Race Car Flow Field Analysis using Autoencoders and Clustering

抄録

ABSTRACT: The aerodynamic development process of a racing car involves the generation of a great amount of data from numerical investigations. A Convolutional Autoencoder (CAE) architecture is applied to optimize the aerodynamic analysis workflow. In this study, flow fields obtained from Reynolds Averaged Navier Stokes (RANS) simulations serve as input for dimensionality reduction and clustering methods. The objective is to relate variations in flow topology to changes of corresponding performance metrics, aiming for an improved understanding of predominant fluidic phenomena. Consequently, inferences of aerodynamically relevant zones around the vehicle provide meaningful insights for future aerodynamic development.