Toolpath optimization in laser metal deposition for manufacturing thin-walled superalloy components based on deep reinforcement learning

Abstract

Complex thin-walled components play a significant role in reducing product weight, enhancing structural strength and performance, and are indispensable in industrial manufacturing. Nevertheless, manufacturing these components poses challenges and involves considerable costs. This study focuses on integrating laser metal deposition technology with deep reinforcement learning for complex thin-walled components. The objective is to optimize the mechanical properties of thin-walled components produced by laser metal deposition along the height direction of deposition. Experimental analysis determined process parameters, including laser power, powder feeding rate, scanning speed, and single-pass overlap rate. The optimization of the deposition path was formulated as a deep reinforcement learning environment using the established process parameters. A reward function was devised to optimize the target, and a model was trained to generate a toolpath for the laser metal deposition process in manufacturing thin-walled components. The mechanical properties of the thin-walled samples produced using the optimized toolpath were analyzed. Additionally, the deposition path design was further refined for a hollow turbine blade model. The characterization results of the thin-walled samples produced in this research showed that the tensile properties of the samples in the deposition height direction were effectively improved. This suggests the practicality of using deep reinforcement learning methods to generate deposition trajectories aimed at improving the mechanical properties of the formed parts.