Current Affairs: Model Tests, Semi-Empirical Predictions ... … · Proceedings of the 28th International Conference on Ocean, Offshore and Arctic Engineering . OMAE2009 . 31 May

Proceedings of the 28th International Conference on Ocean, Offshore and Arctic Engineering OMAE2009

31 May – 05 June, 2009, Honolulu, Hawaii, USA

OMAE2009-80216

CURRENT AFFAIRS: MODEL TESTS, SEMI-EMPIRICAL PREDICTIONS AND CFD COMPUTATIONS FOR CURRENT COEFFICIENTS OF SEMI-SUBMERSIBLES

Guilherme Vaz MARIN

Wageningen, The Netherlands

Olaf J. Waals MARIN

Wageningen, The Netherlands

Harald Ottens Heerema Marine Contractors

Leiden, The Netherlands

Fahd Fathi GustoMSC

Schiedam, The Netherlands

Tim Le Souëf INTECSEA

Perth, Australia

Kwong Kiu INTECSEA

Perth, Australia

ABSTRACT Current loads on stationary vessels have been investigated

as part of the Current Affairs Joint Industry Project (JIP). Model-tests, semi-empirical models and CFD methods were used to predict these loads. This paper examines one configuration out of the eight tested in the JIP; an idealized semi-submersible consisting of two square rounded-corner columns connected with a pontoon. The model experiments, empirical model predictions and CFD results are presented and discussed. ‘Blind’ and ‘Improved’ CFD computations (with and without knowledge of the experimental results) have been carried out by the JIP participants. Comparisons between these results are made, deviations from the experimental data are quantified and conclusions are drawn. Two key issues for modeling accuracy are identified and discussed; the location of the transition to turbulent flow and the control of the numerical errors.

INTRODUCTION Computational Fluid Dynamics (CFD) is now commonly

used for analyzing flows around offshore structures. The availability of user-friendly commercial CFD packages and the low price of hardware have catalyzed its use significantly. The cost and efficiency advantages of CFD compared with tank testing and semi-empirical models are familiar arguments. However, questions still remain regarding the adequacy of CFD for accurately predicting current loads on stationary offshore

structures such as semi-submersibles and mono-hulls. The Current Affairs JIP was initiated in 2007 to consider this topic and to develop tools and guidelines to assist engineers in the assessment of current effects in the different design stages. Successive design stages require differing levels of accuracy in the determination of current loads. In the initial design, semi-empirical methods may be appropriate and in subsequent stages CFD analysis and model testing may be justified. The JIP aims to investigate and quantify the possibilities and limitations of these complementary methods.

Several commercial and in-house CFD codes were used to predict current forces on a range of hull shapes. A semi-empirical tool was developed as part of this JIP. It estimates the current loads on a structure by considering experimental data for its constituent parts in concert with a shielding model. Tank tests were carried out to allow validation of the semi-empirical and CFD results. CFD best-practice-guidelines will be created based on these comparisons and further work.

All computations were carried out at model-scale Reynolds numbers to allow validation with the model-scale experiments. Eight vessel shapes were studied and tested. The configurations ranged from a basic single column model to a model with four columns and pontoons (a full semi-submersible). This paper focuses on a model shape with two rounded-corner square columns and one pontoon with sharp corners (see Figure 1). As agreed among the participants no quantitative values are presented.

1 Copyright © 2009 by ASME

Figure 1: Tank Test Model with Two Rounded-Corner

Square Columns and a Pontoon.

MODEL TESTS

The objective of the model testing was to provide a consistent dataset of current loads for inclusion in the semi-empirical model and to validate the CFD calculations. The model was equipped with force measurement frames to measure all forces and moments in the horizontal plane. The mid frame is connected to the pontoon. The frame to hold the pontoon in place is located inside the columns, without touching the column itself. Each column is mounted on its own force measurement frame. All models were constructed of smooth PVC with a typical surface roughness of 1.5-7µm.

A systematic test series was carried out in a depressurized towing tank at atmospheric pressure. The main dimensions of this basin are 240 x 18 x 8m (l x w x h). The model projected area is less than 1% of the cross section of the basin, sufficiently small to avoid blockage effects.

