Benchmark of Aerodynamic Surrogate Models on Varying Training Datasets

AI-based methods show huge potential assisting engineers in their pursuit to further improve vehicle aerodynamics. To aid future model selections it is important to test the performance of different surrogate models based on the provided training data size and shape. In this study, data of the simulated flow around the 2D NACA 8810 airfoil is used. To achieve the desired amount of varying data for the training and validation datasets the closure coefficients of the kw-SST RANS turbulence model are varied by Design of Experiment. Different levels of dimensional reduction are generated for each dataset by using Principal Component Analysis.     «

AI-based methods show huge potential assisting engineers in their pursuit to further improve vehicle aerodynamics. To aid future model selections it is important to test the performance of different surrogate models based on the provided training data size and shape. In this study, data of the simulated flow around the 2D NACA 8810 airfoil is used. To achieve the desired amount of varying data for the training and validation datasets the closure coefficients of the kw-SST RANS turbulence model...     »