The detailed calculation procedure of REM is given in Guo
and Steen [18].

Table 11
The 0th harmonic pitch and heave in waves.
Lpp /

3 (CFD)

3 (EFD)

5 (CFD)

5 (EFD)

2.453
1.571
1.224
1.090
1.667
0.909
0.625

6.322
6.359
6.333
6.385
6.293
6.201
5.865

6.223
6.409
6.375
6.419
5.431
5.517
6.051

0.125
0.121
0.121
0.134
0.120
0.128
0.127

0.122
0.123
0.125
0.137
0.124
0.137
0.145

Fig. 10. Comparison of added resistance from REM and CFD calculation.

64

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667


Table 12
Added resistance components.
Lpp /

Caw

1.090
1.571
1.224
2.453
1.667
0.909
0.625

4.3032
2.2323
3.1685
2.3719
2.5023
6.8121
1.9755

Caw

0.3068
0.2935
0.3148
0.3251
0.3218
0.2795
0.1051

Caw

4.610
2.5258
2.8537
2.6970
2.8241
7.0916
2.0806

5.6. Wake ow

Fig. 11. Added resistances from free and xed models.

resistance is studied here. The numerical model is xed at the


design load condition in short waves, and added resistances from
free and xed model are compared with experimental results in
Fig. 11. The comparison shows that numerical added resistance
from xed and free hull are almost equal in very short waves, i.e.
Lpp / > 1.57 ( < 0.63 Lpp ), and the difference becomes larger as the
wavelength increases. The results show that the contribution from
ship motion to added resistance can be safely neglected when the
wavelength is less than 0.63 Lpp .
5.5. Analysis of resistance components in waves
The resistance in waves is composed of frictional and pressure
resistance, and both of them are recorded in the numerical simulation. The effect of waves on the pressure and frictional resistance
is analyzed here. The time traces of total resistance, pressure and
frictional resistance of one wave case are shown in Fig. 12.
In Fig. 12, Te is the encounter wave period. The rst order amplitude of frictional force in time domain is very small, compared
with that of the pressure resistance. The results show that pressure
resistance dominates the 1st order wave resistance.
Added resistance can also be divided into two parts, one of them
is due to the change of pressure resistance (Caw P ) and the other
part is due to the change of frictional resistance (Caw f ). Caw T represents ship total added resistance. The corresponding results are
given in Table 12. Table 12 shows that the added resistance due
to change of frictional force is very small and almost constant for
short waves.

Fig. 12. Comparison of total resistance (dash line), pressure resistance (dash dotted
line), and frictional resistance (solid line) at Lpp / = 1.090.

Another attractive advantage of CFD is that it can simulate complicated interaction between the boundary layer and incoming
waves and predict the wake ow, while potential ow codes cannot. This is very important for both propeller design and ship
hull optimization [24]. It is a pity that there is no experimental
data available about wake ow of free KVLCC2 in waves. Thus the
numerically predicted wake ow in calm water is compared with
experimental data in Fig. 13 to validate the numerical model for
wake ow. In order to reduce the effect of mesh on wake ow, both
steady wake and unsteady wake is computed with the ne mesh in
Table 6, which is found to be ne enough for the unsteady simulation. The comparison shows that the numerical model predicts the
wake ow with good accuracy.
What is really important for propeller design is the nominal
wake ow at the propeller disc, which is analyzed here. The nominal wake ow at the propeller disc in calm water is shown in
Fig. 14. The velocity is presented in a ship-xed reference system.
The gure shows the iso-contours of the three Cartesian velocity components non-dimensionalized by ship forward speed U
(U = 1.047 m/s). D = 0.17 m is the propeller diameter.
The wake ow is expected to change due to waves and ship
motions. The wave-induced particle velocity axially and vertically
changes wake ow [25]. Moreover, wave-induced ship motions also
have an important effect on wake ow. The numerical calculation
for wake ow is performed with wavelength /Lpp = 0.9171, and

Fig. 13. Wake eld at y/Lpp = 0.9825 (propeller plane) from CFD compared with EFD
in calm water.

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

65

Fig. 14. The nominal wake ow in calm water.

