You are on page 1of 9

17 6 Vol.17 No.

6
2013 6 Journal of Ship Mechanics Jun. 2013

Article ID 1007-7294 2013 06-0616-09

Investigation on Some Factors Effecting Ship Resistance


Calculation with CFD Code FLUENT

DENG Rui, HUANG De-bo, ZHOU Guang-li, SUN Hua-wei


(Multihull Ship Technology, Key Laboratory of Fundamental Science for National Defence,
Harbin Engineering University, Harbin 150001, China)

Abstract: Some of the viscous Computational Fluid Dynamics methods, when applied to hulls ap-
pear not easy in giving consistent and agreeable results in resistance and other hydrodynamic per-
formance. The authors attempt to find some practical ways by investigating the influential factors such
as mesh gridding, choice of discretization schemes and turbulence models which affect the calculation
results. Through calculation and analysis, specifically with the CFD code FLUENT, an alternative
set of computation parameters for engineering application is suggested. And, the results obtained by
applying the proposed method in this paper to mono-hull show better agreement with corresponding
model tests.
Key words: CFD; resistance; mesh generation; discretization scheme; turbulence model
CLC number: U661.31+1 Document code: A doi: 10.3969/j.issn.1007-7294.2013.06.004

1 Introduction

The flow field around the ship sailing on the interface of air and water is a complicated
three dimensional one. The transition and the separation of the flow, especially when the ef-
fect of the complex profile of the hull are taken into account, make it difficult to simulate the
flow field[1]. Ignoring the viscosity, and treating the flow field as a potential flow can obtain sat-
isfactory results in some problems such as wave making and the flow far away from the hull,
which have indistinctive relationship with the viscosity. Potential flow theory is a useful tool,
but CFD method is more powerful in the research about the details of the flow field [2]. Though
a lot of new measures [3-4] are adopted in the CFD method, some of the difficult problems [5-6]
can be solved by it, and it is successfully used[7], the simulation result is affected by the mesh
generation, the choice of the turbulence models and the discretization schemes.
The research about the mesh generation is focused on the uncertainty [8-9] caused by it,
and some of the researchers have made some investigations on the verification and the valida-
tion[10-12], but few of them contributed any suggestion that can be used in the engineering[13]. The
Received date: 2012-10-09
Foundation item: Supported by the National Natural Science Foundation of China (Grant No.51209048)Research on the
Distribution of Ship Energy Consumption and Energy Efficiency; Fundamental Research Funds for the
Central Universities (Grant No.HEUCF120103; HEUCFR1204; HEUCF110104; HEUCF100106)
Biography: DENG Rui(1981-), male, Ph.D., lecturer of Harbin Engineering University, E-mail: dengrui@hrbeu.edu.cn;
HUANG De-bo(1943-), male, professor/tutor.
6 DENG Rui et al: Investigation on Some Factors Effecting 617

applicability [16] of the turbulence models [14-15] used are seldom investigated. Considering all of
the factors and offering some suggestion[17] that could be used in engineering application is nec-
essary, otherwise, the reliability and the feasibility of CFD method are doubtful [18-19].
In the present work, combined with the comparison with model test data, some factors that
affect the calculation of the resistance, such as the mesh generation, the choice of the dis -
cretization schemes and turbulence models in the simulation of a mono-hull vessel are inves -
tigated, and some suggestions are put forward.

2 Ship model test

The model test is performed at the towing tank in Harbin Engineering University. The
length, width and water depth of the towing tank are respectively 108.0, 7.0 and 3.5 m. The
carriage is controlled by the computer, and the speed of it is between 0.100-6.500 m/s. In this
paper, the model used is of a mono-hull, and the length, width and depth of the full scale ship
are 65.0, 8.3 and 2.7 m. The scaling factor is 1/20, so the length of the water line of the test
mode is 2.8 m. The model is shown in Fig.1. The surging and rolling are restrained during the
experiment.

3 Preliminary numerical simulation

In this study, model construction and mesh generation are performed by the pre-process -
ing software GAMBIT, and 0.92 million non-structural grids are used to discrete the flow field.
The boundary conditions are non-slip wall, velocity inlet, free stream outlet, vertical symmet -
rical face and free surface. The VOF method is adopted to calculate the free surface, and the
k- turbulence model is used. The method used for the coupling of the pressure and velocity
is SIMPLE, and the discretization scheme of the convection is the first order upwind difference.
The numerical ship model is shown in Fig.2.

