Selective Maintenance and Inspection Optimization for Partially Observable Systems

Abstract

AbstractSelective maintenance has been extensively investigated based on the premise that all component states after the last mission are precisely known in advance. However, in industrial scenarios, the inspection activities have to be conducted to identify component states, which might share the same resources with maintenance. Due to the limited maintenance effectiveness and inspection accuracy, both maintenance and inspection may be subject to uncertainty. To optimize the selective maintenance and inspection optimization for partially observable systems, a finite-horizon mixed observability Markov decision process (MOMDP) model was introduced when component states were partially observable while the remaining time resources were fully observable. In the MOMDP model, multiple optional maintenance and inspection actions can be dynamically executed during a break based on the state distribution of all components and the remaining time resources. Two illustrative examples were given to demonstrate the effectiveness of the proposed method.KeywordsSelective maintenanceMulti-state systemsInspection strategyPartially observable systems