Fig. 15. Mean wake ow in waves.

wave height H = 0.138 m, which is the measured wave amplitude


in a model test with nominal wave height H = 0.15 m.
5.6.1. The 0th order harmonic amplitude of wake at propeller disc
The 0th order harmonic amplitude of wake ow is the mean
value in waves, which is illustrated in Fig. 15. The wake velocity
in three different directions is non-dimensionalized with ship forward speed in the same way as in Fig. 14. Comparison with the
wake ow in calm water (Fig. 14) indicates that the area of low axial
velocity becomes smaller in waves and the variation of axial velocity at the top part becomes large. The mean vertical wake velocity
in waves becomes negative over the whole disc.
In order to show the change of wake ow induced by waves
quantitatively, the difference of mean wake velocity along three
directions in waves (Ui , i = 1, 2, 3) and that in calm water (Ui-steady ,
i = 1, 2, 3) as a fraction of ship forward speed U (U = 1.047 m/s) is
shown in Fig. 16. In Fig. 16(a), the largest increase of the mean axial

velocity in waves relative to that in calm water is about 35% of ship


forward speed. The change in axial velocity is mainly due to the fact
that the bilge vortex moves up and down in waves. The region with
low axial velocity in calm water indicating the location of the bilge
vortex disappears in waves, resulting in a wake with higher mean
axial velocity. Fig. 16(b) shows the change of the mean transversal velocity in waves relative to that in calm water. The transversal
velocity is zero at the center plane due to the symmetry condition adopted in the CFD simulation. The mean transversal velocity
shows that the largest change is about 8% of the ship forward speed.
In Fig. 16(c), the change of the mean vertical velocity relative to that
in calm water is negative at the whole propeller disc. The maximum
mean velocity increase is about 25% of ship forwards speed.
5.6.2. The 1st order harmonic amplitude of wake at propeller disc
In Fig. 17(a), the 1st order axial velocity shows very large amplitude at the lower part of the propeller disc, where the maximum

Fig. 16. Difference of the 0th harmonic velocities and steady ow velocities (Ui Ui-steady )/U along three directions at propeller disc.

66

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667

Fig. 17. The 1st harmonic amplitude of wake ow at propeller disc.

value is about 35% of ship forward speed. This value superimposed


on the 0th harmonic amplitude will induce a large axial velocity and
low pressure at the propeller disc. In Fig. 17(b), the 1st order harmonic amplitude of transversal velocity is zero at the center plane
due to the symmetry boundary conditon. The value increases away
from the center plane. Compared with 1st order axial and vertical
velocity, the 1st transversal velocity is smaller. This is not surprising, since purely head sea is considered here. Fig. 17(c) shows that
the maximum vertical rst harmonic amplitude is about 22% of ship
forward speed.

6. Conclusions
Ship motions and added resistance of KVLCC2 advancing in head
waves have been calculated with unsteady RANS code. A systematic
verication and validation analysis of numerical result in both calm
water and waves is performed to show the reliability and accuracy
of the numerical model in terms of the studied sea-keeping problems. Numerical added resistance and ship motions in waves are
veried to be convergent through time step and grid spacing renement studies. The uncertainty is less than 6% for all the studied
variables in calm water.
CFD prediction of ship motions and added resistance matches
the experimental data well for different wavelengths, down to
the shortest wavelength of Lpp / = 2.453. The comparison of ship
motions show that CFD has no advantage over strip theory in predicting ship motion. However CFD predicts added resistance with
higher accuracy than the theoretical calculation REM based on
strip theory, especially in short waves. Although CFD could accurately predict added resistance in short waves, the computational
cost is still very high due to the requirement of very small mesh
size. To save CPU time and also ensure the numerical accuracy, it is
recommended to estimate the added resistance by xing the ship
model when the wavelength is less than 0.63 Lpp .
Pressure resistance dominates the rst order ship resistance in
waves, and the rst order frictional resistance is small. The added
resistance mainly comes from the change of pressure force, and the
added frictional resistance is small and almost constant in short
waves.
Both the 0th and the 1st order axial velocities induced by waves
at the propeller disc predicted by numerical simulation are very
large, and the maximum values are approximately 35% of ship forward speed. Wave-induced mean transversal velocity is quite small
except for the local region, where maximum velocity is about 8% of
the ship forward speed. The 1st order transversal velocity increases
away from the center plane, but is still small. The mean vertical
velocity in waves is negative in the whole propeller disc. Wave
induced maximum 1st-order vertical velocity is about 22% of the
ship forward speed. The numerical simulation indicates that the

