Progress in Computational Fluid Dynamics, Vol. 8, No.

5, 2008

299

Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES A.C. Benim*
Department of Mechanical and Process Engineering, Duesseldorf University of Applied Sciences, Josef-Gockeln-Str. 9, D-40474 Duesseldorf, Germany E-mail: alicemal.benim@fh-duesseldorf.de *Corresponding author

E. Pasqualotto
Development Engineer Combustor, Alstom Power Systems, Turbomachines Group, Alstom (Switzerland) Ltd., Brown Boveri Strasse 7, 5401 Baden, Switzerland E-mail: ennio.pasqualotto@power.alstom.com

S.H. Suh
Department of Mechanical Engineering, Soongsil University, (156-743) 511 Sangdo-Dong, Dongjak-Gu, Seoul, Korea E-mail: suhsh@ssu.ac.kr
Abstract: Incompressible turbulent flow past a circular cylinder is computationally investigated by a cascade of turbulence modelling strategies. Results are compared with experiments, by keeping the focus on the predictability of the drag coefficient. Two-Dimensional Reynolds Averaged Numerical Simulations (2D RANS) and Two-Dimensional Unsteady Reynolds Averaged Numerical Simulations (2D URANS) are performed for a wide range of Reynolds numbers, encompassing the critical regime. It is shown that the drag crisis can qualitatively be predicted by the 2D RANS, provided that the near-wall region is adequately resolved. The 2D RANS underpredicts the experimental drag coefficients for the most Reynolds numbers, seemingly due to the missing contribution of the organised transient motion. The 2D URANS, on the other hand, generally overpredict the drag coefficients, indicating the intimate relationship between the organised transience and three-dimensionality. Computations by three-dimensional transient procedures such as Three-Dimensional Unsteady Reynolds Averaged Simulations (3D URANS), Large Eddy Simulations (LES) and Detached Eddy Simulations (DES) are performed for the Reynolds number of 104. A substantial improvement in the prediction quality is observed. The best agreement with the experiments is obtained by the LES. The DES is also observed to deliver results that are very close to the measurements. Keywords: circular cylinder; turbulent flow; CFD; RANS; URANS; LES; DES. Reference to this paper should be made as follows: Benim, A.C., Pasqualotto, E. and Suh, S.H. (2008) Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES, Progress in Computational Fluid Dynamics, Vol. 8, No. 5, pp.299307. Biographical notes: A.C. Benim (Prof. Dr.-Ing.) received his BSc and MSc Degrees in Mechanical Engineering at the Bosphorus University of Istanbul, Turkey, and his PhD Degree at the University of Stuttgart, Germany, in 1988. Following a post-doctoral period at the University of Stuttgart, he joined ABB Turbo Systems Ltd. in Baden, Switzerland in 1990. He was the manager of the Computational Flow and Combustion Modelling group. Since January 1996, he is Professor for Energy Technology at the Duesseldorf University of Applied Sciences, Germany. E. Pasqualotto (Dipl. Masch. Ing. ETH) received his diploma in Mechanical Engineering in 1989 from the ETH Zuerich, Switzerland. Within the period 19902002, he was employed by the companies ABB Power Generation Ltd. in Baden, Switzerland, Waertsilae NSD Ltd. in Winterthur, Switzerland, HBI Haerter AG in Zuerich, Switzerland and Andritz VA Tech Hydro GmbH in Zuerich, Switzerland, where he worked in development of gas turbine combustors (ABB Power Generation), in development of diesel engine injections systems (Waertsilae NSD), in development of tunnel ventilation systems (HBI Haerter) and in development of Pelton turbines (Andritz VA Tech Hydro). Since 2008, he is employed by Alstom Power Systems

Copyright 2008 Inderscience Enterprises Ltd.

300

A.C. Benim et al.

in Baden, Switzerland, where he is working in the development of combustion systems of industrial gas turbines. S.H. Suh (Prof. Dr.-Ing.) studied Mechanical Engineering in Soongsil University, Seoul, Korea and obtained his BS and MSc Degrees in 1980 and 1984. He obtained his PhD Degree in 1989 from University of Stuttgart, Germany. He is currently Professor of the Soongsil University. He is the head of the Korean Fluid Machinery Association.

