Constructing subgrid-scale models for turbulent channel flows

Subgrid-scale (SGS) models for turbulent channel flows are developed using direct numerical simulation (DNS) data. Large-eddy simulation (LES) offers a substantial reduction in computational cost compared to DNS. However, LES prediction at coarse grid resolutions–where it is practically useful–shows sensitivity to SGS models. Thus, improving the affordability and reliability of LES hinges on improved SGS modeling strategies. For wall-bounded turbulence at high Reynolds numbers, the complex structures of boundary layers render the conventional modeling based on isotropic turbulence less effective. In this study, a family of SGS models is constructed using DNS datasets of turbulent channel flows at low to intermediate friction Reynolds numbers. A recently developed wavelet-based procedure, combined with a tensor representation theory, is employed to identify the SGS model forms that are spectrally optimal with respect to the DNS data. A priori validations are conducted for spectral energy balance, and the wall-normal variation of the model structure is examined.