Large Eddy Simulation of Turbulent Flow over a Backward-Facing Step

This work studies the effect of grid resolution and subgrid scale modeling on the predictive accuracy of Large Eddy Simulation. In particular, the problems considered are: turbulent flow over a backward-facing step and fully-developed turbulent channel flow. A SGS-free model and four subgrid scale models are used: Smagorsinky, k-equation, dynamic k-equation and WALE. Different combinations of streamwise and spanwise grid resolutions are considered along with different strategies for the cell-size distribution in the wall-normal direction. First, a detailed study pertaining to fully-developed turbulent channel flow is presented for target friction Reynolds numbers 180 and 300. A symmetric three-layered wall-normal meshing strategy in conjunction with a SGS-free model is shown to be the best compromise between computational efficiency and agreement with benchmark data. Overall, the accuracy of the results was observed to be most sensitive to the spanwise resolution of the grid. Cell sizes in wall units ranging in between 25-28 and 12-20 in the streamwise and spanwise directions, respectively, yielded excellent agreement with reference data. Second, a detailed study of turbulent flow over a backward-facing step at a step-height Reynolds number of 5100 is conducted. This includes analysis of first- and second-order statistical moments of the velocity field for three different grid refinement levels and several SGS models. Additional quantities are computed on the finest grid, such as high-order statistics, one-dimensional spanwise power spectral density and spatial autocorrelation of the velocity, turbulent kinetic energy budget, as well as the first- and second-order statistics of the vorticity field. In line with the channel flow study, the WALE and SGS-free models produce the best results when compared with experimental and numerical benchmark data. All presented datasets are made available online for public use