Introduction

External flows around bluff bodies represent a major area in fluid mechanics, with a vast number of applications in many disciplines of aero- and hydrodynamics. Following the pioneering work of von Krmn (1911), an enormous and still rapidly growing body of literature has developed on the subject. This can also be seen as a manifestation of the difficulty in understanding and adequately describing the associated complex flow phenomena. Flows exhibiting particularly simple configuration, such as those past a cylinder or sphere have been subject of extensive experimental and computational investigation as generic configurations. In the present work, a hydraulically smooth and infinitely long cylinder blown by a unidirectional, homogeneous flow in unbounded domain is investigated.

1.1 General flow features

An overview of the flow regimes that occur with increasing Reynolds number (Re) was given by several authors (Lienhard, 1966; Achenbach, 1971). In the high Reynolds number range (Re 104) generally, four different flow regimes are defined, namely subcritical, critical, supercritical and transcritical. In all regimes a characteristic feature of the flow is the unsteady vortex shedding, known as the von Krmns vortex street, which becomes rather disorganised in the critical regime. The Reynolds numbers marking the boarders between the regimes can significantly vary depending on many parameters of the experimental setup such as free-stream turbulence, wind-tunnel blockage and model-end conditions, length-to-diameter ratio as well as the actual surface roughness. In the subcritical regime the boundary layer is completely laminar, and the drag coefficient remains fairly constant around 1.01.2. Transition to turbulence takes places in the wake. Within a range of higher Reynolds numbers encompassing the critical regime (approx. 105 < Re < 106) a sudden drop in the drag coefficient to values 0.20.4 is observed. This phenomenon, which is also called as drag crisis is due to a transition to turbulence in the boundary layer prior to separation and the associated downstream movement of the separation point.

1.2 Review of previous computations

In a number of computational investigations (Jordan and Fromm, 1972; Jain and Goel, 1976; Tamura et al., 1980; Nieuwstadt and Keller, 1973; Braza et al., 1986; Behr et al., 1995; Anagnostopoulos and Iliadis, 1996; Ding et al., 2004;

Sahin and Owens, 2004; Ben Richou et al., 2005; Shu et al., 2005; Braack and Richter, 2006; Nobari and Naderan, 2006), unsteady, two-dimensional Navier-Stokes equations are solved for rather low Reynolds numbers, in the range up to 300. In some of those investigations (Behr et al., 1995; Anagnostopoulos and Iliadis, 1996; Sahin and Owens, 2004; Ben Richou et al., 2005) the emphasis was placed on the analysis of the blockage effect, whereas in others (Ding et al., 2004; Braack and Richter, 2006) the main purpose was to use the flow as benchmark. Some authors investigated the effect of cylinder rotation (Shu et al., 2005) and oscillation (Nobari and Naderan, 2006). In Jordan and Fromm (1972), Braza et al. (1986), the two-dimensional, unsteady Navier Stokes analysis was extended up to Re = 1000, where the wake flow is known to be turbulent, but without using any turbulence model. Unsteady, two-dimensional Navier-Stokes equations, without using any turbulence model, were solved computing flows past cylinders also at high Reynolds numbers. Ishii et al. (1985) performed computations for Reynolds numbers in the range 1.0 1057.83 106, using a Beam-WarmingSteger method with improved accuracy. Although a good agreement with the experiments was achieved, it was found (Tamura et al., 1990) that the results were grid dependent. Tamura et al. (1990) used a third order accurate upwind scheme and extended the analysis to three-dimensional formulation. By the three-dimensional modelling, improved results were obtained especially for the critical regime. Nevertheless, the calculations were, in general, highly sensitive to grid fineness. In both investigations, no explicit filtering was applied to set the scale of the dissipative mechanism. Such procedures, which purely rely on the dissipative nature of the numerical scheme, without and explicit subgrid scale model (also designated as no model LES) are controversial (Sagaut, 2002). Unsteady, two-dimensional analysis for high Reynolds numbers, relying on different turbulence models were also presented. Song and Yuan (1989) applied a two-dimensional pseudo-LES model to flows around a cylinder with Reynolds numbers in the range 1.4 1058.4 106. The lack of three-dimensionality, which is by definition an essential feature of LES, was compensated by adopting the modelling to the particular application by modifying the subgrid scale constant and introducing ad hoc modifications in the boundary layer. Franke (1991) applied a URANS approach to the flow past a circular cylinder at Re = 1.4 105 using two different turbulence models, namely a two-zone k- model (k: turbulence kinetic energy, : dissipation rate of k), and a two-zone Reynolds stress

Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES model. Here, two-zone refers to solving only the k-equation in a layer near the wall, whereas is determined from an empirical algebraic expression with features accommodating for low-Reynolds number effects and transition. Steady-state, two-dimensional RANS computations were performed by Majumdar and Rodi (1985) for Re = Re=1.4 105, and by Celik and Shaffer (1995) for Reynolds numbers in the range 1.0 105Re = 1.0 107. Such an approach assumes that the unsteadiness associated with the vortex shedding can sufficiently be accounted for, in the time mean, by the applied turbulence model. In both investigations, a standard k- model was used. For the near-wall flow, a wall-functions approach was adopted. The principal inadequacy of the latter for the present flow was compensated by introducing rather empirical modifications for considering transitional effects and determining the transition point. Along with improving computational capabilities, recent investigations of the flow past a circular cylinder have increasingly been based on rather expensive transient, three-dimensional procedures such as the truly Large Eddy Simulations (LES) or Detached Eddy Simulations (DES). Kato and Ikegawa (1991) presented one of the early LES studies, using the Smagorinsky subgrid scale model. They imposed a condition of relaminarisation within a layer near the wall. The thickness of the this layer was rather arbitrarily set to of 20% of the cylinder diameter. They demonstrated the superiority of truly (three-dimensional) LES against two-dimensional pseudo LES approaches, by comparisons with measurements for Re = 1.4 105. Breuer (1998a, 1998b) performed an LES analysis for Re = 3900, using three different approaches for subgrid scale modelling, namely the Smagorinsky model, the dynamic model and the no-model approach. Role of the discretisation scheme and grid resolution were also investigated. It was concluded that the dynamic model combined with a central differencing scheme yields the best solution. Lbcke et al. (2001) presented results for Re = 3900 and Re = Re = 1.4 105, which were obtained by LES based on a dynamic model and URANS based on different approaches. It was shown that a comparable accuracy to LES could be obtained by the proposed non-equilibrium explicit algebraic stress model. Franke and Frank (2002) presented LES results for Re = 3900. Near walls, the damping function of Van Driest was used. An emphasis of the investigation was the realisability, as the computed subgrid scale stresses were analysed and found to be always realiseable, due to the sufficiently fine grid resolution. Fujisawa et al. (2005) investigated the flow around a rotationally oscillating cylinder by LES. DES of the flow past circular cylinder were presented by Travin et al. (1999) and Strelets (2001) for Re = 5 104, 1.4 105 and Re = 3 106. Many important aspects of the flow physics, modelling and numerics were discussed (Travin et al., 1999; Strelets, 2001) as the DES was based on the one-equation Spalart-Allmaras model.

301

1.3 Present contribution

As summarised above, different aspects of the flow past a circular cylinder were investigated in detail by many researchers, using different modelling strategies and tools. The purpose of the present paper is to present a rather coherent validation study for a cascade of different modelling strategies for turbulent flow, extending from two-dimensional RANS and URANS, to three-dimensional URANS, DES and LES. Doing so, no special ad-hoc modifications of the models tailored to this type of flow are applied. This coherent validation study based on consistent/comparable strategies for the numerical disretisation and gridding, is believed to be of additional value for the research community. In the present analysis the focus is placed upon the predictability of the drag coefficient.

Modelling

Incompressible, turbulent flow of a Newtonian fluid is considered. At inlets, a unidirectional flow with spatially and temporally uniform distributions is assumed. Inlet boundary conditions for turbulence quantities are derived by assuming a very low (0.1%) turbulence intensity and a turbulent-to-laminar viscosity ratio (0.1%). At the outlet, a zero gauge pressure is prescribed along with zero-gradient conditions for remaining variables. At walls, no-slip conditions apply. In addition to the above-mentioned boundaries, solution domains are enclosed by symmetry planes. In the three-dimensional computations, the solution domain is bounded by periodic boundaries in the third dimension. The computational analysis is performed using the general purpose CFD codes ANSYS-CFX (2006) and Fluent (2006). Block-structured grids with conformal block interfaces are used. In the RANS and URANS analysis, the Shear Stress Transport (SST) model (Menter, 1994) is used as the main turbulence model. The k- model (Launder and Spalding, 1974) is also used in some RANS computations. Using the SST model, no wall-functions are used for the near-wall flow, whereas the k- model is applied in conjunction with the wall-functions approach (Launder and Spalding, 1974). For a proper application of both strategies for the near-wall region, it is essential that the wall-layer is adequately discretised. The important parameter in this context is the non-dimensional wall-distance y+ of the next-to-wall cells, which is defined as y+ = (y/)(w/)0.5, where y, , w and denote the cell-to-wall distance, kinematic viscosity, wall shear stress, and density, respectively. If the wall layer is to be resolved, the condition of y+ < 1 needs to be fulfilled. On the contrary, applying the wall-functions approach, y+ shall be large enough (y+ > 30), preferably within the range 100 > y+ > 30. In grid independency studies of the present investigation, grids are always generated by taking care for obtaining most favourable y+ values for the employed turbulence model.

