Reinforced Scan: A Reinforcement Learning Enabled Optimal Laser Scan Path Planning in PBF Additive Manufacturing

Additive Manufacturing (AM) is an innovative technology that fabricates parts layer by layer. However, in powder bed fusion (PBF), printed metal parts often exhibit residual stresses, deformations, and other defects due to uneven temperature distribution during printing. To address this, an optimized scanning sequence within each layer can help mitigate temperature inconsistencies. Traditional optimization methods are based on domain knowledge, employing try-and-error or heuristic methods. Nonetheless, these methods are not universal and cannot achieve the optimal. The challenge of improving the scanning strategy is the large search space for optimizing the scanning sequence for the scanning tracks within the layer. To fill this gap, this work proposes an innovative scan strategy aiming to optimize scanning sequences for achieving uniform temperature distribution in PBF. The proposed Reinforced Scan approach uses reinforcement learning methods to determine the scanning sequence with a customized reward function intelligently. This reward function not only considers the temperature variance but also the spatial uniformity of temperature distribution. This method can significantly reduce the computational burden involved in scanning sequence optimization. The effectiveness of the proposed Reinforced Scan is validated through Netfabb Local Simulation involving laser scanning on a Ti64 thin plate, where its performance is compared with existing heuristic scan sequences. The simulation results demonstrate that Reinforced Scan yields superior outcomes, achieving reduced residual stress compared to conventional heuristic methods.