Machine learning-augmented turbulence modeling for RANS simulations of massively separated flows

Abstract

We combine data assimilation and machine learning to correct the RANS Spalart-Allmaras turbulence model. The final neural-network contribution is a Boussinesq-correction, rather than a turbulent eddy-viscosity adjustment. Flows over periodic hills at distinct Reynolds numbers and geometries were selected to demonstrate the potential gain of machine learning-augmented turbulence models.