Deep Reinforcement Learning Based Toolpath Generation for Thermal Uniformity in Laser Powder Bed Fusion Process
Mian Qin,Junhao Ding,Shuo Qu,Xu Song,Charlie C.L. Wang,Wei-Hsin Liao
DOI: https://doi.org/10.1016/j.addma.2023.103937
IF: 11
2023-12-23
Additive Manufacturing
Abstract:Laser powder bed fusion (LPBF) is a widely used metal additive manufacturing technology. However, the accumulation of internal residual stress during printing can cause significant distortion and potential failure. Although various scan patterns have been studied to reduce possible accumulated stress, such as zigzag scanning vectors with changing directions or a chessboard-based scan pattern with divided small islands, most conventional scan patterns cannot significantly reduce residual stress. The proposed adaptive toolpath generation (ATG) algorithms, aiming to minimize the thermal gradients, may result in extremely accumulated temperature fields in some cases. To address these issues, we developed a deep reinforcement learning (DRL)-based toolpath generation framework, with the goal of achieving uniformly distributed heat and avoiding extremely thermal accumulation regions during the LPBF process. We first developed an overall pipeline for the DRL-based toolpath generation framework, which includes uniformly sampling, agent moving and environment observation, action selection, moving constraints, rewards calculation, and the training process. To accelerate the training process, we simplified the data-intensive numerical model by considering the turning angles on the toolpath. We designed the action spaces with three options, including the minimum temperature value, the smoothest path, and the second smoothest path. The reward function was designed to minimize energy density to ensure the temperature field remains relatively stable. To verify the effectiveness of the proposed DRL-based toolpath generation framework, we performed numerical simulations of polygon shape printing domains. In addition, four groups of thin plate samples with different scan patterns were compared using the LPBF process. The results indicated that using our proposed DRL algorithms for toolpath generation reduced the maximum distortion of the sample by approximate 47% compared to Zigzag patterns, 29% compared to Chessboard patterns, and 17% compared to ATG patterns. Our work demonstrated the effectiveness of implementing machine learning algorithms in the toolpath generation of the LPBF process.
engineering, manufacturing,materials science, multidisciplinary