Value of information in sequential decision making: Component inspection, permanent monitoring and system-level scheduling. (October 2016)

• Record Type: Journal Article
• Authors: Memarzadeh, Milad; Pozzi, Matteo
• Is Part Of: Reliability engineering & system safety. Volume 154(2016:Oct.)
• Page Start: 137
• Page End: 151
• Publication Date: 2016-10
• Subjects: Inspection scheduling -- Pre-posterior analysis -- System maintenance; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2016.05.014
• Languages: English
• ISSNs: 0951-8320

Abstract

Abstract: We illustrate how to assess the Value of Information (VoI) in sequential decision making problems modeled by Partially Observable Markov Decision Processes (POMDPs). POMDPs provide a general framework for modeling the management of infrastructure components, including operation and maintenance, when only partial or noisy observations are available; VoI is a key concept for selecting explorative actions, with application to component inspection and monitoring. Furthermore, component-level VoI can serve as an effective heuristic for assigning priorities to system-level inspection scheduling. We introduce two alternative models for the availability of information, and derive the VoI in each of those settings: the Stochastic Allocation (SA) model assumes that observations are collected with a given probability, while the Fee-based Allocation model (FA) assumes that they are available at a given cost. After presenting these models at component-level, we investigate how they perform for system-level inspection scheduling. Highlights: On the Value of Information in POMDPs, for optimal exploration of systems. A method for assessing the Value of Information of permanent monitoring. A method for allocating inspections in systems made up by parallel POMDPs.