Active transfer learning for SHM of bridges under changing environmental conditions

This is the authors' abstract. We don't add key points for this paper.

Obtaining data for training structural health monitoring (SHM) systems is often expensive and/ or impractical, particularly labelled data. Population-based SHM (PBSHM) presents a potential solution to this issue by considering the available data across a population of structures. However, inherent dissimilarities among structures will result in divergent training and testing distributions. Consequently, specialised machine learning methods are required to ensure models can effectively generalise between structures. As such, a transfer learning methodology is proposed to exploit data pertaining to various damage-states and environmental conditions from a source structure, with the aim of enhancing predictions in a target structure with limited data; an active learning framework is used to improve the transfer learning method. The effectiveness of this methodology is evaluated on a novel population of experimental bridges, enabling robust validation of transfer learning for PBSHM in realistic scenarios. Specifically, this population includes data corresponding to several damage-states, as well as, a comprehensive set of environmental conditions; for example, freezing conditions and waterlogging of the deck.