|About this Abstract
||MS&T23: Materials Science & Technology
||Additive Manufacturing: Equipment, Instrumentation and In-Situ Process Monitoring
||Reinforcement Learning for In-situ Melt Pool Control during Laser Powder Bed Fusion
||Anant Raj, Latif Adurzada, Benjamin Stegman, Charlie Owen, Hany Abdel-Khalik, Xinghang Zhang, John W. Sutherland
|On-Site Speaker (Planned)
Low part quality repeatability during laser powder bed fusion (LPBF) restricts its wider adoption for manufacturing. This is a consequence of process fluctuations, like the variation in the shield gas flow across the build plate, leading to variations in the attenuation of the laser by the plume, ultimately resulting in location-specific property variations. Our previous work demonstrates that signatures of this variation can be captured using co-axial melt pool monitoring. In this work, we generate synthetic co-axial melt pool signals using Flow3D simulations. Using reinforcement learning (RL), we control the printing process to counter random perturbations in laser-plume interactions. We train deep Q-network and proximal policy optimization based RL-agents using melt pool signals as input and demonstrate that the agents can control the printing parameters to obtain targeted melt pool profiles in the presence of process fluctuations. The approach is expected to enhance repeatability in LPBF.