Generative Adversarial Networks Guided Lightweight Design Based on Shakedown Strength Constraint

Abstract

Abstract Compared to the classically performed elastic limit rule in structural lightweight optimization design, the shakedown analysis can help determine the non-failure external load region which is greater than the elastic limit but less than the plastic limit, without accounting for the loading history. By sacrificing some abundant load-bearing capacity, a lightweight design of the structure can be achieved. Combining the shakedown analysis with parameter optimization can help resolve parameter optimization intervals while considering structural fatigue in complex loading conditions. This approach ultimately translates into solving a max-min optimization problem. However, due to the nonlinear relationship between the shakedown limit and massive design parameters, a rugged ‘performance landscape’ which is considered to be extremely unfriendly will be led through a traditional surrogate model, sometimes it is not even possible to generate. Moreover, the total time-consuming of simulation data acquisition is unacceptable. The current research utilizes Generative Adversarial Networks (GANs) to enrich the surrogate model with shakedown strength boundary constraints. Benefiting from the unsupervised learning feature and the similarity between the generated data and the real data, GANs are able to build relationships in the above optimization problem quickly, which is an innovative and executable technique to the lightweight design under structural strength constraints. The research first presents the shakedown theory and its numerical implementation. Then, a shakedown strength constrained for mass minimization form and the GANs flowchart for solving this problem are built, and a benchmark example is presented to validate the suggested technique. By using this effective workflow combining shakedown theory and GANs to optimize lightweight frame structures, the structure was able to quickly achieve a lighter configuration design while avoiding failure forms like incremental collapse, ratcheting, and alternate plasticity. Furthermore, the method is used to determine an optimal lightweight design for a manned airtight module. The parametric study presented in the paper demonstrated the effectiveness of the proposed method in reducing the total mass of the structure. Moreover, the study highlights the improvement in design performance by allowing redundant shakedown load capacity and reducing material usage.