Optimizing Injection Molding Quality with Reinforcement Learning: A Comparative Study of Algorithms

Injection molding is widely used to produce complex plastic parts, but process disturbances can affect product quality. This study investigates reinforcement learning (RL) for quality control by comparing four RL algorithms: AC, TD3, SAC, and SACa. Algorithms were trained to control part weight and one geometric dimension in a simulated injection molding process using neural networks based on actual process conditions. Finally, SAC achieved the best results in both simulation and subsequent real-world testing. Further research aims to reduce the data- and timeintensive training process by training the RL algorithm directly or in hybrid form on actual injection molding machines.