Foundations and applicability of transfer learning for structural health monitoring of bridges

Abstract

The number of bridges worldwide is extensive, making it financially and technically challenging for the authorities to install a structural health monitoring (SHM) system and collect large quantities of data for every bridge. Transfer learning has gained relevance in the last few years to extend the SHM concept for most bridges, while minimizing costs with monitoring systems and time with data measurement. It can be especially suitable for bridges structurally similar and replicated extensively, like overpasses integrated into highways. Therefore, this paper intends to lay down the foundations of transfer learning for SHM of bridges and to highlight the importance of the quality of knowledge transferred across different bridges for damage detection. Transfer Component Analysis, Joint Distribution Adaptation, and Maximum Independence Domain Adaptation methods are applied to data sets from different bridges, where classifiers have access to labeled training data from one bridge (source domain) and unlabeled monitoring test data from another bridge (target domain) that present similarities. The effectiveness of those methods is compared through the classification performance using real-world monitoring data sets collected from the Z-24 Bridge in Switzerland, and the PI-57 and PK 075+317 Bridges in France.