A novel porosity prediction framework based on reinforcement learning for process parameter optimization in additive manufacturing

Abstract

Machine learning (ML) has generated great interest in additive manufacturing (AM) thanks to its ability to predict complex patterns and behaviors through data. Examples include design optimization, process control, and cost minimization. In this paper, we develop a novel framework based on reinforcement learning (RL) for porosity prediction in metal laser-powder bed fusion (L-PBF). The novelty of this approach is twofold: it is the first approach that integrates RL in L-PBF for porosity prediction where the state space consists of permutations of three parameters (laser power, scan speed, and hatch spacing) for optimal parameter combinations; furthermore, through an appropriately formulated reward function, we embed physics-informed principles based on the Eagar-Tsai thermal model for training. The proposed framework has been experimentally validated on L-PBF high-strength A205 Al alloy. The experimental results demonstrated high fidelity with the predicted optimal parameters, despite few outliers, demonstrating the potential of this approach.