Large eddy simulations of a circular cylinder at Reynolds numbers surrounding the drag crisis

Large eddy simulations of the flow around a circular cylinder at high Reynolds numbers are reported. Five Reynolds numbers were chosen, such that the drag crisis was captured. A total of 18 cases were computed to investigate the effect of gridding strategy, turbulence modelling, numerical schemes and domain width on the results. It was found that unstructured grids provide better resolution of key flow features, when a 'reasonable' grid size is to be maintained. When using coarse grids for large eddy simulation, the effect of turbulence models and numerical schemes becomes more pronounced. The dynamic mixed Smagorinsky model was found to be superior to the Smagorinsky model, since the model coefficient is allowed to dynamically adjust based on the local flow and grid size. A blended upwind-central convection scheme was also found to provide the best accuracy, since a fully central scheme exhibits artificial wiggles, due to dispersion errors, which pollute the solution. Mean drag, fluctuating lift Strouhal number and base pressure are compared to experiments and empirical estimates for Reynolds numbers ranging from 6.31 x 10(4) to 5.06 x 10(5). In terms of the drag coefficient, the drag crisis is well captured by the present simulations, although the other integral quantities (rms lift and Strouhal number) show larger discrepancies. For the lowest Reynolds number, the drag is seen to be more sensitive to the domain width than the spanwise grid spacing, while at the higher Reynolds numbers the grid resolution plays a more important role, due to the larger extent of the turbulent boundary layer. (C) 2015 Elsevier Ltd. All rights reserved.