Random Voronoi Lattice Design and Optimization with Reinforcement Learning

This paper presents a design framework for generating and optimizing heterogeneous lattice structures using reinforcement learning (RL). Inspired by natural selection and biological structures, the paper focuses on non-periodic, random-like lattices such as those generated via Voronoi seeding theory. These structures can offer more isotropic behavior and local adaptability, features commonly found in natural structures like bones and honeycombs. The design loop integrates lattice geometry generation, structural simulation, and machine learning in a fully automated loop. It uses the Proximal Policy Optimization (PPO) RL algorithm to guide the optimization process, enabling the agent to propose new lattice designs and evaluate them based on simulation results. Random Voronoi lattices are created, and their mechanical performance is assessed through finite element analysis and compression testing. Preliminary results indicate that the framework can successfully identify more optimal lattice designs that improve specific mechanical properties, such as elastic modulus and Poisson’s-like effects. While further refinement and longer training are required to fully understand its potential, the framework lays the groundwork for future use of RL in lattice design and optimization. The modularity of the framework enables easy updating and refining of the setup, to support use across a wide range of engineering applications.