Sequential inspection policy optimization for hierarchical multistate systems

Inspections, aiming at identifying the system’s health status by sensing techniques or experts’ observations, serve a crucial role in health management of engineered systems. In real-world applications, systems often exhibit multistate and hierarchical characteristics, allowing inspections to be implemented across multiple physical levels. Nevertheless, due to the limited inspection resources and the uncertainty associated with the system’s health status, it necessitates planning the multilevel inspection policy sequentially. In this article, a novel sequential inspection policy optimization framework is proposed for hierarchical multistate systems, wherein the optimization problem is captured within a finite-horizon partially observable Markov decision process (POMDP). To effectively solve the resulting POMDP, a tailored value iteration algorithm with condensed spaces is developed. The POMDP’s state space is reduced by merging all possible system states, and the POMDP’s action space is cut down by eliminating ineffective inspection actions. Several propositions are provided to validate the algorithm’s effectiveness. A three-component pipeline system, along with an electro-mechanical actuator system, is given to demonstrate the effectiveness of the proposed framework. The results indicate that the state and action spaces can be significantly reduced while the sequential inspection policy outperforms the traditional static policy in terms of revealing the system’s health status.