A parameterized-loading driven inverse design and multi-objective coupling optimization method for turbine blade based on deep learning

Abstract

Inverse design is an important part of the initial design stage of blade profile, while the traditional inverse design methods are highly dependent on design experience and physical model simulation, resulting in long design cycles. To accelerate the inverse design, this paper proposes an inverse design method based on load distribution by combining the Variational Auto-Encoder model with the Generative Adversarial Network (VAE-GAN) and U-Net (U-Net1 and U-Net2) using a multi-objective genetic algorithm (GA). The inverse design process is divided into three stages. First, the VAE-GAN model is combined with the GA to generate a batch of load distributions that match the target form. Then the U-Net1 network is used to establish the connection between the load distribution and the blade aerodynamic performance as well as the geometric parameters. The optimal load distribution under multi-objective constraints is filtered by designing the fitness function of the GA. Finally, the blade profile corresponding to the optimal load distribution is inversely designed by the U-Net2 network. Numerical simulation results show that the method can rapidly generate the blade profile corresponding to the specified load distribution, and realize the automatic optimization of aerodynamic performance under certain geometric constraints in the inverse design process.