A Comparison of DES Methods for the DrivAer Generic Realistic Car Model on a Wall Resolved and a Wall Function Mesh

The DrivAer realistic generic car model is now established as one of the benchmark geometries to assess the aerodynamic flow field characteristics associated with passenger vehicles. Since its introduction in 2012, the database of experimental studies has grown and provides excellent validation opportunities for analytical methods. This paper compares Computational Fluid Dynamics (CFD) simulations for integral forces, surface pressure distribution and velocity flow fields for the DrivAer model in the notchback configuration. Transient CFD data are obtained by employing hybrid Reynolds Averaged Navier-Stokes (RANS) and Large Eddy Simulation methods (Detached Eddy Simulation - DES) using the finite volume solvers Simcenter Star-CCM+ and the openFOAM based flow solver IconCFD. Computational results are calculated using Wall Resolved Meshes (WRM), where y+ < 1, and Wall Function Meshes (WFM), where 30 < y+ < 100. Results for both flow solvers are obtained using common meshes generated with Beta-CAE ANSA and with the native mesh generators of each flow solver. Sensitivities of the computational results to turbulence model and DES formulation are assessed by comparison with data measured on Ford’s full-scale DrivAer model in the Pininfarina Wind Tunnel. Qualitatively good agreement for the predicted pressure and velocity distribution around the vehicle is demonstrated when compared to physical tests. In general, predicted integral values for rear lift coefficient (C_Lr) are similar to measured test data, whereas integral values for drag coefficient (C_D) are over-predicted by up to 10% and front lift coefficient (C_Lf) are significantly under-predicted by up to 250%. Despite all predictions replicating the measured vehicle’s macroscopic flow field, there are notable differences between predicted trends that are related to flow solver and turbulence model selection and DES formulation. Sensitivities related to computational mesh strategy are less pronounced.