302

A.C. Benim et al. wall-functions and 80 cells for the grid resolving the wall layer. In Figure 1, the predicted drag coefficients are compared with empirical/theoretical values (Streeter and Wylie, 1975) for laminar and turbulent boundary layer. As expected, one can see that the k- model with wall-functions (k- + WF) performs poorly in the initial (laminar) parts of the boundary layer, and approaches empirical values for turbulent flow for large Re. The SST model performs fairy well for low Reynolds numbers. For Re > 104, it shows a very good agreement with the laminar, theoretical curve, and predicts a rather sharp transition to turbulence about Re 34 105. It is interesting that the SST model (resolving the wall-layer) can predict transition fairly well, although no empirical information is fed to the model.
Figure 1 CD vs. Re for flat plate boundary layer

The applied DES model (ANSYS-CFX (2006)) uses the SST model (Menter, 1994) as the background URANS model and controls the URANS-LES switch by operating on the length scale in the dissipation term of the turbulence kinetic energy transport equation. A blending is applied between the URANS and LES zones. A special feature of this DES implementation (ANSYS-CFX, 2006) is the so-called wall shielding, which ensures the activation of the URANS model in a certain vicinity to the walls. This is originally introduced to prevent phenomena such as grid induced separation. In the present case, the applied grids being quite homogeneous and smooth, the latter is not thought to be a critical issue. Nevertheless the wall shielding is still activated for the present analysis. As a further modelling strategy, LES (Sagaut, 2002) are employed. The Smagorinsky model (Smagorinsky, 1963) is used as the subgrid scale model, applying a constant model coefficient of CS = 0.1. The Van Driest damping function (Hinze, 1975) is used for computing the subgrid scale viscosity in the vicinity of solid boundaries. In DES and LES, it is ensured that the condition of y+ < 1 is fulfilled everywhere by the employed grid. In LES computations the convection terms are discretised by central differencing. In RANS and URANS computations, the high resolution scheme (Barth and Jespersen, 1989) is used for the momentum equations. In DES, for the momentum equations, the central differencing is used in the LES regions, whereas the high resolution scheme is used in the URANS regions, with a certain blending in the transition zones. For the turbulence quantities, which are required within the framework of URANS and DES formulation, the upwind scheme is used. For the temporal discretisation a second order backward Euler scheme is applied in all computations (ANSYS-CFX, 2006). The time step size is chosen such that maximum cell Courant numbers (Peyret, 1996) below unity.

4 3 Preliminary analysis
In this preliminary analysis, the flat plate boundary layer is investigated for testing the ability of turbulence models to predict natural transition. Length of the solution domain is adjusted to achieve Rex = 1 106 (Rex = Ux/, where U, and x denote the free stream velocity, kinematic viscosity, and coordinate in flow direction measured from the leading edge, respectively). In the y (transverse) direction, the extension of the domain is defined to be ten times larger than the boundary layer thickness that is theoretically expected to occur (Streeter and Wylie, 1975) at the outlet. The rectangular solution domain is enclosed by a symmetry boundary on the top (a pressure boundary allowing mass transfer would be more convenient, but was not practicable due to convergence difficulties in the present configuration). Generated grids had an equidistant distribution in the x direction with 1000 cells. In y direction, the grid lines were concentrated near the wall, expanding into the domain by geometric expansion factors not larger than 1.2. This resulted in 30 cells for the grid to be used with the