The tests were carried out at the Reynolds and Froude numbers presented in Table 1. These are based on column diameter, D. The results discussed here are for ReD=2.26E5 and Fr=0.16. The Froude number is low and during the experiments no relevant bow waves were visible.

Table 1: Reynolds and Froude Numbers for Model Testing. Reynolds Number Froude Number

1.13E5 0.08 2.26E5 0.16 2.83E5 0.21

The system was checked prior to the model tests by applying a known load to the setup (Figure 2). The loads measured on the force frames were then compared to this load. An example of the results measured for Fx are given in Table 2 for two load levels. The column where the load was applied is highlighted in italics and should be 100%. The results show that the forces were measured with 2% accuracy for this test setup.

Figure 2: Force-Check Setup.

Table 2: Force-Check Results for Fx.

Column 1 Pontoon Column2 98.8% 0.8% 0.0% 99.2% 0.5% 0.0%

Column1 Pontoon Column2

0.7% 98.2% -0.1% 0.5% 98.4% 0.0%

Column1 Pontoon Column 2

0.2% -0.2% 99.1% 0.2% -0.1% 99.2%

The sign convention defines a positive force in the positive

x-direction. The front column was named ‘column1’ and the downstream column was named ‘column2’. The drag and lift coefficients were computed according to;

, 2,

2, 2 , towing-speed.1

2

x yx y ref ref

ref ref

FC A D V

V Aρ= = =

The reference time scale was defined by /ref refT D V= .

Figure 3 presents an example of the measured forces for the test case in this paper. It was observed that column2 is pulled forward into the wake of the upstream column. This suction force is a known phenomenon, described in [1]. The downstream column experiences larger fluctuations in the drag force than the upstream column.

Figure 3: Time Traces of the Fx Forces on the Two

Columns.

Fx4

Fy4

Fy1

Fx1 PULL WITH KNOWN FORCE


The time traces of the forces have been filtered for background noise and other non-physical, high-frequency content. A 4rad/s cutoff frequency was used. A statistical average was then calculated and an average value for drag and lift was computed and plotted against the inflow angle as illustrated in Figure 4. In this case, the draft (T) of the model is also modified, where T2 = 2T1. A factor of ~2 separates the coefficients for the two drafts, as expected.

0 90 180 270 360

Heading (deg.)

Coe

ffici

ent

Cx_T1 Cx_T2 Cy_T1Cy_T2

Figure 4: Average Lift and Drag Coefficients for the Two

Drafts.

SEMI-EMPIRICAL MODEL

An empirical shielding model was developed based on a wake model behind bluff structures. The method was originally developed for wind loads in [2]. The drag coefficients are taken from the ESDU data sheets [3], which provide empirical formulations based on wind tunnel tests for various shapes, Reynolds numbers and roughness values. The relative shielding between columns is computed using a potential flow based model, originally developed by Parkinson and Jandall [4].

Figure 5: The Semi-Empirical Model.

Figure 5 shows a screen-shot of the semi-empirical model.

The user selects environmental conditions (Reynolds number), the shape of the column and the dimensions of the sub-components of the structure. Based on the information available in the database and the shielding model, the forces on the total structure are computed for the complete range of

inflow angles. The results for the total drag on the structure are compared to the tank model tests in Figure 6 and show good agreement. There are some additional parameters in the semi-empirical model (e.g. turbulence background levels, roughness) which could improve the results. These deserve further investigation.

0 90 180 270 360

Heading (deg.)

Coe

ffici

ent

Cx Model TestCx Semi-EmpCy Model TestCy Semi-Emp

Figure 6: Semi-Empirical Results Versus Tank Test Results.

CFD CALCULATIONS

The CFD calculations were carried out in three phases: 1) Blind Computations: computations carried out without

knowledge of the tank test results; 2) Improved Computations: computations improved for

better agreement with the tank test data; 3) Additional Computations: extra numerical studies (e.g.

turbulence, roughness and wall functions) and physical studies (e.g. influence of the draft, inter-column distance and free-surface effects). In this paper, the major findings for the Blind and Improved phases are studied in detail.

