Modeling the Pressure-Strain-Correlation in Differential Reynolds Stress Models using Feature Engineering and a Genetic Evolution Algorithm

Kurzfassung

Simulating turbulent flows using Reynolds-Averaged-Navier-Stokes-Equations (RANS) models is a very efficient approach for most practical applications, as they aim to produce accurate results, while being more performant than scale-resolving methods. However all RANS approaches require the modeling of unknown terms, one of them being the pressure-strain-correlation (PSC) in the differential Reynolds Stress Model. A state-of-the-art model for this is the Speziale-Sarkar-Gatski (SSG) model, which tries to represent the PSC term using a set of basis tensors and (in its simplest form) five coefficients that need to be calibrated to achieve the desired behavior. These coefficients are usually assumed to be constant, however, theoretical considerations suggest they should be functions of the current flow-state, considerably increasing the models complexity and capabilites. A feature engineering process for a genetic evolution programming optimization is proposed to determine functions improving the original models results for flows with adverse pressure gradients.