关键词: 多钉铆接结构, 条件生成对抗网络, 结构优化设计, 轻量化, Wasserstein距离

Abstract: Multi-rivet fastening is one of the key technologies for aircraft structures, exerting a crucial influence on the overall load-bearing capacity of aircraft structures. Traditional optimization methods have limitations in balancing high-dimensional data processing, computational cost control and engineering applicability. This paper proposes a multi-rivet joint structural design optimization method based on Conditional Generative Adversarial Networks (CGAN). By integrating physical laws, lightweight objectives and engineering constraints into the deep learning model, efficient collaborative optimization is achieved. Aiming at the defect that pure data-driven models tend to violate mechanical principles, the multi-rivet load distribution and shear failure criteria are embedded into the loss function of the CGAN generator. To realize structural lightweight design, the minimization of the multi-rivet structure mass is considered when constructing the generator loss function. In addition, engineering manufacturing constraints are incorporated into the neural network model to ensure that the generated design parameters meet the requirements of manufacturing feasibility. A finite element model based on the Johnson-Cook constitutive model and failure criterion is established. After calibration with test results, a "test - simulation" integrated dataset is formed, covering a wider range of parameters. The Wasserstein distance is introduced to prevent gradient vanishing or explosion, thus improving the stability of neural network training. The training results show that the generator and the discriminator can achieve stable adversarial equilibrium, and both the strength and weight loss functions converge, which verifies the model's ability of collaborative optimization for the "strength-lightweight" multi-objective. Under the tensile condition of the multi-rivet joint structure, by automatically optimizing the rivet diameter, arrangement parameters and the thickness of connected components, the structural weight reduction of more than 9% is achieved while satisfying the strength design requirements, and the optimized design scheme can be generated within 300ms. The research shows that the CGAN-based optimization method proposed in this paper can rapidly generate structurally optimized design schemes that are in line with engineering practice, greatly shorten the design cycle, reduce the R&D cost, and provide a new technical means for the high-performance and lightweight design of aircraft multi-rivet joint structures.

Key words: Multi-rivet joint structure, Conditional Generative Adversarial Network, Structural optimization design, Lightweight, Wasserstein distance