Fig.1 Experimental ship model Fig.2 Numerical ship model


The velocities in the calculations are 0.854, 1.281, 1.601, 1.922, 2.349, 2.776, 3.203,
3.630 m/s, respectively, and the comparison between numerical result and the experimental
data will be shown in Fig.9, and the numerical result agrees with the experimental data gen -
erally at high speed, and it suggests the feasibility of the simulation. There is a considerable
discrepancy between the calculation result and the experimental data, which may be caused by
the method of mesh generation, size and quantity of the grid, choice of discretization scheme
618 17 6

and turbulence model. All those factors should be discussed, attempting to get some sugges -
tions that could be useful in practical application.

4 Discussion of the mesh generation

Mesh generation research includes three parts: grid size on the hull surface, the height of
the first layer of the grid and the distribution ratio of the grid.
For example, the velocity is 3.630 m/s, the smallest size of the grid 6.3 mm is used to
mesh the hull surface, and the corresponding grid number generated for the flow field is with -
in the reach of computer used by us. The grid size on the hull surface is increased by a ratio
2 and considering the relationship of the nodes when structural grid is used, some of the
grid sizes are changed, so the grid sizes are, 9.5 mm, 12.7 mm, 19.0 mm, 25.3 mm and 40.0
mm respectively, the corresponding percentage of the water line length are 0.23%, 0.34%,
0.45%, 0.68%, 0.90% and 1.43%, respectively. The largest size on the ship surface is 40 mm,
and any larger size is avoided to keep a better fitted ship form. The quantities of the grid in
these cases are correspondingly 1.7, 0.6, 0.2, 0.07, 0.03 million and 9 thousand respectively.
The grid distributions of the free surface in each case are shown in Fig.3.

(a) l=6.3 mm (b) l=9.5 mm

(c) l=12.7 mm (d) l=19.0 mm

(e) l=25.3 mm (f) l=40.0 mm


Fig.3 Grid distribution of the free surface
The resistances which are changing with the grid sizes in different cases are compared in
Fig.4.
6 DENG Rui et al: Investigation on Some Factors Effecting 619

As above, the resistance is affected by


the grid size on the hull surface. The details
of the flow field can not be calculated well
in the case that the hull surface is meshed
by the large size grid, and the calculation of
the velocity and pressure around the vessel
is not accurate enough. The accuracy of the
numerical result is improved while decreas- Fig.4 Comparison of the resistance with different
ing the size of the grid on the hull surface, grid sizes on hull surface
but the huge number of grid caused by the small grid size will produce a considerable error
that affect the calculation result, so the accuracy of the result in the case of the grid size of
6.3 mm is worse than that of 9.5 mm. There is a balance between the quantity of the grid and
the accuracy of the result during the mesh generation, and the suggestion is that for the resis -
tance calculation of the mono-hull vessel, the proper grid size on the hull surface may be
about 0.34% of the water line length when taking the two factors above into account.
The flow field around the hull surface is complex, and the gradient of the velocity and the
turbulence intensity is greater than anywhere also, so much more grid is used around and near
the hull to refine the mesh in order to get an accurate result. There are two important factors
for mesh generation around the hull: the factor y which means the height of the first layer of
the grid and the factor r* which means the grid distribution ratio. These two factors relate to
each other, because the grid distribution around the hull is in a way of geometric progression
usually, y indicates the first item of it, and r* indicates the scale factor.
The height of the first layer of the grid is in concert with a non-dimensional factor y + ,
and the formulation used to calculate y+ is shown as follows:
y 0.9
y+=0.172 Re
L
where y is the height of the first layer of the
grid, L is the length of the waterline, Re is
the Reynolds number relative to the L. The
factor y + should meet the condition 11.63 ~
30y+200~500 under usual circumstance.
In the condition that the grid size on the hull
surface is 9.5 mm and the calculation speed Fig.5 Comparison of the results of resistance with
is 3.63 m/s. Four different heights of the first grids of different heights of first layer
layer are chosen, they are y=0.10, 0.53, 1.06 and 1.77 mm, and the factor y + are 11.5, 60,
120 and 200, respectively. Comparison of the results of resistance in these conditions is shown
in Fig.5.
It can be seen from the comparison that there is inapparent difference between each re-
sult in the cases of different condition. The maximal discrepancy between numerical and exper-
620 17 6