Codes Used

Three different viscous-flow codes have been used in the JIP: two commercial packages, CFX [5] and STAR-CCM+ [6], and one in-house code, FreSCo [7]. All codes solve the unsteady RANS equations, together with eddy-viscosity-based turbulence models. Some also have LES capabilities which were not investigated in this project. Several participants used CFX, allowing an assessment of how different users using the same code have tackled the flow problem. The emphasis is on showing the different codes, settings and grids used, and to correlate these with the results obtained, in order to create best-practice guidelines. The four participants are denoted ‘P1’ through ‘P4’.

Table 3 summarizes the codes used and their main characteristics. The codes employ the finite volume method. All codes use the MPI parallel scheme and sub-domain decomposition. CFX is weakly-coupled, solving mass and momentum conservation in one matrix. STAR-CCM+ and


FreSCo are segregated codes, solving mass and momentum separately, with the coupling carried out iteratively. While STAR-CCM+ and FreSCo can deal with any kind of unstructured elements (any polyhedral cell type), CFX can only handle general types (hexahedrals, tetrahedrals, prisms and pyramids). Numerical Set-up and Boundary Conditions

All computations were done for the model-scale Reynolds number (ReD=2.26E5) without considering free-surface effects. The model basin was modeled with the real side and bottom wall dimensions. The inlet and outlet were positioned 10D and 20D from the center of the structure respectively. Figure 7 shows a typical numerical set-up used for the computations. The boundary conditions were as follows: Inlet: Inflow boundary with velocity field

and turbulence quantities

prescribed. ( ,0,0)refV=V

Outlet: Outflow boundary with zero normal gradients.

Sides/Bottom: Free Slip boundary. Water surface: Symmetry boundary. Structure: No Slip boundary, no wall-functions.

Figure 7: Boundary Surfaces.

Grids

Typical grids used by the participants are illustrated in Figure 8. The grids do not have the same number of nodes. Three types of grids were used; structured hexahedral grids with an O-O topology (P1 and P3); an unstructured tetrahedral grid with a prism layer for boundary layer resolution (P2); and an unstructured polyhedral grid with a prism layer for boundary resolution (P4). The connection between the columns and the pontoon for P1 is not modeled in the same way as for the other participants. The rounded corners are abandoned close to the pontoon, which increases the area of the columns. P2 did not comply with the basin dimensions in order to save grid nodes.

The grid types were not changed between the Blind and Improved phases.

P1

P2

P3


Figure 8: Example Grids Used by the Participants.

Blind Computations

The numerical settings used by the participants for the Blind phase are presented in Table 4. All CFX users have chosen the CFX-high-resolution convection spatial-discretisation scheme, which is a blended scheme and has an order of accuracy between 1 and 2. Participant P3 and P4 used higher than 2nd order schemes. Only P4 used a 1st order time-

discretization scheme. The other participants used a second-order backwards time-discretisation scheme. P4 The number of outer-loops is related to the solving strategy used by the codes. Coupled solvers require fewer outer loops than segregated solvers. The convergence tolerance used by P3-FreSCo is smaller than the other participants. The differences in the time steps were also large. No participants used wall-functions. Considering the fact that one expects a highly 3D unsteady massive separated flow around a blunt body (far from a 2D steady friction-driven flat plate flow) is justified, near-wall modeling errors being then avoided. Only P3-CFX used the SAS-SST turbulence model [8]. The other participants used the k-ω SST turbulence model [9].

Figure 9 presents results for the drag coefficient Cx and compares them with the experimental data and semi-empirical estimate. Several trends are evident. All CFD computations underestimate the average Cx coefficient when compared with the experiments. The differences between the CFD results are smaller than the differences between the best CFD result and the average experimental result. The semi-empirical prediction overestimates the average Cx value. The unsteady behaviour of the drag coefficient is underestimated by all CFD computations. Repeating the experiments several times should give some insight into these peaks and improve the quality of the comparison with the CFD results.

Table 3: Comparison of Codes Used in this Study.

Participant Code Name

Code Type Version Discretization Method

Solver Strategy

Parallelization Method

Grid Types Possible

P1 P2 P3

CFX Commercial 11.0SP1, 12.0 Finite Volume

Coupled mass & momentum

Decoupled turbulence

MPI Unstructured

P4 STAR-CCM+ Commercial 3.06.006 Finite Volume Segregated MPI

Arbitrary Unstructured Polyhedrals

P3 FreSCo In-house v0.9 Finite Volume Segregated MPI Arbitrary Unstructured

Polyhedrals

Table 4: Numerical Settings for Blind Computations.

