Reinforced scan: a reinforcement learning enabled optimal laser scan path planning in laser powder bed fusion additive manufacturing

Abstract

Additive Manufacturing is an innovative technology that fabricates parts layer by layer. However, in Laser Powder Bed Fusion (LPBF), printed metal parts often exhibit residual stresses, deformations, and other defects due to non-uniform temperature distribution during the printing process. To mitigate these issues, an optimized scan sequence within each layer can improve thermal uniformity. Traditional optimization methods, which rely on domain knowledge and employ trial-and-error or heuristic approaches, often fail to achieve optimal solutions due to the complex nature of the problem. One major challenge in improving scan strategies lies in the vast search space required to optimize the scan sequence for individual scan tracks within each layer, making it difficult to identify the best solution. To overcome this challenge, this work proposes an innovative scan strategy, Reinforced Scan, that leverages reinforcement learning to intelligently determine the optimal scan sequence. The method introduces a novel reward function that accounts not only for temperature variance but also for the spatial uniformity of the temperature field. By structuring the optimization problem into multiple hierarchical levels, the approach significantly reduces computational demand and enhances the manageability of the optimization process. The effectiveness of the proposed Reinforced Scan is validated through Netfabb™ Local Simulation and real-world laser scanning experiments on a Ti-6Al-4V thin plate. Its performance is compared against conventional heuristic scan sequences. Both simulation and experimental results demonstrate that Reinforced Scan achieves superior outcomes, notably reducing residual stress compared to traditional methods.