Machine learning models for sub-grid-scale (SGS) term

Asahi, Yuichi   ; Maeyama, Shinya*; Fujii, Keisuke*

We have developed a deep-learning model to surrogate the effect of small-scale on large-scale fluctuations. We have constructed the sub-grid-scale (SGS) models based on the Mori-Zwanzig projection operatormethod and neural networks. We have performed large eddy simulations (LESs) of the Kuramoto-Sivashinsky turbulence with these SGS models. We have demonstrated that the time averaged energy spectrumof LESs agree with that of the dynamic numerical simulation (DNS).