A self-adaptive group maintenance policy for multi-state systems under postponed replacement: A state-action interaction approach

Condition-based group maintenance powered by sensor/inspection information is a foundation for guaranteeing operational safety of complex deteriorating systems. Existing group maintenance approaches, however, rest predominately on fixed group partitions within pre-specified time horizons, rarely accounting for dynamic cooperation of scheduled and unscheduled group maintenance for cost-efficient downtime mitigation. This paper addresses this gap by developing a self-adaptive group maintenance framework under an interactive state-action updating paradigm. Postponed maintenance is tentatively scheduled for individual components, following which both group maintenance partitions and system state space are updated in a global manner. Specifically, distinct postponed maintenance windows are arranged for each component based on its current health state. Building upon individual scheduling outcome, both preventive and opportunistic group maintenance are optimized through unified feedback from historical maintenance actions and the latest component states. This constitutes a state-action interaction mechanism whereby the joint states of all components determine group maintenance sequences adaptively, and in turn, group maintenance actions reshape components' health states instantly. A dynamic programming-based optimization algorithm is developed to identify the optimal courses of maintenance. A case study on a train bogie system demonstrates the applicability of the proposed framework in group maintenance organization and cost reduction.