Quantification of Value of Information associated with Optimal Observation Actions within Partially Observable Markov Decision Processes

Maintenance is inevitable in the structural life cycle, due to excessive loads, poor construction, corrosion, and mechanical deterioration. Infrastructure maintenance is a sequential decision-making process, which can be modeled by partially observable Markov decision processes (POMDPs). This model constructs a general framework to parameterize the maintenance process, including intervention and observations. The former improves the mechanical performance of the system, while the latter action pertains to the detection of the component symptoms (e.g., inspection and monitoring). Information gathering from observation enhances the decision-making capacity of manager. However, imperfect measurements, such as signal noise, reduce the value of information (VoI). These measurements may induce an investment deficit in the structural management system. Therefore, the primary objective of this study is to quantify the VoI associated with different observation system. Illustrative results show that VoI presents the exponentially decreasing associated with the reducing accuracy. This property is reflected in the empirical equation which uses the logarithmic term to describe the relationship between observation accurate level and VoI. Finally, the numerical results and empirical formula indicate that the low-accuracy structural health monitoring (SHM) system should be avoided as the plausible observation may cause component failure without timely repair.