A FDQN-based tolerance closed-loop optimization model for thin-wall components in aircraft assembly

With the cross- generational development of aircraft products, the leap in assembly accuracy and performance has also brought about a significant increase in manufacturing costs. Tolerance optimization is a key research direction to reduce cost and increase efficiency without sacrificing assembly accuracy and performance requirements. However, when confronting complex deviation relationships and nonlinear optimization in aircraft assembly, the tolerance optimization process is constrained by factors such as large sampling volumes, high-frequency iterative calculations, relative independence, and low intelligence levels, gradually revealing its limitations. To address these challenges, a tolerance closed-loop optimization model (TCOM) based on a High Resolution-Precision Generative Adversarial Network (HiRes-PreciGAN) and improved Fick Deep $Q$-networks (FDQN) is proposed in this study. The model incorporates Latin Hypercube Sampling (LHS) and HiRes-PreciGAN to achieve efficient tolerance analysis of rigid and flexible assembly deviations. On this basis, the model synthesizes several optimization objectives, and uses FDQN to deeply optimize the tolerance allocation scheme to maximize the cost-effectiveness. In addition, an exploration strategy based on physical inspiration is designed to achieve nonlinear regulation through Fick diffusion mechanism to improve the convergence of model optimization. Through a specific industrial case, the proposed model is evaluated from the aspects of optimization performance and optimization scheme. The evaluation results show that the improved exploration strategy has a 12.36 % improvement in convergence compared with the previous epsilon-greedy strategy, and is superior to any single meta-heuristic algorithm when dealing with the tolerance optimization problem of aircraft assembly. In terms of tolerance optimization, the model significantly improves several key indicators related to cost-effectiveness. This study provides a new idea to tolerance optimization that aims to reduce costs and enhance efficiency, and facilitate the intelligent transformation in high-performance assembly for the aerospace and other fields.