A survey on generative methods in automotive aerodynamic shape optimization

Abstract Electric motors surpass internal combustion gasoline engines in energy efficiency. While air resistance is less critical for internal combustion vehicles, it significantly impacts electric vehicle energy consumption. However, existing aerodynamic drag-reduction technologies have not kept pace with the rapidly advancing electric vehicle industry. Over the past 5 years, notable advancements have been made in high-dimensional aerodynamic shape optimization, particularly as deep neural network-based methods have transitioned from laboratory research to practical engineering applications in the aeronautical sector. This paper explores cutting-edge technologies and potential applications within the automotive industry. The study focuses on two main aspects of generative form-optimization methodologies: shape representation and shape-generation techniques. Shape representation methods are categorized and analyzed from three perspectives: dimensionality reduction, mesh-based approaches, and point-cloud-based methods. Additionally, generative methodologies are examined, with an emphasis on variational autoencoders, generative adversarial networks, and deep learning architectures that incorporate attention mechanisms. This review provides a comprehensive overview of current and emerging artificial intelligence-based optimization strategies for reducing aerodynamic drag in vehicle design.