Deep reinforcement learning for condition-based maintenance planning of multi-component systems under dependent competing risks

Condition-Based Maintenance (CBM) planning for multi-component systems has been receiving increasing attention in recent years. Most existing research on CBM assumes that preventive maintenances should be conducted when the degradations of system components reach specific threshold levels upon inspection. However, the search of optimal maintenance threshold levels is often efficient for low-dimensional CBM but becomes challenging if the number of components gets large, especially when those components are subject to complex dependencies. To overcome the challenge, in this paper we propose a novel and flexible CBM model based on a customized deep reinforcement learning for multi-component systems with dependent competing risks. Both stochastic and economic dependencies among the components are considered. Specifically, different from the threshold-based decision making paradigm used in traditional CBM, the proposed model directly maps the multi-component degradation measurements at each inspection epoch to the maintenance decision space with a cost minimization objective, and the leverage of deep reinforcement learning enables high computational efficiencies and thus makes the proposed model suitable for both low and high dimensional CBM. Various numerical studies are conducted for model validations.