Data-driven Reynolds Stress Modeling Using High Fidelity Flow Database

Abstract

In this study, Reynolds stress modeling using sparse dataset was tested to check the improvement of Reynolds-averaged Navier-Stokes (RANS) modeled Reynolds stress. The Reynolds stress modeling used in the study is to construct a regression function using machine learning algorithms. Inputs of the regression, flow features are computed from RANS modeled variables. Outputs of the regression, Reynolds stress discrepancies between RANS modeled Reynolds stress and high-fidelity simulation modeled Reynolds stress is used. Periodic hill is selected for baseline RANS simulations. The random forest algorithm is selected for the regression function. Reynolds stress modeling using sparse dataset shows good agreement of predicted Reynolds stress compared to high-fidelity simulation modeled Reynolds stress.