Results

4.1 2D RANS and URANS

The solution domain and boundary types are shown in Figure 2, for 2D RANS. Boundaries of the blocks used to generate the block-structured grid are also indicated in the figure. For 2D URANS, the domain is, of course, twice as large, mirrored around the lower symmetry plane of Figure 2, exhibiting no symmetry plane through the cylinder.
Figure 2 Solution domain, boundaries, grid topology for 2D RANS

Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES To find the optimal domain size, a study is performed by varying H (Fig. 2), for Re = 104, using the 2D RANS SST model by resolving the wall layer. Figure 3 shows the variation of the predicted CD with H. One can see the blockage effect becomes small beyond H/D = 6. In the present computations, H/D = 10 is used.
Figure 3 CD vs. H (2D RANS SST, Re = 104)

303

analysis of Franke and Frank, 2002). In the present analysis at least 120 cells equidistant cells are used to discretise the half cylinder circumference. The largest aspect ratios that occur on the cylinder surface varied between 5 and 30. Away from the cylinder surface, the grid lines are geometrically expanded by rather mild expansion ratios (1.02 in the circular domain around the cylinder and 1.031.05 in the outer domain (Fig. 2)). The size of the grids for the 2D RANS SST computations by resolving the near-wall layer varied between 4.0 104 and 2.5 105 for the considered range of Reynolds numbers. Figure 5 shows a detail view of the grid used for 2D SST computations for Re = 104.
Figure 5 Detail view of grid for Re = 104 (used for 2D SST)

The apparent inadequacy of the wall-functions approach in predicting transition was already demonstrated (Fig. 1). Another problem in using the wall-functions in the present flow is that y+ shows a continuous variation within a wide range, encompassing stagnation and separation zones, making it difficult to attain favourable y+ values throughout, and a meaningful average value. 2D RANS computations are performed by the k- model using the standard wall-functions, for different grids resulting in different mean y+ values. In all computations, the overall grid is sufficiently fine and do not vary except in the vicinity of the wall. In Figure 4, CD is shown as a function of the mean y+ value (area-averaged over the surface) for different Re. It is interesting to see that there is hardly a plateau for the CD values, implying the impracticability of obtaining grid independent results using the standard wall-functions approach for the present flows.
Figure 4 CD as function of average y+ (2D RANS k- + WF)

Streamlines predicted by 2D RANS SST in the near-field of the cylinder are shown in Figure 6 for Re = 105 and Re = 5 105. One can see that the separation point moves further downstream and the recirculation zone becomes smaller as the Reynolds number increases.
Figure 6 Streamlines for Re = 105 (above) and Re = 5 105 (below) (2D RANS SST)

Figure 7 shows the streamlines at a time step as obtained by 2D URANS SST, for Re = 105, where the asymmetry of the flow can clearly be observed.
Figure 7 Streamlines at a time-step for Re = 105 (2D URANS SST)

In grid generation, as far as the resolution along cylinder circumference is concerned, it was found that 100 cells for half cylinder provide sufficient accuracy in RANS and URANS computations (This figure was about 80 in the LES

304

A.C. Benim et al. for all cases. The used time step size has been 1.0 105 s, guaranteeing cell Courant numbers being smaller than unity for all computations. Figure 9 shows the grid on the cylinder surface as well as the grid on the periodic plane.
Figure 9 A view of 3D grid (Re = 104)

Predicted (2D RANS k- , 2D RANS SST, time averaged 2D URANS SST) drag coefficients within a range of Reynolds numbers are presented in Figure 8. RANS k- results were obtained for mean near-wall y+ values varying between 30 and 100. RANS SST and URANS SST results were obtained by resolving the near-wall layer, fulfilling the condition of y+ < 1. Measurements of other authors that are borrowed from a compilation of Bohl and Elmendorf (2005) are also shown in the figure.
Figure 8 CD vs. Re: measurements (Bohl and Elmendorf, 2005) and predictions (2D RANS k- , 2D RANS SST, time averaged 2D URANS SST)

It is interesting to note that the measurements also show variations within a rather broad range, depending on the particular experimental conditions. One can see that RANS k- computations do not predict any transition, as expected. (Variations in the drag coefficients predicted by the wall-functions may be caused by the strong dependency of CD on the grid, i.e., on the mean y+ as demonstrated in Fig. 4). One can see that the RANS SST model resolving the wall-layer can qualitatively predict the reduction of the drag coefficient over the critical regime. Quantitatively, the predicted drag coefficients are observed to underestimate measurements for all Reynolds numbers. This is apparently due to the contribution of the organised transient motion, which could not be modelled by the RANS approach. On the other hand, one can see that the inclusion of transience within a two-dimensional formulation does not necessarily improve the results. In URANS SST computations, the contribution of the transient motion is predicted mostly in an exaggerated manner, leading to an over prediction for most Reynolds numbers (Fig. 8).