imental data is about 4.3%. For the mono-hull ship researched in this paper, when the size of
the mesh grid used on the hull surface is about 0.34% of the water line length and structural
grid is used for the flow field, resistance is not affected seriously by the alteration of the height
of the first grid layer in the normal range of the factor y+ used in the CFD calculation. Taking
the quantity of the grid into account, 1.77 mm can be used for the parameter y which means
the height of the first grid layer.
The grid distribution ratio r*=1.091, 1.189
and 1.414 are adopted for the structural grid
generation of the flow field on the basis of
the research above, and the velocity for the
numerical ship model is 3.63 m/s, the com-
parison of the resistance results under these
conditions is shown in Fig.6. Fig.6 Comparison of the resistance of the grid in the
As it is shown in the comparison, the case of different grid distribution ratio
resistance is not changed conspicuously with the alteration of the parameter r* in each condi-
tion after the size of the grid on the hull surface and the height of the first grid layer are de -
termined. In the case that r* equals to 1.189, the resistance appears a oscillatory convergence
trend with time, and which is different from the others. Comparison of the numerical result and
the experimental data shows a maximum discrepancy 6.7%. The comparison indicates that the
resistance is affected little by the grid distribution ratio r* when the mesh generation scheme
suggested in this paper is used, but if the condition permits, a smaller r* is suggested.
In this paper, the most remarkable impact of the resistance is caused by the grid size on
the hull surface, and a proper accuracy of the result can be obtained while the grid size is
about 0.34% of the water line length. The resistance is affected little by the height of the first
grid layer and the grid distribution ratio, and a y + within the normal required range of CFD
along with a smaller ratio are suited to the application of engineering.

5 Discussion of the discretization schemes

There are some discretization schemes in the Finite Volume Method with different accu -
racy in solving the convection term of the N-S equation. Although the convection term is only
a first-order derivative, but it is the hardest part to solve because of the intense directionality.
The choice for a discretization scheme has a great impact on the accuracy of the result, stabili-
ty of the computation process and the condition needed for the calculation, so a proper dis -
cretization scheme can accelerate the calculation and improve the accuracy. In the present
time, the discretization schemes used are the first, second and third-order upwind schemes, fi-
nite analysis and central difference scheme. Some research points out there is no way to con -
firm the best discretization scheme, but the suggested one is the third-order upwind scheme [1].
6 DENG Rui et al: Investigation on Some Factors Effecting 621

The high order discretization scheme is the


method with conditional stability, and it is
hard to obtain a convergent result with high
order discretization scheme in the case that
the mesh quality is not good enough.
The grid size on the hull surface is
0.34% of the length on water line, the non-
dimensional parameter y + is 200, the grid Fig.7 Comparison of the results of different
distribution ratio r* is about 1.091 and the discretization scheme
calculation velocity is 3.63 m/s, in this condition, the resistances calculated with first-order,
second-order upwind, QUICK and third-order MUSCAL schemes are compared in Fig.7.
It is shown in Fig.8 that the asymptotic rate of convergence and the stability of the four
discretization schemes are similar. The low-order discretization scheme like first-order upwind
scheme costs less time for calculation and shows no oscillation of the result, which means a
absolute stability in any condition, but appears poor accuracy. Some high-order discretization
schemes like second-order upwind scheme possess the advantages of the low-order scheme
and show improved accuracy. No marked improvement is found in accuracy and asymptotic rate
of convergence when the higher-order scheme like QUICK is adopted, as alleged in theory,
and it costs more computation time. There is a special requirement about the mesh quality and
the computation time while the third-order scheme like MUSCAL is used, but no remarkable
difference between it and the second-order one in asymptotic rate of convergence and accura -
cy. It is difficult to get a high quality mesh due to the complexity of the ship form, and the
flow field is also complex, so the second-order discretization scheme is suggested in this con -
dition for a commendable result in accuracy, stability and convergence rate.

