Data-Enabled Parameter Modification for the Shear-Stress-Transport Model with Progressive Neural Networks

Turbulence models based on the Reynolds-averaged Navier–Stokes method are widely employed for simulation in engineering and research. Nonetheless, these models have some limitations in simulating flows with adverse pressure gradients due to the inclusion of various assumptions, such as the eddy viscosity hypothesis. As a result, many researchers have focused their efforts on improving turbulence models, including parameter calibration. In this paper, a progressive neural network framework is utilized to modify the dissipation coefficients of the shear-stress transport (SST) model into a function of physical quantities. Firstly, the zero pressure gradient flat plate case is employed to obtain the calibrated model SST-FP. Subsequently, the flow with a turbulent separation bubble is adopted to progressively acquire the calibrated model SST-TSB. The enhanced model performs well in predicting the mean velocity profile, friction coefficient, and other variables of training cases, owing to joint corrections on the k and ω equations. Furthermore, verification research demonstrates that the SST-TSB model mitigates the potential damage to the wall-law prediction. It can also adapt the forecast properly in circumstances where the original model predicts larger or smaller separation zones, avoiding the drawbacks that Bayesian inference and other methods have when applied to parameter calibration.