Participant Grid Type Grid Size

Convec. Scheme

Time Scheme

Outer Loops

Conv. RMS

Time Step

Simulation Time

Turbulence Model

BLayer

P1-CFX Structured Hexahedrals

1.8M CFX-high-resolution

2nd order 4 1E-4 Tref/2 75 Tref K-Omega SST

y+<1

P2-CFX Unstructured 246K CFX-high resolution

2nd order 4 1E-4 Tref /7.5 60 Tref K-Omega SST

y+<5

P3-CFX Structured Hexahedrals

356K CFX-high-resolution

2nd order 2 1E-3 Tref /100 50 Tref SAS-SST y+<1

P4-STAR Unstructured Polyhedrals

700K 2nd order upwind

1st order 10 2E-3 Tref /15 70 Tref K-Omega SST

y+<1

P3-FreSCo Structured Hexahedrals

356K QUICK 2nd order 10 1E-6 Tref /100 100 Tref K-Omega SST

y+<1


Figure 10 presents the lift force coefficient Cy time

histories. The lack of modulation in the CFD curves indicates that vortex shedding is not occurring. Only P2 and P3-CFX show some modulation. The semi-empirical method can only predict average static loads and therefore the lift prediction is, by definition, zero.

The CFD computations have the same global trends and the flow field patterns are also similar. Figure 11 shows a typical dimensionless vorticity field ( / refD Vω ) at time t/Tref =

50.

Figure 9: Cx Results for Blind Computations.

Figure 10: Cy Results for Blind Computations.

Figure 11: Total Vorticity at z=0 and y=0 Planes.

The force results are partly explained by the flow field

images. The flow separates close to the rounded corners. The separation bubble extends until the second column. The separation bubble extends under the structure from the first pontoon sharp corner for almost the complete length of the structure. This leads to a complete dead-water region between the two columns, which does not permit any vortex shedding from the first column and affects the inflow to the second column. Additionally, the shallow draft inhibits the vortex shedding after the second column.

For these reasons, the CFD computed flow for the blind-computation settings is almost steady. In this case, the added-value of a complex CFD computation to derive load values is low compared with a simple semi-empirical prediction. Nevertheless, the errors for the average drag are lower and the flow field around the structure can only be visualized and studied in detail with the help of CFD. Improved Computations The numerical settings used by the participants for the Improved Computations phase are described in Table 5. The grids used are the same as shown in Figure 8 but with a different number of nodes. All changes with respect to the Blind computations are shown in italics in Table 5. Except for P1, all participants refined the grids. P1 used the SAS+SST turbulence models. P2 increased the grid density and the time resolution. P3-CFX increased the grid density. P4 increased the order of the discretization of the time scheme, the number of outer loops and the time resolution. P3-FreSCo increased the space and time resolution, and the number of outer-loops, in order to decrease the iterative error.

Figure 12 and Figure 13 present the time histories of the improved force coefficients and compare them with the experimental data and semi-empirical estimates. The visible trends are:


Table 5: Numerical Settings for Improved Computations.

Participant Grid Type Grid Size

Convection Scheme

Time Scheme

Outer Loops

Conv. RMS

Time Step

Simulation Time

Turbulence Model BLayer

P1-CFX Structured Hexahedrals 1.8M CFX-high-

resolution 2nd order 4 1E-4 Tref/2 330 Tref SAS-SST y+<1

P2-CFX Unstructured 390K CFX-high resolution 2nd order 4 1E-4 Tref/30 93 Tref

K-Omega SST y+<1

P3-CFX Structured Hexahedrals 651K CFX-high-

resolution 2nd order 2 1E-3 Tref/100 50 Tref SAS-SST y+<1

P4-STAR Unstructured Polyhedrals 700K 2nd order

upwind 2nd order 25 1E-3 Tref/30 80 Tref K-Omega

SST y+<1

P3-FreSCo Structured Hexahedrals 650K QUICK 2nd order 35 1E-6 Tref/200 100 Tref

K-Omega SST y+<1

Table 6: Errors for All Methods [%] for Improved Computations.

