Semi-supervised risk-based active learning using inspection and maintenance information

Gaining the ability to make informed decisions regarding the operation and maintenance (O&M) of structures and infrastructure provides motivation for development of asset management technologies such as structural health monitoring systems and digital twins. The current paper proposes a methodology for improving risk-based active learning and consequently decision-making performance. This approach is achieved by leveraging the Markovian structure of the decision processes in which classification model are used. Historical predictions for the salient health states are updated by applying latent-state smoothing algorithms. Subsequently, these improved predictions are used to update the statistical classifier in a semi-supervised manner. It is demonstrated that these updates allow the classifier to more rapidly converge to yield well-defined decision boundaries that result in few inspections being made and enhanced decision-making performance.