Preventive maintenance optimization of production system based on multi-agents deep reinforcement learning

Intelligent manufacturing system, which is advanced and productive, is vulnerable to internal degradation and external impacts in industrial environment. The preventive maintenance activities are usually conducted to reduce random failure and environmental impacts by repairing and replacing aged production machines. However, the dynamic characteristics of production system impede the operators to obtain the optimal maintenance policy. To solve this problem, the deterioration process of series production line is modeled as Markov decision process. Multi-agent deep reinforcement learning algorithm is presented to achieve the optimality, including maximizing the production rate while minimizing the economic costs. To alleviate the curse of dimension and curse of history, deep neural networks with self-learning mechanism are adopted to approach the optimal policy function. A simulation study is conducted to validate the effectiveness of the proposed model.