Implicit near-wall domain decomposition approach for large eddy simulation of turbulent flow separation

To mitigate the high computational cost associated with resolving small near-wall eddies in large eddy simulation (LES) while achieving acceptable accuracy, this work extends the implicit near-wall domain decomposition (INDD) method to a hybrid Reynolds-averaged Navier–Stokes (RANS)/LES zonal decomposition approach. In this framework, the LES solution is first computed using a Robin-type (slip) wall boundary condition on a coarse mesh. This solution is then iteratively corrected in the near-wall region based on an updated solution derived from simplified one-dimensional (1D) RANS equations. It is important that the composite solution is smooth thanks to the specially designed slip boundary conditions. The performance of the proposed approach is assessed for flow over a plain channel at different Reynolds numbers and for a periodic hill at ReH=10,595, using very coarse meshes. Applying the INDD technique to both flows results in a significant accuracy improvement compared to unresolved LES, demonstrating good generalization capability across different turbulent viscosity models, such as algebraic or RANS turbulence models. Meanwhile, when applying the k – l RANS model to channel flow cases, the quality of predictions by INDD remains relatively independent of interface locations. Compared to reference wall-resolved LES, the INDD technique significantly reduces the overall computational cost by using simplified RANS equations and more importantly, by requiring far fewer grid points.