Towards a Robust Deep Reinforcement Learning for Optimization of Heating Setup in Thermoforming Process

Thermoforming, a commonly used technique in thermoplastics and composites manufacturing, involves interaction of various mechanical components influencing the quality of the final product. Among those, a precise selection of heaters setting plays a pivotal role in pre-optimizing the process. While the traditional control theories have been historically employed for process optimization problems, recent advancements in Artificial Intelligence (AI) have encouraged its adoption across diverse manufacturing domains. Nevertheless, the AI application in thermoforming remains rather limited to date. This case study harnesses a Deep Reinforcement Learning (DRL) to enhance the thermoforming's primary operation: optimizing the input heating setting given a target temperature profile, and possibly saving energy consumption. We showcase the implementation of two widely used DRL algorithms: Proximal Policy Optimization (PPO) and Deep Q-Networks (DQN). A comparison is made to evaluate their effectiveness under diverse process conditions, including both low and high temperature profiles along with single-versus multi-heater utilizations. The PPO algorithm demonstrated a higher performance across all simulated conditions.