ABSTRACT
This paper introduces a novel machine-learning approach based on non-invasive ground-based radar (GBR) time series data for classifying short-span bridge vehicle crossing events. GBR is used to remotely measure the bridge displacement, which is otherwise difficult to acquire but is an essential quantity for Structural Health Monitoring (SHM). For a comprehensive Bridge SHM, monitoring traffic is beneficial to gain knowledge about the actual loading situation. This can be challenging with global responses like displacement. This study indicates that it is possible to classify crossings as single- or multi-presence from displacement signals using tree-based learners and MiniRocket. Thus, our approach serves as a proof of concept to establish remote and data-driven displacement approaches in the context of BWIM. We rely on recordings of the bridge deck taken by an unmanned aerial vehicle as reference data. Despite a small, imbalanced, and biased dataset, we achieve a balanced accuracy of 90%.
