Effects of modelling, resolution and anisotropy of subgrid-scales on large eddy simulations of channel flow

Abstract

In this paper, the effect of subgrid-scale (SGS) modelling, grid resolution and anisotropy of the subgrid-scales on large eddy simulation (LES) is investigated. LES of turbulent channel flow is performed at Re _τ=934, based on friction velocity and channel half width, for a wide range of resolutions. The dynamic Smagorinsky model (DS), the high-pass filtered dynamic Smagorinsky model (HPF) based on the variational multiscale method and the recent explicit algebraic model (EA), which accounts for the anisotropy of the SGS stresses are considered.

The first part of the paper is focused on the resolution effects on LES, where the performances of the three SGS models at different resolutions are compared to direct numerical simulation (DNS) results. The results show that LES using eddy viscosity SGS models is very sensitive to resolution. At coarse resolutions, LES with the DS and the HPF models deviate considerably from DNS, whereas LES with the EA model still gives reasonable results. Further analysis shows that the two former models do not accurately predict the SGS dissipation near the wall, while the latter does, even at coarse resolutions. In the second part, the effect of SGS modelling on LES predictions of near-wall and outer-layer turbulent structures is discussed. It is found that different models predict near-wall turbulent structures of different sizes. Analysis of the spectra shows that although near-wall scales are not resolved at coarse resolutions, large-scale motions can be reasonably captured in LES using all the tested models.

Keywords:

• large eddy simulation
• subgrid-scale modelling
• resolution effects
• subgrid-scale anisotropy
• channel flow

Acknowledgements

Support from the Swedish research council through grant numbers 621-2010-6965 and 621-2007-4232 and computer time provided by the Swedish National Infrastructure for Computing (SNIC) are gratefully acknowledged. The authors would also like to thank Prof. Javier Jiménez for providing DNS data at Re _τ=934.