wake ow is strongly affected by the incoming waves. The effect of


waves on the wake is surprisingly strong, and more studies should
be performed to verify our ndings. Also, the effect of the change of
wake due to waves on the propeller performance should be investigated.
Acknowledgements
The work is part of the research project SeaPro, sponsored by
Rolls-Royce Marine and the Research Council of Norway.
References
[1] Choi J, Yoon SB. Numerical simulations using momentum source wave-maker
applied to RANS equation model. Coastal Engineering 2009;56:104360.
[2] Zhang ZR, Liu H, Zhu SP, Zhao F. Application of CFD in ship engineering design
practice and ship hydrodynamics. In: Conference of global Chinese scholars on
hydrodynamics. 2006. p. 31522.
[3] Larsson L, Stern F, Bertram V. Benchmarking of computational uid dynamics for ship ows: the Gothenburg 2000 workshop. Journal of Ship Research
2003;47(1):6381.
[4] Wilson R, Paterson E, Stern F. Unsteady RANS CFD for naval combatants in
waves. In: Proceedings of the 22nd symposium on naval hydrodynamics. 1998.
[5] Sato Y, Miyata H, Sato T. CFD simulation of 3-dimensional motion of a ship in
waves: application to an advancing ship in regular heading waves. Journal of
Marine Science and Technology 1999;4:10816.
[6] Rhee SH, Stern F. Unsteady RANS method for surface ship boundary layer and
wake and wave eld. International Journal for Numerical Methods in Fluids
2001;37:44578.
[7] Hideo O, Hideaki M, Orych M. Evaluation of added resistance in regular incident
waves by computational uid dynamics motion simulation using an overlapping grid system. Journal of Marine Science and Technology 2003;8:4760.
[8] Hideo O. Comparison of CFD simulations with experimental data for a tanker
model advancing in waves. International Journal of Naval Architecture and
Ocean Engineering 2011;3:18.
[9] Hino T.Proceedings of CFD workshop 2005 Tokyo. 2005.
[10] Deng GB, Queutey P, Visonneau M. RANS prediction of the KVLCC2 tanker in
head waves. Journal of Hydrodynamics 2010;22(Supplement (5)):47681.
[11] Weymouth GD, Wilson RV, Stern F. RANS computational uid dynamics predictions of pitch and heave ship motion in head seas. Journal of Ship Research
2005;49(2):8097.
[12] Stern F, Wilson RV, Coleman HW, Parson EG. Comprehensive approach to verication and validation of CFD simulations-part 1: methodology and procedures.
Journal of Fluids Engineering 2001;123:793802.
[13] Deng GB, Queutey P, Visonneau M. Seakeeping prediction for a container ship
with RANS computation. In: The 22nd Chinese conference in hydrodynamics.
2009.
[14] Larsson L, Stern F, Visonneau M.Proceedings of a workshop on numerical ship
hydrodynamics. 2010.
[15] Eca L, Hoekstra M. An evaluation of verication procedures for CFD applications.
In: Proceedings of the 24th symposium on naval hydrodynamics. 2002.
[16] NUMECA. Theoretical manual FINETM /Marine v2.2; 2010. http://www.ecnantes.fr.
[17] Azcueta R. Steady and unsteady RANS simulations for littoral combat ship.
In: Proceedings of the 25th symposium on naval hydrodynamics. Canada: St.,
Johns Newfoundland and Labrador; 2004.
[18] Guo B, Steen S. Experiment on added resistance in short waves. In: Proceedings
of the 28th symposium on naval hydrodynamics. Pasadena, CA, USA: California
Institute of Technology; 2010.
[19] Roache PJ. Quantication of uncertainty in computation uid dynamics. Annual
Review of Fluid Mechanics 1997;29:12360.

B.J. Guo et al. / Applied Ocean Research 35 (2012) 5667


[20] Guo B, Steen S. Computational study on ship resistance and ow eld of KVLCC2.
In: Proceedings of Gothenburg 2010: a workshop on numerical ship hydrodynamics. 2010. p. 58593.
[21] Salvesen N, Tuck EO, Faltinsen OM. Ship Motion and Sea Load. The Society of
Naval Architects and Marine Engineers 1970;78(6):130.
[22] Gerritsma J, Beukelman W. Analysis of the resistance increase in waves of a fast
cargo ship. International Shipbuilding Progress 1972;19(217):28592.
[23] Fujii H, Takahashi T. Experimental study on the resistance increase of a ship in
regular oblique waves. In: Proceeding of 14th ITTC, vol. 4. 1975. p. 35160.

67

[24] Carrica PM, Wilson RV, Stern F. Unsteady RANS simulation of the
ship forward speed diffraction problem. Computers & Fluids 2006;35:
54570.
[25] Longo J, Shao J, Irvine M, Stern F. Phase-averaged PIV for the nominal
wake of a surface ship in regular head waves. Journal of Fluids Engineering
2007;129:52440.
[26] Eca L, Vaz G, Hoekstra M. Code verication, solution verication and validation
in RANS solvers. In: Proceedings of 29th international conference on ocean,
offshore and arctic engineering. 2010. OMAE 2010-20338.

Você também pode gostar