关键词: 部分可观察马尔可夫决策过程模型, 检测维修策略, 视情维修, Perseus算法

Abstract: The principles of Partially Observable Markov Decision Processes (POMDP) and the basic procedure of Perseus approximate algorithm were introduced. Machine state assessment method was given based on the processing defective rate of machine and the vibration signal diagnostic information of some of the core components of machine. An inspection and maintenance decision POMDP model of machine was constructed based on the condition-based maintenance taking into account the limitation of inspection method and the uncertainty of inspection result. This model was served to minimize expected discounted cost by an example of a machining center of car engine production line. Furthermore, the Perseus approximate algorithm was used in order to solve this model. Then, an approximate optimal policy of inspection and maintenance and approximate optimal discounted cost value was obtained in a finite-horizon.

Key words: partially observable Markov decision processes model, inspection and maintenance policy, condition-based maintenance, Perseus algorithm