For displaying the vortical structures, the so-called Q criteria is utilised (Hunt et al., 1988). The Q isosurfaces (for Q = 1000) obtained by 3D URANS SST at a time step are presented in Figure 10.
Figure 10 Q isosurfaces at a time step for 3D URANS SST (Re = 104) (see online version for colours)

4.2 3D URANS SST, LES and DES

3D URANS SST, LES and DES computations are performed only for a single Reynolds number, i.e., for Re = 104. The depth of the solution domain is defined to be equal to the cylinder circumference, i.e., D. For the boundaries in the third dimension, periodicity is assumed. The grid structure of the 2D analysis is extruded in the third direction, by using 66 equidistant cells in the third direction. The resulting 3D block-structured grid has 2.6 106 hexahedral cells. The same grid is used for 3D URANS, LES and DES computations, fulfilling the condition y+ < 1 The importance of three-dimensionality can be observed in the displayed structures. In addition to the three-dimensional vortices extending in the third direction, vortices extending in the perpendicular direction, i.e., in the main flow direction are also detected (Fig. 10). Figure 11 shows the distribution of y+ on the cylinder surface for the LES computations at a time step. One can see that the maximum y+ values remain below 0.8. Distribution of the cell Courant number at a time step, on the periodic plane, for the LES computations is shown in

Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES Figure 12. As can be seen in Figure 12, the maximum cell Courant number remains below 0.25.
Figure 11 y+ distribution on cylinder surface at a time step for LES (Re = 104) (see online version for colours)

305

The surface separating the URANS and LES regions in the DES computations is displayed in Figure 14. One can see that the URANS is mainly active in the front region for distance comparable to the cylinder diameter, extending into the shear layers on both sides of the cylinder.
Figure 14 Surface separating URANS and LES regions a time step for DES (Re = 104) (see online version for colours)

Figure 12 Cell courant number distribution on periodic plane at a time step for LES (Re = 104) (see online version for colours)

Q isosurfaces obtained by DES at a time step are displayed in Figure 15.

Figure 15 Q isosurfaces at a time step for DES (Re = 104) (see online version for colours)

Q isosurfaces obtained by LES at a time step are displayed in Figure 13. Comparing to Figure 10, one can see that much finer structures are now captured by LES (Fig. 13), as expected, in comparison to 3D URANS SST.
Figure 13 Q isosurfaces at a time step for LES (Re = 104) (see online version for colours)

The predicted pressure coefficients along the cylinder surface (using the angular position from 0 to 180 as the co-ordinate) are presented in Figure 16. For 2D URANS, 3D URANS, LES and DES, the time-averaged values are presented, which are also averaged between the upper and lower surfaces. For 3D URANS, LES and DES, the values are also area-averaged along the cylinder height at each angular position. As can be seen, the predicted pressure coefficient values vary within a wide range, the maximum and minimum values being provided by 2D RANS and 2D URANS predictions, respectively. The 3D URANS, LES and DES predictions are comparably close to each other and lie roughly in the middle of 2D RANS and 2D URANS results.

306

A.C. Benim et al. the experiments is obtained by the LES. The DES is also observed to deliver results that are very close to the measurements.

Figure 16 Variation of time-averaged pressure coefficient along cylinder surface (Re = 104)

Acknowledgements
Mr. A. Nahavandi (Dipl.-Ing.), Mr. M. Cagan (MSc) and Mr. G. Peker (MSc) are acknowledged for their help. Support by the Soongsil University is Acknowledged.

