Reliability modeling, state estimation, maintenance policy, and system design for multi-component systems under dynamic self-configuration

Abstract

Engineering systems usually comprise multiple functionally connected components that jointly provide a desirable performance of the system. In practice, components inevitably deteriorate over time. Owing to the connection in functionalities among components, certain seriously deteriorated components may negatively affect the good-conditioned components and thus significantly reduce the performance of the system. To overcome this problem, some systems are designed to have the ability of dynamic self-configuration. Specifically, these systems can dynamically adjust their configurations by isolating/retrieving certain components from/back to the system in real time, so that the systems always operate at their highest performance. Despite the abundant literature that focuses on multi-component systems with fixed configuration, reliability modeling, state estimation, maintenance policy, and system design for multi-component systems under dynamic self-configuration still remain challenging and underexplored. This thesis establishes a model-based framework that covers the above four reliability-related topics. First, we formally define and analytically model the dynamic self-configuration mechanism, based on which the reliability and reliability metrics of the proposed system are obtained. Then, we develop a marginalized particle filtering algorithm that utilizes both the continuous part and discrete part of the observation data to estimate the hidden system state. In addition, a dynamic reliability prediction method is proposed by integrating the proposed reliability models and filtering algorithm. In practice, maintenance is indispensable to the long-term operation of the system. We propose and optimize a bi-level inspection-and-maintenance policy that utilizes both the system performance obtained by continuous-time monitoring and the components’ conditions obtained by discrete-time inspections to make maintenance decisions for the proposed system. Finally, we propose a new scenario of system design where the dependency among the components can be modified to improve the performance of the system. We jointly optimize the system design and maintenance policy to maximize the system’s long-term profit rate. Comprehensive numerical studies are conducted and numerous managerial insights are obtained.