Dynamic maintenance model for a repairable multi-component system using deep reinforcement learning

Using artificial intelligence for maintenance planning is useful for many industries to have a smart decision-making tool that delivers the best maintenance policy to minimize the expected maintenance costs. In this paper, a deep reinforcement learning method is used to provide a new dynamic maintenance model for a degrading repairable system subject to degradation and random shock. At any time, the degradation level of the system can be considered as the state of the system, and based on the available actions, it transits to different levels. The gamma process is used to formulate the degradation form of the system. The maintenance problem is formulated as a Markov decision process, and Deep Q learning algorithm is used to solve the problem. For most of the models in the literature, the degradation state of the system must be discretized. However, discretization of the degradation states brings inaccuracy and inefficiency to the model. In this paper, instead of discretizing the degradation state, we consider the exact level of degradation as the state of the system. The Deep Q learning method tries to recognize patterns instead of mapping every state to its best action. A neural network is trained during the learning process of the algorithm, and it can be used as a decision-making tool for the maintenance team to find the best maintenance action based on the current degradation level of the system. A numerical example illustrates how the deep reinforcement learning algorithm can be applied to find the optimal maintenance action at each degradation level.