Investigation of reinforcement learning applied to gradient voronoi energy absorption structures design

Voronoi structures can dissipate impact kinetic energy through plastic deformation. Their energy absorption capacity depends on design parameters like cell size and distribution. However, traditional trial-and-error design is inefficient. This study proposes an integrated Deep Reinforcement Learning (DRL) and Finite Element Method (FEM) framework to efficiently design high-performance Voronoi structures. Using Polylactic Acid (PLA), finite element simulations and collision experiments were conducted to validate the reliability of the numerical model under the investigated impact condition. A DRL+FEM framework was developed. A Double Deep Q-Network (DDQN) agent was trained to maximize energy absorption. Results showed that negatively graded Voronoi structures provide improved overall performance under the adopted optimization criterion. After optimization, the two-layer structure achieved a specific energy absorption and crushing force efficiency of 6.51. The three-layer structure saw improvements of 1.08 times and 1.71 times in these metrics, respectively, while the four-layer structure’s crushing force efficiency increased by 0.97 to 1.16 times compared to pre-optimization levels. The framework successfully reveals the influence of gradient characteristics.