Multi-agent deep reinforcement learning based Predictive Maintenance on parallel machines

In the context of Industry 4.0, companies understand advantages performing Predictive Maintenance (PdM). However, when moving towards PdM, several considerations must be carefully examined. First, they need to have a sufficient number production machines and relative fault data generate maintenance predictions. Second, adopt right approach, which, ideally, should self-adapt machinery, priorities organization, technician skills, but also able deal with uncertainty. Reinforcement learning (RL) is envisioned as key technique in this regard due its inherent ability learn by interacting through trials errors, very few RL-based frameworks been proposed so far literature, or are limited respects. This paper proposes new multi-agent approach that learns policy performed technicians, under uncertainty multiple machine failures. comprises RL agents partially observe state each coordinate decision-making scheduling, resulting dynamic assignment tasks technicians (with different skills) over set machines. Experimental evaluation shows our outperforms traditional policies (incl., corrective preventive ones) terms failure prevention downtime, improving ?75% overall performance.