AI-driven generative and reinforcement learning for mechanical optimization of 2D patterned hollow structures
-
Graphical Abstract
-
Abstract
2D patterned hollow structures have emerged as advanced materials with exceptional mechanical properties and lightweight characteristics, making them ideal for high-performance applications in aerospace and automotive industries. However, optimizing their structural design to achieve uniform stress distribution and minimize stress concentration remains a significant challenge due to the complex interplay between geometric patterns and mechanical performance. In this study, we develop an integrated framework combining conditional generative adversarial networks (cGANs) and deep Q-networks (DQNs) to predict and optimize the stress fields of 2D-PHS. We generated a comprehensive dataet comprising 1000 samples across five distinct density classes using a custom grid pattern generation algorithm, ensuring a wide range of structural variations. The cGAN accurately predicts stress distributions, achieving a high correlation with finite element analysis (FEA) results while reducing computational time from approximately 40 s (FEA) to just 1–2 s per prediction. Concurrently, the DQN optimizes design parameters through scaling and rotation operations, enhancing structural performance based on predicted stress metrics. Our approach resulted in a 4.3% improvement in average stress uniformity and a 23.1% reduction in maximum stress concentration. These improvements were validated through FEA simulations and experimental tensile tests on 3D-printed thermoplastic polyurethane samples. The tensile strength of the optimized samples increased from an initial average of 5.9–6.6 MPa under 100% strain, demonstrating enhanced mechanical resilience. This study demonstrates the efficacy of combining advanced AI techniques for rapid and precise material design optimization, providing a scalable and cost-effective solution for developing superior lightweight materials with tailored mechanical properties for critical engineering applications.
-
-