A framework for in-situ process control in metal additive manufacturing using anomaly-driven reinforcement learning

Metal additive manufacturing (AM) enables the production of complex geometries, customized parts, and high-performance components. However, it faces challenges ensuring consistent part quality due to process stochasticity, necessitating in-situ process control. Current process control approaches often lack integration of multiple process parameters, are typically limited to one static control objective, and rely on assumptions about process dynamics. This paper introduces a framework for in-situ process control in AM utilizing anomaly-driven reinforcement learning (AD-RL) to address these issues. Reinforcement learning enables training a comprehensive control policy that can simultaneously adjust multiple process parameters. However, reward functions based on manual rules and process assumptions constrain the policy’s scope and its capacity to dynamically adjust to process variations. Therefore, this framework utilizes anomaly detection to automate the reward function. The framework contributes to improving process control in AM, with the benefits of AD-RL suggesting broader implications for advancing control in manufacturing systems.