Condition-based maintenance for production systems: Inferring degradation states from quality characteristics

With the growing complexity of production systems, direct monitoring of system degradation has become increasingly challenging, posing difficulties for decision makers to plan appropriate maintenance actions. This paper presents a systematic framework that utilizes the quality characteristics of products to develop the optimal maintenance strategy for production systems. The degradation process of the system is modeled by a Wiener process. Product quality characteristics and underlying system states are mapped by a random function. The variability in the mapping mechanism is characterized by a normal distribution. Leveraging the partially observable Markov decision process, the optimal dynamic maintenance strategy is determined, and its structural properties are explored via in-depth stochastic analyses. Analytical results indicate that the optimal maintenance strategy is a control limit policy, with a threshold that divides the decision space into two mutually exclusive subspaces. Furthermore, numerical examples motivated by maintenance of vacuum molding machines are elaborated to illustrate the proposed model. Comparisons with common methods demonstrate the effectiveness of the proposed approach, whose performance outperforms that of the benchmarks under both finite and infinite horizons. It is also found that the increased uncertainty in the mapping mechanism between product quality and underlying states leads to a noticeable change in the optimal strategy and a higher maintenance cost.