Aerodynamic shape optimization of hypersonic aircraft using data-driven generative nonlinear parameterization

Aerodynamic shape optimization of hypersonic vehicles is critically important yet profoundly challenging. The difficulties arise from the need to manage multiple competing objectives, complex three-dimensional geometries, and the extreme computational cost of high-fidelity aerodynamic simulations across subsonic, transonic, and hypersonic regimes. Despite recent advances, an effective global optimization strategy for hypersonic aircraft design remains limited, largely hindered by the curse of dimensionality. To remove this barrier, we propose a data-driven generative nonlinear shape parameterization framework for efficient aerodynamic design of hypersonic aircraft. This framework begins by constructing diverse hypersonic aircraft shapes that cover the feasible sub-domains of a high-dimensional design space. A linear dimension reduction method is used to transform the high-dimensional point-cloud database to a low-dimensional modal space. Subsequently, a nonlinear generative model is trained to learn the statistical distribution feature of the linear mode coefficients. The resulting generative latent space provides an efficient, low-dimensional, and expressive parameterization of aerodynamic shapes. The proposed method is validated in both single-point and multi-point optimization of hypersonic aircraft, demonstrating superior efficiency and effectiveness compared with conventional parameterization approaches. This study presents an efficient roadmap for aerodynamic shape parameterization and global optimization of next-generation aircraft.