References
Achenbach, E. (1971) Influence of surface roughness on the cross-flow around circular cylinder, Journal of Fluid Mechanics, Vol. 46, pp.321335. Anagnostopoulos, P. and Iliadis, G. (1996) Numerical study of the blockage effects on viscous flow past a circular cylinder, International Journal for Numerical Methods in Fluids, Vol. 22, pp.10611074. ANSYS-CFX-10 (2006) Solver Manual, ANSYS Inc. Barth, T.J. and Jespersen, D.C. (1989) The Design and Application of Upwind Schemes on Unstructured Meshes, AIAA Paper No. 89-0366. Behr, M., Hastreiter, D.D., Mittal, S. and Tezduyar, T.E. (1995) Incompressible flow past a circular cylinder: dependence of the computed flow field o the location of lateral boundaries, Computer Methods in Applied Mechanics and Engineering, Vol. 123, pp. 309316. Ben Richou, A., Ambari, A., Lebey, M. and Naciri, J.K. (2005) Drag force on a circular cylinder midway between two parallel plates at Re<<1 Part 2: moving uniformly (numerical and experimental), Chemical Engineering Science, Vol. 60, pp.25352543. Bohl, W. and Elmendorf, W. (2005) Technische Strmungslehre, 13th ed., Vogel, Wrzburg. Braack, M. and Richter, T. (2006) Solutions of 3D Navier-Stokes benchmark problems with adaptive finite elements, Computers & Fluids, Vol. 35, pp.372392. Braza, M., Chassaing, P. and Minh, H.Ha. (1986) Numerical study and physical analysis of the pressure and velocity fields in the near wake of a circular cylinder, Journal of Fluid Mechanics, Vol. 165, pp.79130. Breuer, M. (1998a) Large eddy simulation of the subcritical flow past a circular cylinder: numerical and modelling aspects, International Journal for Numerical Methods in Fluids, Vol. 28, pp.12811302. Breuer, M. (1998b) Numerical and modelling influences on large eddy simulations for the flow past a circular cylinder, International Journal of Heat and Fluid Flow, Vol. 19, pp.512521. Celik, I. and Shaffer, F.D. (1995)Long time-averaged solutions of turbulent flow past a circular cylinder, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 56, pp.185212. Ding, H., Shu, C., Yeo, K.S. and Xu, D. (2004) Simulation of incompressible viscous flows pasta circular cylinder by hybrid FD scheme and meshless least square-based finite difference method, Computer Methods in Applied Mechanics and Engineering, Vol. 193, pp.727744. Fluent 6.2 (2006) Users Guide, Fluent Inc.

The predicted time-averaged drag coefficients are compared with the experimental value in Figure 17. In addition to the graphical representation, the predicted drag coefficients, and, in brackets, their percentage deviations from the experimental value are also written above each column. One can see that LES provides an excellent agreement with the experiments, whereas the DES prediction comes also very close to the measured value. As 3D URANS can still be considered to provide a fair agreement with the measurements, discrepancies of 2D RANS and 2D URANS from the measured value are observed to be quite large.
Figure 17 Predicted time-averaged drag coefficients compared with the experimental value (Re = 104)

Conclusions

It is shown that the drag crisis can qualitatively be predicted by 2D RANS, provided that the near-wall region is adequately resolved. The 2D RANS underpredicts the experimental drag coefficients for the most Reynolds numbers due to the missing contribution of the organised transient motion. The 2D URANS on the other hand, generally overpredict the drag coefficient, indicating the relationship between the organised transience and three-dimensionality. Computations by three-dimensional transient procedures delivered a substantial improvement in the prediction quality for Re = 104. The best agreement with

