[BK세미나] 9/16(화) Prof. Ryozo Nagamune(The University of British Columbia) "Deep Reinforcement Learning for Melt Pool Sol

기계공학부 구성원들의 많은 관심과 참여 부탁드립니다.

▣ 주 제: Deep Reinforcement Learning for Melt Pool Solidification Cooling Rate Control in Directed Energy Deposition

▣ 연 사: Prof. Ryozo Nagamune

▣ 소 속: The University of British Columbia

▣ 일 시: 2025. 9. 16.(화) 15:00

▣ 장 소: 제1공학관 A685호

▣ 초 록

In this talk, a method based on reinforcement learning (RL) will be presented to optimize process parameters in directed energy deposition (DED) metal additive manufacturing (AM) process. The process parameters to be considered are layer-wise constant laser power, laser scan velocity, and inter-layer dwell time. These parameters are trained to make layer-wise average melt pool solidification cooling rate (SCR) trajectory track a specified target trajectory, and at the same time, to minimize the manufacturing time. The training is based on an RL method called deep deterministic policy gradient method. Simulation results using a physics-based finite-difference model demonstrate that SCR tracking errors can be achieved with minimized manufacturing time. The proposed RL-based method facilitates process parameter tuning in DED metal AM processes for efficiently manufacturing products with controlled microstructure and mechanical properties.