Maintenance decision methodology integrating twin data with reinforcement learning

Condition-based maintenance (CBM) involves making decisions on maintenance or repair based on the actual deterioration conditions of the components. The long-run average cost is minimized by choosing the right maintenance action at the right time. In this study, considering that previous decision-making relied on probabilities and assumptions for state transitions, lacking clear mechanistic guidance and disconnecting from the actual processes of change, this paper proposes a maintenance decision-making method that integrates digital twin technology with reinforcement learning. In this method, the digital twin is introduced into the maintenance decision method to realize the intelligent operation and maintenance of mechanical equipment and parts by simulating the CBM decision-making problem as a continuous semi-Markov decision process (CSMDP). By comparing this method with current multi-objective optimization decision-making approaches, it demonstrates a 60% reduction in maintenance costs while extending the equivalent lifespan, and a 50% increase in lifespan under the condition of maintaining equivalent maintenance costs, thereby illustrating the efficacy and superiority of this method.