kεNet湍流模型研究及其在低雷诺数槽道流中的应用

kεNet:discovering the turbulence model and applying for low Reynolds number turbulent channel flow

我们提出了一种基于物理信息的深度学习网络(kεNet),可用于RANS方程中发现封闭的湍流模型.kεNet由一个传统的典型神经网络结构和若干个基于物理信息的方程组成,如雷诺应力方程、k方程和ε方程.以低雷诺数下的槽道流动的湍流模型的修正为例,通过训练基于物理信息的神经网络,模型参数得到了修正.修正后的湍流模型参数应用于OpenFOAM软件进行计算,能够非常好地预测Re_(τ)=5200和2000下的槽道流动.

We propose a physics informed deep learning neural network,termed kεNet,for discovering the closure model of Reynolds stress in Reynolds-averaged Navier-Stokes(RANS)equations.The kεNet consists of a traditional physics uninformed neural network and several physics informed equations,e.g.,the formula of Reynolds stress,the transport equations of turbulence kinetic energy k and turbulence dissipation rateε.A low Reynolds number k-εcorrection for turbulent channel flow is considered as an example.Based on the kεNet,all model parameters of the RANS are calibrated simultaneously with the specified mathematical formula,through training the neural network with data from direct numerical simulation.The gained turbulence model is utilized in OpenFOAM,an open source computational fluid dynamics(CFD)code,and predicts reasonable good results for a turbulent channel flow with Re_(τ)=5200 and 2000.