Optimization of an aperiodic sequential inspection and condition-based maintenance policy driven by value of information

The issue of the optimal planning of inspection and maintenance actions for a randomly deteriorating system constitutes a difficult sequential decision-making problem in which the objective is generally to achieve minimal life-cycle cost. For mathematical tractability, most approaches rely either on the consideration of specific maintenance strategies, e.g. Periodic Inspection and Replacement (PIR), whose defining parameters are optimized, or on time-and-space-state discretization using Markov Decision Process (MDP) models and resolution through policy iteration. In both cases, optimality may be hard to guarantee. In this paper, the decision-theoretic concept of Value of Information (VoI) is used as a metric to guide resource prioritization in time, that is, to schedule inspections in a piecewise optimal manner. An aperiodic sequential inspection policy is proposed, where the determination of the next best time for inspection, or replacement, is based on the current condition and on the computed expected gain from possible inspections, i.e. on a VoI metric. This policy can be implemented when the current condition is known from imperfect inspection or processing of condition-monitoring data. Also, more generally, a discussion is proposed on the use of VoI as a guide for information collection in life-cycle management.