关键词: 生成式模型, 飞行器设计, 气动布局设计, 生成对抗网络, 空气动力学

Abstract: Aerodynamic configuration design plays a critical role in the conceptual phase of aircraft development, where diverse design options must be efficiently generated and evaluated to identify optimal configurations. However, conventional aerodynamic design methodologies heavily rely on expert experience, which limits the exploration of innovative configurations and restricts the diversity of feasible solutions. Meanwhile, mainstream aerodynamic optimization approaches typically commence with the refinement of initial design stages, concentrating on local optimization within predefined configurations, rather than exploring the diversity of aerodynamic layout alternatives during the conceptual design phase.Recently, generative artificial intelligence has offered a novel solution paradigm for aircraft configuration design.This study presents a generative aerodynamic configuration design methodology based on Generative Adversarial Network (GAN), with application to a small general aviation aircraft. First, the GAN and its variants are evaluated for their capability in representing parametric space and generating aerodynamic configurations. Based on this analysis, a Conditional Wasserstein GAN with Gradient Penalty (CWGAN-GP) is established for efficient generation of aerodynamic configurations. Second, for wing design under low-speed cruise conditions (Mach number 0.2, angle of attack 2 degree), the representation capability of different generative models in the parameter space and their ability to generate configurations under given conditions were analyzed, demonstrating the advantages of CWGAN-GP in generative design. Finally, for the conceptual aerodynamic configuration design of a general aviation aircraft under typical cruise conditions (Mach number 0.6, angle of attack 2 degree), the CWGAN-GP model successfully generates a variety of aerodynamic configurations using the lift coefficient and lift-to-drag ratio as conditional parameters. Furthermore, by extending beyond the training parameter ranges, the proposed method exhibits strong generalization and extrapolation capabilities, capable of producing different aerodynamic shapes outside the original condition space. These results indicate that the generative design methodology presented offers a promising new paradigm for next-generation aircraft aerodynamic configuration exploration and innovation.

Key words: generative models, aircraft design, aerodynamic configuration design, GAN, aerodynamics