Construction of a data-driven turbulence model for automotive external flow simulation

To address the insufficient accuracy of traditional Reynolds-averaged Navier-Stokes (RANS) turbulence models in predicting automotive separated flows, a method combining conditional flow field inversion and symbolic regression is proposed to modify the baseline SST turbulence model. Using the NASA Hump and curved backward-facing step (CBFS) cases as the training set, an SST-CND (shear-stress transport-conditional non-linear dissipation) modified model is developed. The model is subsequently validated using two standard automotive aerodynamic models, namely the Ahmed body and the SAE reference model. The results show that the SST-CND modified model improves the prediction accuracy of separated flows over both the Ahmed body and the SAE reference model. The drag coefficient error for the SAE model is reduced to less than 4%. Compared with the baseline SST model, the SST-CND model captures a wake vortex structure in better agreement with experimental results. Additionally, the conditional flow field inversion method effectively confines the corrections to regions outside the attached boundary layer, preserving the simulation accuracy of the baseline model for attached flows.