PRODUCTION LINE CONTROL MODELING VIA MDP AND MULTIAGENT DEEP REINFORCEMENT LEARNING

This work presents a methodological proposal based on Multi-Agent Reinforcement Learning (MARL) for the distributed control of machine speeds in industrial production lines. The problem is formulated as a Markov Decision Process (MDP), in which local agents make decisions based on information from adjacent buffers and the machine itself, while respecting the natural delay in the impact of actions throughout the system. The approach aims to replace fixed operational rules with adaptive policies learned in simulated environments, aligning with Industry 4.0 trends. The relevance of the proposal is underscored by recent advancements in the literature, which highlight the potential of MARL to handle scenarios characterized by high complexity and variability.