Method Aver Cx

Max Cx

Min Cx

Max Cy

Min Cy

Experiments 0 0 0 0 0 Semi-

Empirical +41.0 n/a n/a n/a n/a

P1-CFX -26.2 -51.9 10.7 -76.8 -76.1

P2-CFX -31.5 -49.5 1.3 -67.4 -89.1

P3-CFX -31.5 -46.7 -4.0 -56.8 -70.7

P4-STAR -26.2 -53.8 26.7 -97.9 -93.5

P3-FreSCo -8.4 -37.6 33.3 -25.3 -29.3

• All CFD results have improved towards the

experimental values. Greater lift and drag modulation is visible in the computed time traces.

• Table 6 presents the errors of all methods and computations with respect to the experimental values. These errors are computed as the relative difference between the numerical and the experimental results (in percentage of the experimental value). The numerical results considered for the computation of the error are the ones after the decaying of numerical transients. The minimum errors for the average of Cx, and the maximum of Cy, are 8% and 25% respectively.

• The semi-empirical estimate is further from the tank test experimental result than are the CFD results.

• The lowest errors are found for P3-FreSCo when using finer time-steps, higher-order schemes, tight control of the iterative convergence, medium-size grids and large computational times (7 days using 32 processors). CPU times are not compared here because different machines were used by each participant. However, for all other participants the computational times were of the order of a few days (1-4) using less than 4 processors.

Figure 12: Cx Results for Improved Computations.

Figure 13: Cy Results for Improved Computations.


Figure 14 presents the results for the normalized total vorticity at two different planes at t/Tref=50 for all participants. The choice of the same time t/Tref does not ensure a reliable comparison. Due to different initial conditions and numerical transients, the solution for the same t/Tref may not be the same for the different calculations. Nevertheless, this comparison permits an assessment of the main differences between the computed flow fields by the participants.1 The main trends visible are as follows.

• P1’s results show that the separation bubble appearing on the side of the first column extends up to the second column, inhibiting vortex shedding between the columns. A separation bubble starts at the sharp corner under the pontoon and extends up most of the pontoon length. This is also visible in P4’s and P3-F’s results. Von Karman vortex shedding is not visible behind column1 or column2 in these results. This can be explained by the very coarse time-step that the participants used.

1 Note: some of the oscillations visible here may be due to the way the

visualization software handles the different type of grids, solution storage and interpolation algorithms.

P3

P3-F

P4

P1

Figure 14: Total Vorticity at the z=0 and y=0 Planes for P1-P4.

P2 • P2’s results show similar trends as the other partners,

with some point-to-point oscillations due to the non-smooth grids used. The results are qualitatively similar to the results presented by P1 (same code) but the thickness of the detached bubble is larger for P2 and does not touch the second column.


• P3’s results are typical of a SAS simulation. The vortex structures are less coherent than for eddy viscosity based turbulence model (e.g. P4 or P3-F). The separation bubbles detaching from the first column are smaller than the ones seen in P1 and this results in more vortex-shedding and larger modulation.

• P3-F’s results show highly coherent vortical structures between the columns and behind the aft column. These structures are too coherent however; this is a feature commonly associated with URANS and eddy-viscosity based turbulence models. The vortices are too strong and the separation bubble detaching from the first column is thin compared with the results of the other participants, which may indicate smaller drag forces on the first column.

• P4’s results show coherent vortex structures but they’re weaker than those of P1, P3 and P3-F. Vortex shedding occurs but is also weaker. The separation bubble on the side of the first column is thicker than for the other computations.

The forces on the individual columns were also analyzed

(see Figure 15). The results show that P3, P4 and P3-F have produced similar results for the drag force on column1, while P1 has produced larger values and better agreement with the experiments. P3-F over-predicts the experimental data for column2. The other results under-predict the experimental data. This is due to the larger lift force at the first column, which affects the inflow to the second column due to the increased vorticity. This comparison also shows that, even if the agreement for the total structure is correct, it is possible that for the individual components larger deviations may occur (e.g. P3-F’s results). Only P3 and P3-F produce results comparable with the experimental lift loads.

Extra Numerical Studies

Three participants performed additional sensitivity studies subsequent to the Improved phase which contributed to further understanding of the results. P1 investigated the application of a transition model, P4 studied laminar-turbulent flow transition by means of turbulence suppression and P3-F investigated the numerical accuracy of the results in detail.

