Large-Eddy and Wall-Modeled Simulations of Turbulent Flow Over Two-dimensional River Dunes

Thumbnail Image
Mellati Nokhandan, Mojtaba
Hybrid RANS/LES , Two-Dimensional Dunes , Wall-Modeling , DES
The Large-Eddy Simulation (LES) technique is only applicable to wall-bounded flows at moderate Reynolds numbers due to its computational cost. Wall-layer models are designed to extend the LES concept to engineering flows at high Reynolds numbers by bypassing the near-wall region and modeling its effects on the outer flow. In this work, large-eddy and hybrid Reynolds-Averaged Navier-Stokes (RANS)/LES calculations are performed for turbulent flow over a two-dimensional dune geometry. A Lagrangian dynamic eddy-viscosity model is applied for resolved LES computations. Also, Wall-Modeled LES (WMLES) calculations are performed using models of the Detached-Eddy Simulation (DES) family based on Spalart & Allmaras (1992) turbulence model, including DES, Delayed DES (DDES), and Improved DDES (IDDES). Simulations are carried out at two different Reynolds numbers, Re_b = 18 900 and 50 000, based on bulk velocity and average dune height. Wall-resolved LES calculations are used to investigate the accuracy of the present wall-layers. Compared to attached, equilibrium flows or flows with a mild separation, a higher accuracy was achieved in prediction of the mean and second order statistics using all the present hybrid methods. Large, three-dimensional eddies are generated due to the instability of the shear layer after flow separation at the dune crest. As they convect downstream with the mean flow, these eddies interact with the near-wall flow and act as a mechanism to increase the eddy-generation at the RANS/LES interface. As a result, the interface between the RANS and LES regions moves toward the bed and leads to a more realistic turbulence cycle in the RANS region. Considering both the first and second order statistics, as well as the flow physics included in the instantaneous turbulent field of the simulation, the IDDES model resulted in a higher level of accuracy compared to the DES and DDES methods at all the tested Reynolds numbers. However, the same grid density as the two other models might lead to noise propagations over the reattachment region. To resolve this issue it might be necessary to refine the mesh in wall-parallel directions (streamwise and spanwise directions) when using IDDES as the wall-layer model.
External DOI