ABSTRACT
In recent years, AI and image-based inspection technologies have gained attention for infrastructure maintenance, offering cost-effective solutions compared to traditional methods. Japan’s Ministry of Land, Infrastructure, Transport and Tourism is promoting their adoption. However, the accuracy of these new technologies often lags behind conventional methods, posing risks of maintenance mismanagement due to unreliable inspection results. To address this, this study proposes a methodology using a partially observable Markov decision process (POMDP) to optimize inspection and repair policies while considering uncertainties in inspection accuracy. A case study will evaluate the methodology’s application to real-world infrastructure management, determining the minimum accuracy and cost thresholds necessary for practical use. This approach aims to balance cost-efficiency and reliability, enabling informed decisions on the adoption of new inspection technologies.