P1-CFX: transition model application

There is a laminar-turbulent transition model in CFX called the Local Correlation-Based Transition Model (see [10]) which adds two extra transport equations; one for the boundary layer intermittency and one for the momentum-thickness Reynolds number. The method forms a framework for the implementation of correlation-based models into general-purpose CFD. This model should be applied to marine problems with care, since it was designed for aerodynamics applications. Figure 16 presents the forces calculated by P1 with, and without, the transition model.

Figure 15: Cx and Cy for the Improved Computations,

column1 (top) and column2 (bottom).

Figure 16: Cx and Cy Results for the P1-CFX+ Transition

Model Computations, column1 (top) and column2 (bottom).


I

ess turbulence until x/D = 0.5 on the because of the low,

y studies with e-step and number of

er

ts application increases the drag load on the first column anddecreases it on the second column. The lift loads on both columns are increased towards the experimental values. For the first column, it’s likely that the overestimate of Cx is due to the different column geometrical modeling. For the second column, the underestimation of the drag is likely due to the fact that the transition model was not intended to operate behind a shear-layer. The lift prediction improves considerably for the overall structure, while the agreement with the experimental drag force is not greatly improved.

P4-STAR: turbulence suppression studies

P4 decided to supprfirst column. This decision was taken model-scale Reynolds number and because it was known that turbulence models sometimes predict transition to turbulence too soon. The transition location was estimated and may not correspond with the real transition point. However, this action postponed the transition to a point further downstream of what is usually considered by any turbulence model. The effect was considerable. Figure 17 shows that the drag prediction improves for column1 and there is greater lift. For column2, the drag decreases and the lift is also larger. This method is not practical for a complex structure but it illustrates the sensitivity of the results to the turbulence separation point. P3-FreSCo: numerical accuracy studies

P3 performed additional numerical accuracFreSCo. They studied the effect of the timouter-loops for three different grids levels. This permitted an assessment of the influence of the numerical errors (space-discretization, time-discretization and iterative error) on the solution. This participant has substantial computer power available. It may not be practical to perform these studies for each new semi-submersible.

Figure 18 presents an example error analysis of average drag value for the complete structure. The error is computed in the same way as for Table 6. This is a first step towards an uncertainty analysis [12]. For unsteady computations, the space-discretisation ( 1/i ih Ncells= , with i the index corresponding to the grid used and i=1 corresponding to the finest grid) and the time ror ( /-discretisation j reff t T= Δ , with

1f denoting the finest time-step used) play a role. Ideally,

0, 0i jh f→ → . The iterative error is rep y the

mber of outer loops and by the size of the symbols. The is the lowest for the largest number of outer-

loops. The error analysis shows that the difference to the experiments is decreasing in general for finer time-steps and finer grids. However, it also shows that the behaviour is highly non-linear and that it is possible to obtain better results with a medium grid, combined with a medium time-step, than with a very coarse grid, together with a very fine time-step, or a fine

resented b

nuiterative error

Figure 17: Cx and Cy Results for the P4-STAR+ Turbulence Suppression Computations, column1 (top) and column2

advisable to perform grid refinement studies for a

(bottom).

Figure 18: Numerical Error Analysis for Average Cx by P3-FreSCo.

grid combined with a coarse time-step. Hence, it is not and time

coarse grid and to extrapolate the results for a fine grid. For the coarse grid with 1/ 2.8ih h = (2.8 times coarser than the finest grid) the trends show that the time-step does not influence the results and the av rror is around -30%. However, for the medium grid it is seen that time-step has a large influence. The error decreases from -30% to -8% when using finer time-

erage Cx e


steps. The iterative error adds extra complication due to the non-linear solution process and to the observed fact that using larger time-steps requires larger number of outer-loops to achieve the same iterative error. This must be investigated further.

CONCLUSIONS Tank model tests, empirical-model estimates and CFD

been carried out for the current loads on sem

es. The result is a fair initial estim

get an accurate time histo

s have been lear

nd time discretisation) are small and

into the numerical errors and uncertainties. Phy

computations have i-submersibles and mono-hulls. Results for one semi-

submersible configuration have been presented and analyzed in this paper. The average loads have been considered and the time histories of the loads have been scrutinized. The following conclusions have been made.

