Reduced-Order Modeling of Vehicle Aerodynamics via Proper Orthogonal Decomposition

Aerodynamic optimization of the exterior vehicle shape is a highly multidisciplinary task involving, among others, styling and aerodynamics. The often differing priorities of these two disciplines give rise to iterative loops between stylists and aerodynamicists. Reduced-order modeling (ROM) has the potential to shortcut these loops by enabling aerodynamic evaluations in real time. In this study, we aim to assess the performance of ROM via proper orthogonal decomposition (POD) for a real-life industrial test case, with focus on the achievable accuracy for the prediction of fields and aerodynamic coefficients. To that end, we create a training data set based on a six-dimensional parameterization of a Volkswagen passenger production car by computing 100 variants with Detached-Eddy simulations (DES). Based on this training data, we reduce the dimension of the solution space via POD and interpolate the base coefficients with Kriging (aka Gaussian Process Regression) for predictions of the flow field at unseen parameter combinations. The error analysis of the fields and drag coefficient predictions reveal that 100 training samples are sufficient for this six-dimensional test case in order to meet the necessary accuracy requirements for an application during the aerodynamic development process. We conclude that ROM via POD+Kriging at the core of an interactive aerodynamic design process enables aerodynamicists and stylists to find geometries which equally satisfy both aerodynamic and esthetic requirements in a joint session.