TY - JOUR
T1 - Modeling subgrid-scale forces by spatial artificial neural networks in large eddy simulation of turbulence
AU - Xie, Chenyue
AU - Wang, Jianchun
AU - Weinan, E.
N1 - Publisher Copyright:
© 2020 American Physical Society.
PY - 2020/5
Y1 - 2020/5
N2 - Spatial artificial neural network (ANN) models are developed for subgrid-scale (SGS) forces in the large eddy simulation (LES) of turbulence. The input features are based on the first-order derivatives of the filtered velocity field at different spatial locations. The correlation coefficients of SGS forces predicted by the spatial artifical neural network (SANN) models with reasonable spatial stencil geometry can be made larger than 0.99 in an a priori analysis, and the relative error of SGS forces can be made smaller than 15%, much smaller than that of the traditional gradient model. In a posteriori analysis, a detailed comparison is made on the results of LES using the SANN model, implicit large eddy simulation (ILES), the dynamic Smagorinsky model (DSM), and the dynamic mixed model (DMM) at grid resolution of 643. It is shown that the SANN model performs better than the ILES, DSM, and DMM models in the prediction of the spectrum and other statistical properties of the velocity field, as well as the instantaneous flow structures. These results suggest that artificial neural network with consideration of spatial characteristics is a very effective tool for developing advanced SGS models in LES of turbulence.
AB - Spatial artificial neural network (ANN) models are developed for subgrid-scale (SGS) forces in the large eddy simulation (LES) of turbulence. The input features are based on the first-order derivatives of the filtered velocity field at different spatial locations. The correlation coefficients of SGS forces predicted by the spatial artifical neural network (SANN) models with reasonable spatial stencil geometry can be made larger than 0.99 in an a priori analysis, and the relative error of SGS forces can be made smaller than 15%, much smaller than that of the traditional gradient model. In a posteriori analysis, a detailed comparison is made on the results of LES using the SANN model, implicit large eddy simulation (ILES), the dynamic Smagorinsky model (DSM), and the dynamic mixed model (DMM) at grid resolution of 643. It is shown that the SANN model performs better than the ILES, DSM, and DMM models in the prediction of the spectrum and other statistical properties of the velocity field, as well as the instantaneous flow structures. These results suggest that artificial neural network with consideration of spatial characteristics is a very effective tool for developing advanced SGS models in LES of turbulence.
UR - http://www.scopus.com/inward/record.url?scp=85087783706&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85087783706&partnerID=8YFLogxK
U2 - 10.1103/PhysRevFluids.5.054606
DO - 10.1103/PhysRevFluids.5.054606
M3 - Article
AN - SCOPUS:85087783706
SN - 2469-990X
VL - 5
JO - Physical Review Fluids
JF - Physical Review Fluids
IS - 5
M1 - 054606
ER -