Wall-Modeled Large-Eddy Simulations of Flows With Curvature and Mild Separation

The performance of wall-modeled large-eddy simulation (WMLES) based on hybrid models, in which the inner region is modeled by Reynolds-averaged Navier–Stokes (RANS) equation and the outer region is resolved by large-eddy simulation (LES), can make the application of LES attainable at high Reynolds numbers. In previous work by various authors, it was found that in most cases a buffer region exists between the RANS and LES zones, in which the velocity gradient is too high; this leads to an inaccurate prediction of the skin-friction coefficient. Artificially perturbing the RANS∕LES interface has been demonstrated to be effective in removing the buffer region. In this work, WMLES has been performed with stochastic forcing at the interface, following the previous work by our group on two nonequilibrium complex flows. From the two flows studied, we conclude that the application of stochastic forcing results in improvements in the prediction of the skin-friction coefficient in the equilibrium regions of these flows, a better agreement with the experiments of the Reynolds stresses in the adverse pressure gradient and the recovery region, and a good agreement of the mean velocity field with experiments in the separation region. Some limitations of this method, especially in terms of CPU requirements, will be discussed.