Artificial-neural-network-based subgrid-scale models in the strain-rate eigenframe for large-eddy simulation of compressible turbulent channel flow

In this paper, we propose artificial-neural-network-based (ANN-based) subgrid-scale (SGS) models in the strain-rate eigenframe for large-eddy simulation (LES) of compressible turbulent channel flow. In the ANN modeling, both input and output variables are transformed to the local eigenframe of the strain-rate tensor to ensure that the model is coordinate-invariant. In addition, the grid information is integrated into input variables of the nerual network to address the grid anisotropy of the turbulent channel flow. The proposed coordinate-invariant ANN model achieves a higher correlation coefficient and lower relative error than traditional SGS models, including the dynamic Smagorinsky model (DSM), the Vreman model, and the wall-adapting local eddy-viscosity (WALE) model, in an a priori test. In an a posteriori test, the proposed model performs better in predicting structures and statistics of velocity and temperature, and it also gives a more accurate prediction of SGS fluxes. Furthermore, the proposed model exhibits high accuracy in situations with an untrained Reynolds number and a Mach number, demonstrating its generalization ability.