Optimizing Laser Powder Bed Fusion (L-PBF) for Improved Performance: A Reinforcement Learning Approach for Choosing Process Parameters

Abstract

Laser Powder Bed Fusion (L-PBF) is a high-performance manufacturing process for parts with complex geometries. However, a poor choice of process parameters induces the formation of defects, suggesting the need for an improved control system to optimize the result. This paper presents a control policy using Reinforcement learning (RL) for optimizing key process parameters such as laser power (P) and scanning speed (V), aiming to maintain melt pool depth. In this approach, Q learning algorithm iteratively learns how to choose optimal process parameters by maximizing a reward function designed to minimize deformation. A melt pool is simulated using Flow 3D for SS316L, while part-scale deformations are investigated with Finite Element method (FEM). Response surface methodology (RSM) is used to statistically analyse as well as validate deformation behaviour. The proposed approach is the first step towards a data-driven, intelligent control system aimed at optimizing process parameters for stable melt pool depth and minimizing defect formation in L-PBF.