Modelling turbulent flow past a circular cylinder by RANS, URANS, LES and DES
Franke, R. (1991) Numerische Berechnung der instationren Wirbelablsung hinter zylindrischen Koerpern, PhD Dissertation, Department of Civil Engineering, University of Karlsruhe, Germany. Franke, J. and Frank, W. (2002) Large eddy simulation of the flow past a circular cylinder at ReD = 3900, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 90, pp.11911206. Fujisawa, N., Asano, Y., Arakawa, C. and Hashimoto, T. (2005) Computational and experimental study of flow around a rotationally oscillating circular cylinder in uniform flow, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 93, pp.137153. Hinze, J.O. (1975) Turbulence, 2nd ed., McGraw-Hill, New York. Hunt, J.C.R., Wray, A.A. and Moin, P. (1988) Eddies, Stream and Convergence Zones in Turbulent Flows, Report CTR-S88, Center for Turbulence Research, Stanford University. Ishii, K., Kuwahara, K., Ogawa, S., Chyu, W.J. and Kawamura, T. (1985) Computation of flow around a circular cylinder in a super critical regime, AIAA 18th Fluid Dynamics, Plasmadynamics and Lasers Conf., AIAA Paper 851660, Cincinatti, Ohio 1618 July, pp.125132. Jain, P.C. and Goel, B.S. (1976) Shedding of vortices behind a circular cylinder, Computers & Fluids, Vol. 4, pp.137142. Jordan, S.K. and Fromm, J.E. (1972) Oscillatory drag, lift and torque on a cylinder in a uniform flow, The Physics of Fluids, Vol. 15, pp.371376. Kato, C. and Ikegawa, M. (1991) Large eddy simulation of unsteady turbulent wake of a circular cylinder using the finite element method, Proceedings of the Symposium on Advances in Numerical Simulation of Turbulent Flows, First Joint ASME-JSME Fluids Engineering Conference, Portland, Oregon, 2327 June. Launder, B.E. and Spalding, D.B. (1974) The numerical computation of turbulentflows, Computer Methods in Applied Mechanics and Engineering, Vol. 3, pp.269289. Lienhard, J.H. (1966) Synopsis of Lift, Drag and Vortex Frequency Data for Rigid Circular Cylinders, Bulletin 300, College of Engineering Research Division, Technical Extension Service, Washington State University, Pullman, Washington. Lbcke, H., Schmidt, St., Rung, T. and Thiele, F. (2001)Comparison of LES and RANS in bluff-body flows, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 89, pp.14711485. Majumdar, S. and Rodi, W. (1985) Numerical calculation of turbulent flow past circular cylinders, Proceedings of the Third Symposium on Physical Aspects of Aerodynamic Flows, Long Beach, California, 2124 January.

307

Menter, F.R. (1994) Two equation eddy-viscosity turbulence models for engineering applications, AIAA Journal, Vol. 32, pp.15981695. Nieuwstadt, F. and Keller, H.B. (1973) Viscous flow past circular cylinders, Computers & Fluids, Vol. 1, pp.5971. Nobari, M.R.H. and Naderan, H. (2006) A numerical study of flow past a cylinder with cross flow and inline oscillation, Computers & Fluids, Vol. 35, pp.393415. Peyret, R. (1996) Handbook of Computational Fluid Mechanics, Academic Press, San Diego. Sagaut, P. (2002) Large Eddy Simulation for Incompressible Flows An Introduction, 2nd Ed., Springer, Berlin. Sahin, M. and Owens, R.G. (2004) A numerical investigation of wall effects up to high blockage ratios on two-dimensional flow past a confined circular cylinder, Physics of Fluids, Vol. 16, pp.13051320. Shu, C., Qu, K., Niu, X.D. and Chew, Y.T. (2005) Numerical simulation of flow past a rotational circular cylinder by Taylor-series-expansion and least squares-based lattice Boltzmann method, International Journal of Modern Physics C, Vol. 16, pp.17531770. Smagorinsky, J. (1963) General circulation experiments with the primitive equations. I: the basic experiment, Month. Weath. Rev., Vol. 91, pp.99164. Song, C.C.S. and Yuan, M. (1989) Simulation of vortex shedding flow about a circular cylinder at high Reynolds numbers, ASME Journal of Fluids Engineering, Vol. 112, pp.155163. Streeter, V.L. and Wylie, E.B. (1975) Fluid Mechanics, 6th ed. McGraw-Hill, London. Strelets, M. (2001) Detached eddy simulation of massively separated flows, Proceedings of the 39th AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 2001-0879, 811 January, Reno, NV. Tamura, H., Kiya, M. and Arie, M. (1980) Numerical study on viscous shear flow past a circular cylinder, Bulletin of the JSME, Vol. 23, pp.19521958. Tamura, T., Ohta, I. and Kuwahara, K. (1990) On the reliability of two-dimensional simulation for unsteady flows around a cylinder-type structure, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 35, pp.275298. Travin, A., Shur, M., Strelets, M. and Spalart, P. (1999) Detached-eddy simulations past a circular cylinder, Flow, Turbulence and Combustion, Vol. 63, pp.293313. von Krmn, Th. (1911) ber den mechanismus des widerstandes, den ein bewegter krper in einer flssigkeit erzeugt, Nachr. Ges. Wiss. Gttingen, Math. Phys. Klasse, pp.509517.