The use of a semi-empirical model provides the designer only with time-averaged forc

ate in terms of speed/quality/complexity. However, errors of 40% for the average Cx load have been observed. The estimated loads are larger than for the model-tests, leading to conservative designs but the designer gets no dynamic and flow field information from using these models.

The CFD computations carried out by the different participants showed that it is not simple to

ry of the drag and lift loads. During the Blind phase the results had errors larger than 20% for the average drag and most of the computations did not exhibit any lift. For the Improved computations, the best result had an 8% error for the average drag and 25% error for the maximum lift. The experiments also revealed large variations in the loads over time. These experiments should be repeated several times in order to permit a more reliable statistical analysis.

Additional studies have been presented to explain the discrepancies between the results and some lesson

ned. Two key issues play a role; laminar-turbulent transition and numerical errors. Because the model-scale Reynolds number is relatively low, part of the flow is likely to be laminar. The turbulence models used with RANS are known to predict the transition too early. This explains some of the deviations of the numerical results from the experimental results. It has been shown that delaying the turbulent transition improves the CFD results. In the experiments, turbulence-tripping-strips could be used to minimize these errors.

The CFD user must also ensure that the numerical errors (iterative, space discretisation a

the uncertainties should be quantified if possible. It has been shown here that this is important for obtaining accurate results. Fine grids with relatively coarse time-steps, or relatively coarse grids with fine time-steps, are likely to lead to unphysical results.

Future work will include further study into the turbulence transition issues and

sical aspects such as the free surface effects and scale effects will then be considered.

ACKNOWLEDGMENTS The following participants in the Current Affairs JIP are

acknowledged; Amog Consulting, Petrobras, GustoMSC, Projemar, Heerema, Statoil Hydro, Keppel Fels, Total, INTECSEA and MARIN. The authors thank Christiaan Klaij and Arjen Koop of MARIN for their contribution to the computations, for their review of this paper and for their input on numerical issues. Thank you to Michael Gachet from GustoMSC for his comments on the paper.

REFERENCES [1] Finnigan, T. “Current Blockage Effects on Model-Scale Offshore Platforms”. Civil Engineering in the Oceans V, 1992.

[2] van Walree, F., and van den Boom, H.J.J. “Wind, Wave and Current Loads on Semi-Submersibles”. OTC1991-6521, 1991.

[3] ESDU Data sheet 80025. Mean Forces, Pressures and Flow Field Velocities for Circular Cylindrical Structures: Single Cylinder with Two dimensional Flow.

[4] Parkinson,G.V. and Jandall,T. “A Wake Source Model for Bluff Body Potential Flow”. J.Fluid Mech, Vol. 40, Pt.3, pp. 577-594, 1970.

[5] ANSYS Inc., ANSYS CFX11.0 Users Manual, 2008.

[6] CD-Adapco, User Guide STAR-CCM+, version 3.06.006, London, 2008.

[7] Toxopeus, S., and Vaz, G. “Calculation of Current or Manoeuvring Forces using a Viscous Flow Solver”. OMAE2009-79782. Honolulu, Hawaii, USA. 2009.

[8] Menter, F. R., and Egorov, Y. “A Scale-Adaptive Simulation Model using Two-Equation Models”. AIAA Journal, 1095, 2005.

[9] Menter, F. “Two-equation Eddy Viscosity Turbulence Models for Engineering Applications”. AIAA Journal, 32, pp. 1299–1310. 1994.

[10] Langtry, R., and Menter, F. R., “Transition Modelling for General CFD Applications in Aeronautics”. AIAA Journal, 522, 2005.

[11] Vaz, G., Mabilat, C., van der Wal, R., Gallagher, P. “Viscous Flow Computations on Smooth Cylinders. A Detailed Numerical Study with Validation”. OMAE2007-29275. San Diego, California, USA. 2007.

[12] Eça L, Hoekstra M. “An Evaluation of Verification Procedures for CFD Applications”, 24th Symposium on Naval Hydrodynamics, Fukuoka, Japan. 2002.


Current Affairs: Model Tests, Semi-Empirical Predictions ... … · Proceedings of the 28th International Conference on Ocean, Offshore and Arctic Engineering . OMAE2009 . 31 May

Documents