Reinforcement learning for optimal policy learning in condition-based maintenance

Condition-based maintenance (CBM) involves taking decisions on maintenance or repair based on the actual deterioration conditions of the components. The long-run average cost is minimised by choosing the right maintenance action at the right time. In this study, the CBM decision-making problem is modelled as a continuous semi-Markov decision process (CSMDP). It consists of a chain of states representing various stages of deterioration, a set of maintenance actions, their costs and scheduled inspection policy. The application of a reinforcement learning (RL) algorithm based on the average reward for CSMDPs in CBM is described. The RL algorithm is used to learn the optimal maintenance decisions and inspection schedule based on the current health state of the component.