Prediction of turbulent channel flow using Fourier neural operator-based machine-learning strategy

The implicit U-Net enhanced Fourier neural operator (IUFNO) combines the loop structure of implicit FNO (IFNO) with U-Net, leading to enhanced long-term predictive ability in the large-eddy simulations (LES) of turbulent channel flow. It is found that the IUFNO outperforms the traditional dynamic Smagorinsky model (DSM) and the wall-adapted local eddy-viscosity (WALE) model at coarse LES grids. The predictions of both the mean and fluctuating quantities by IUFNO are closer to the filtered direct numerical simulation (fDNS) benchmark compared to the traditional LES models, while the computational cost of IUFNO is much lower.