A unified method of data assimilation and turbulence modeling for separated flows at high Reynolds numbers

In recent years, machine learning methods represented by deep neural networks (DNNs) have been a new paradigm of turbulence modeling. However, in the scenario high Reynolds numbers, there are still some bottlenecks, including lack high-fidelity data and stability problem coupling process models Reynolds-averaged Navier–Stokes (RANS) solvers. this paper, we propose an improved ensemble Kalman inversion method as unified approach assimilation modeling for separated flows at numbers. A novel design based on transfer regularizing strategy proposed to improve method. The trainable parameters DNN optimized according given experimental surface pressure coefficients framework mutual between RANS solvers eddy viscosity models. way, model training integrated into one step get agree well with experiments directly. effectiveness is verified cases around S809 airfoil Through few states, can generalizing both attached different angles attack, which also perform robustness. errors lift attack significantly reduced more than three times compared traditional Spalart–Allmaras model.