6 Discussion of the turbulence models

On the basis of the suggestion for the resistance calculation of the mono -hull vessel
above, the resistances are calculated and compared. The turbulence models adopted are Stan -
dard k-, Realizable k-, RNG k-, Stan-
dard k -, SST k -, Reynolds Stress and
Large Eddy Simulation models, respectively.
The calculation diverges when Standard k-
model and Reynolds Stress model are adopt-
ed, and the results with the rest turbulence
models are compared in Fig.8.
It is difficult to obtain a convergent re- Fig.8 Comparison of the resistance results using
sult when the Reynolds Stress model is different viscous models
622 17 6

adopted due to the poor computational sta-


bility caused by the hypothesis and simplifi-
cation in the modeling process of the equa-
tions. Many applications have shown that the
differential forms of Reynolds Stress models
are not much better than the two -equation
turbulence model. Most of the parameters in
the Standard k - model are obtained by Fig.9 Ship resistance comparison between the
classic experiments, and they are not fit for calculation result and experimental data
application, so a convergent result can not be obtained with them. It is shown in Fig.9 that the
results with the rest turbulence models have similar computational stability and rate of con -
vergence. The k- SST turbulence model integrates the advantages of k- and k- models,
and the curved form of the wall is also taken into account, in addition, no more memory or com-
putation time is needed when it is adopted compared with the other two -equation models.
There is a harsh requirement in mesh generation for the application of Large Eddy Simulation,
which means additional cost in computer memory and computation time, so it seems be seldom
adopted in engineering application. Comparing these turbulence models, we suggest, for ship
resistance prediction, k- SST turbulence model is applied.

7 Ship resistance calculation

Through the above discussion, for the mono-hull ship resistance calculation, the follow -
ing method is determined, in which the grid size on the hull surface is 0.34% of the water line
length, the non-dimensional parameter y + is 200, the grid distribution ratio r* is about 1.091,
and the second-order upwind scheme and the k- SST turbulence model are adopted. The
ship resistances of the velocity 0.854, 1.281, 1.601, 1.922, 2.349, 2.776, 3.203 and 3.630 m/s
are calculated respectively by the method. The result of the preliminary numerical simulation
and the one obtained by the method are compared with the experimental data, as it is shown in
Fig.9.
Where result 1 is the previous result, and the result 2 is the one with the method sug-
gested here. It is shown in Fig.9 that the result 2 agrees with the experimental data much bet -
ter than that of result 1, and it indicates the method can be applied to the simulation of the
viscous flow field of the mono-hull ship and the calculation of the resistance. The effect of the
mesh generation, choice of the discretization scheme and turbulence model on the calculation
of the hydrodynamic performance are considered in the method, and compared with result 1,
the mean discrepancy of result 2 is reduced. The ship motions are restrained in the simulation,
which means the wetted area is not changed with ship speed, different from that in experiment,
so it may be one of the main sources in discrepancy between the numerical result and the test
data, and it is also the point for further research.
6 DENG Rui et al: Investigation on Some Factors Effecting 623

8 Conclusions

The simulation of the viscous flow field around a mono-hull vessel is investigated, sys -
tematically somewhat, in mesh generation, choice of the discretization scheme and turbulence
model, and a method for resistance calculation is suggested. The resistance obtained with and
without this method are compared with the test data, and some conclusions are drawn:
(1) The factor that affects the numerical result of resistance most is the grid size on the
hull surface for the mono-hull vessel in this paper, and the proper grid size is 0.34% of the
water line length, the accuracy of the result can thereby be improved and the cost of the cal -
culation is not increased.
(2) The alteration of the height of the first grid layer does not affect the resistance obvi -
ously when the grid size on the hull surface is about 0.34% of the waterline length, and the
result does not change much while the non-dimensional parameter y + is altered between 11.5
and 200.
(3) The effect of the grid distribution ratio to the resistance is not significant if structural
grid is used for the flow field and the techniques in (1) and (2) are adopted, but a smaller ra -
tio r* is suggested.
(4) A proper accuracy of the simulation can be obtained for the mono-hull vessel in the
paper, if the calculation is carried out with the second-order upwind discretization scheme, and
the accuracy of the calculation is similar to those from QUICK or third -order MUSCAL
scheme.
(5) The k- SST turbulence model is suggested for the numerical calculation of the mono-
hull vessel on the basis of the comparison of several turbulence models.
(6) The accuracy of the resistance calculated by the method suggested is improved, and
the mean discrepancy is reduced.
A further research of the factors that affect the numerical result is desired, and the focus
of it is to derive some suggestion of the parameters for engineering application in order to re -
duce the cost of the numerical simulation and improve the accuracy. The work in this paper is
on a specified ship, the comprehensive suitability to other forms is certainly to be confirmed
with more simulations, some of our simulations for other ships show the same tendency which
will be reported later.

