Aerodynamic Optimization Design Using Two-Stage Optimization Based on BézierGAN Parameterization

This paper proposes an efficient aerodynamic optimization design method by combining BézierGAN with particle swarm optimization (PSO). It applies a two-stage optimization (TSO) method based on BézierGAN to improve the efficiency of aerodynamic optimization. BézierGAN integrates rational bézier curves parameterization with information maximizing generative adversarial networks (InfoGAN) and reduce the dimensionality of the design space by learning from existing airfoil databases. Essentially, BézierGAN is a geometric parameterization method that divides the design variables into two categories: latent code and noise variables. Based on the dual-parameter property of BézierGAN the two-stage optimization method optimizes the latent code and all design variables sequentially. The results demonstrate that BézierGAN can effectively reduce the dimensionality of the design space by representing the geometric shape with fewer design variables. The study of airfoil inverse design shows that the aerodynamic optimization design method based on BézierGAN and the two-stage optimization framework can reduce the iteration of optimization to improve the optimization efficiency.