AI-Driven optimization of the filling process: a comparison of reinforcement learning methods

Abstract

This study proposes an artificial intelligence-based approach to overcome the limitations of traditional PID controllers in the control of industrial filling processes. The precision requirements of the filling process can be affected by variables such as material type, temperature, and flow rate, which may cause classical control methods to be inadequate. Therefore, a dynamic and data-driven control model has been developed using reinforcement learning (RL) methods. Monte Carlo (MC), Temporal Difference (TD), and Q-Learning methods have been compared, and experimental analyses have been conducted to determine the most effective strategy. Simulation results have demonstrated that the MC method is more suitable for process modeling, and the optimal transition points between coarse and fine feeding in the filling process have been identified.