Integration of Reinforcement Learning Agent to Reduce Coordinator Limitations in Cyber-Physical Production Systems

The complexity and dynamism of cyber-physical production systems (CPPS) pose significant coordination challenges. Traditional mechanisms are often insufficient to optimize these complex systems, due to rigid decision-making processes and limited human capacity. This paper presents a new approach to enhance the coordination of CPPS, based on reinforcement learning (RL). Faced with the challenges posed by the complexity and dynamics of modern production environments, this approach aims at increasing the performance of CPPS systems by overcoming the traditional limits of the coordinator. We demonstrate the practical implementation of this methodology through a simulated aluminium rim production line consisting of two workstations, using specialized simulation software. The goal is to demonstrate that the proposed approach can significantly improve the performance of CPPS systems by reducing the limitations faced by the coordinator. Finally, we analyse and compare the results of the approach that integrates the RL agent with those of a configuration without RL agent. This analysis allows to evaluate the effectiveness of the proposed solution, offering a new perspective on reducing the coordinator’s limitations in managing CPPS.