Reinforcement learning inclusion to alter design sequence of finite element modeling

Abstract

The study explores possibilities on how to approach cross-field methods, such as the design of mechanical systems via finite element modeling, with the contribution of reinforcement learning as a machine learning technique for guidance in design space. The application of the epsilon-greedy algorithm for optimizing parametric finite element model is illustrated by simulations through practical examples, namely the design of a cantilever beam and a JetVest. The results obtained clearly show that this approach can be beneficial in the field of rapid prototyping.