Modeling subgrid-scale force and divergence of heat flux of compressible isotropic turbulence by artificial neural network

Xie, Chenyue1; Li, Ke2; Ma, Chao3; Wang, Jianchun1

2019-10-16

10.1103/PhysRevFluids.4.104605

Physical Review Fluids   影响因子和分区

2469-990X

In this paper, the subgrid-scale (SGS) force and the divergence of SGS heat flux of compressible isotropic turbulence are modeled directly by an artificial neural network (ANN), which serves as a data-driven SGS modeling tool for large-eddy simulations (LESs). The unclosed SGS force and divergence of SGS heat flux are modeled based on the local stencil geometry with Galilean invariance. The input features include the first-order and second-order derivatives of filtered velocity and temperature, filtered density, and its first-order derivative. It is shown that the proposed ANN-F7 model shows an advantage over the gradient model in the a priori test. Specifically, the ANN-F7 model gives larger correlation coefficients and smaller relative errors than the gradient model. In an a posteriori analysis, the ANN-F7 model performs better than the dynamic Smagorinsky model (DSM) and dynamic mixed model (DMM) in the prediction of the statistical properties of flow fields at the Taylor microscale Reynolds number Re-lambda ranging from 180 to 250. The DSM and DMM models lead to the typical tilted spectral distribution of velocity, where low wave numbers are too energy rich, while those near the cutoff are damped too strongly. In contrast, it is shown that the velocity spectrum predicted by the ANN-F7 model almost overlaps with the filtered direct numerical simulation data. Besides, the ANN-F7 model reconstructs the probability density functions of SGS force and divergence of SGS heat flux much better than the DSM and DMM models. An artificial neural network with reasonable physical input features can deepen our understanding of turbulence modeling.

SCI ; EI

英语

第一 ; 通讯

Young Elite Scientist Sponsorship Program by CAST[2016QNRC001]

Physics

Physics, Fluids & Plasmas

WOS:000490483000004

AMER PHYSICAL SOC

20194507627991

Large eddy simulation ; Neural networks ; Probability density function ; Reynolds equation ; Reynolds number ; Turbulence ; Turbulent flow

Fluid Flow:631 ; Fluid Flow, General:631.1 ; Heat Transfer:641.2 ; Mathematics:921 ; Probability Theory:922.1

Web of Science

1.Southern Univ Sci & Technol, Dept Mech & Aerosp Engn, Ctr Complex Flows & Soft Matter Res, Shenzhen Key Lab Complex Aerosp Flows, Shenzhen 518055, Peoples R China
2.Chinese Acad Sci, Inst Computat Math & Sci Engn Comp, Beijing 100190, Peoples R China
3.Princeton Univ, Program Appl & Computat Math, Princeton, NJ 08544 USA

Xie, Chenyue,Li, Ke,Ma, Chao,et al. Modeling subgrid-scale force and divergence of heat flux of compressible isotropic turbulence by artificial neural network[J]. Physical Review Fluids,2019,4(10).

Xie, Chenyue,Li, Ke,Ma, Chao,&Wang, Jianchun.(2019).Modeling subgrid-scale force and divergence of heat flux of compressible isotropic turbulence by artificial neural network.Physical Review Fluids,4(10).

Xie, Chenyue,et al."Modeling subgrid-scale force and divergence of heat flux of compressible isotropic turbulence by artificial neural network".Physical Review Fluids 4.10(2019).