An approach to wall modeling in large-eddy simulations

TLDR

Large‑eddy simulation of high‑Reynolds‑number channel flow up to Reτ≈80 000 is performed at moderate cost using a detached‑eddy subgrid‑scale model. The approach incorporates wall modeling and a grid that scales logarithmically with Reynolds number, without flow‑specific tuning. The three codes agree, but show a ~15 % underprediction of skin‑friction and a mismatch between modeled and resolved logarithmic velocity layers, with trade‑offs between viscous, modeled, and resolved shear stresses as Reynolds number and grid refinement vary.

Abstract

Channel flow with friction Reynolds number Reτ as high as 80 000 is treated by large-eddy simulation at a moderate cost, using the subgrid-scale model designed for detached-eddy simulations. It includes wall modeling, and was not adjusted for this flow. The grid count scales with the logarithm of the Reynolds number. Three independent codes are in fair agreement with each other. Reynolds-number variations and grid refinement cause trades between viscous, modeled, and resolved shear stresses. The skin-friction coefficient is too low, on the order of 15%. The velocity profiles contain a “modeled” logarithmic layer near the wall and some suggest a “resolved” logarithmic layer farther up, but the two layers have a mismatch of several units in U+.