Degradation process of many infrastructure systems, such as road networks, water and wastewater networks can be modeled in a probabilistic framework, incorporating the effect of the maintenance policy based on the information collected. Information helps the decision maker to reduce uncertainty about the components’ condition and minimize the corresponding management cost. However information is expensive and resources are limited, so the approach to collect information has to be carefully selected: specifically, inspection scheduling for time-variant components is a key problem in maintenance optimization. While it is understood that agents should assess the value of available pieces of information and scheduling information collection based on that, this pre-posterior analysis is computationally intractable for most applications. Leveraging our recent work, in this paper we propose a framework for inspection scheduling in the context of Partially Observable Markov Decision Processes (POMDPs). Specifically, we propose a heuristic to approximate the value of information for inspecting components modeled by independent POMDPs. The proposed approach assumes that future inspections are allocated by a simple stochastic process, independently for each component and time. The computational complexity of the proposed approach is linear with respect to the number of components. We present the framework and evaluate its performance with a numerical and an analytical example.
doi: 10.12783/SHM2015/307