References

[1] Gao Qiuxin. A survey on computational naval hydrodynamics[J]. Journal of Ship Mechanics, 1998, 3(4): 74-78.
[2] Cai Rongquan. Current situation of ship CFD and our knowledge[J]. Ship & Boat, 2002, 1: 29-36.
[3] De Palma P, de Tullio M D, Pascazio G, Napolitano M. An immersed-boundary method for compressible viscous flows[J].
Comput. Fluids, 2006, 7(35): 693-702.
[4] Zhang S C Smoothed Particle Hydrodynamics (SPH) method (A review)[J] Chinese Journal of Computational Physics,
1996, 13(4): 385-397.
624 17 6

[5] Carrica P M, Wilson R V, Noack R W, Stern F. Ship motions using single-phase level set with dynamic overset grids[J].
Computers & Fluids, 2007, 36: 1415-1433.
[6] Qun Zhao, Steve Armfield, Katsutoshi Tanimoto. Numerical simulation of breaking waves by a multi-scale turbulence model
[J]. Coastal Engineering, 2004, 1(51): 53-80.
[7] ITTC. Proceedings of the 24th ITTC-V[C]. 2005, 1.
[8] ITTC QM Procedure 7.5- 03- 01- 01[S]. 2002.
[9] ITTC QM Procedure 7.5- 03- 02- 01[S]. 2002.
[10] Weymouth G D, Wilson R V, Stern F. RANS CFD predictions of pitch and heave ship motions in head seas[C]// Proc. 8th
Int. Conf. Numerical Ship Hydrodynamics. Busan, Korea, 2003.
[11] Zhang Zhirong. Verification and validation for RANS simulation of KCS container ship without/with propeller[C]// 9th In-
ternational Conference on Hydrodynamics. Shanghai, China.
[12] Zou Lu, Larsson L, Orych M. Verification and validation of CFD perdictions for a manoeuvring tanker[C]// 9th Interna-
tional Conference on Hydrodynamics. Shanghai, China.
[13] Deng Rui, Huang Debo, Li Jia, Cheng Xuankai, Yu Lei. Discussion of grid generation for catamaran resistance calcula-
tion[J]. Journal of Marine Science and Application, 2010, 9: 187-191.
[14] Yakhot V, Orszag S A. Renormalization group analysis of turbulence I: Basic theory[J]. Journal of Scientific Computing,
1986, 1: 3-11.
[15] Jones W P, Launder B E. The calculation of low-Reynolds-number phenomena with a two-equation model of turbulence
[J]. International Journal of Heat and Mass Transfer, 1973, 16(6): 1119-1130.
[16] Zang Zhirong, Zhao Feng, Li Baiqi. Application of k- turbulence model to computation of viscous flow field around a
ship[J]. Journal of Ship Mechanics, 2003, 7(1): 33-37.
[17] Deng Rui, Huang Debo, Yu Lei, Cheng Xuankai, Liang Hongguang. Research on factors of a flow field affecting catama-
ran resistance calculation[J]. Journal of Harbin Engineering University, 2011, 32(2): 141-146.
[18] Chen Qingyan, Zhai Zhiqiang. How realistic is CFD as a toll for indoor environment design and studies without experi-
ment[C]// The 4th International Symposium on HVAC. Beijing, 2003: 62-77.
[19] Fred Stern, Robert V W, Hugh W C, et al. Verification and validation of CFD simulations[R]. IIHR(Iowa Insititute of Hy-
draulic Research) Report, No.407, University of Iowa, 1999.

CFD


150001

:
(FLUENT )

: CFD
: U661.31+1 : A
1981-
1943- /
1969- /
1985-

You might also like