ABSTRACT
Due to the increasing age of and load on traffic bridges, there is a high demand for systems to monitor the structure of traffic bridges and detect possible damage. This paper presents a data-driven approach to detect damage in bridge structures. Strain data was measured while the bridge was excited by single-vehicle crossings. The proposed feature generation methods focus on ratio-based features that are weight- and velocity-independent of the generally unknown vehicle parameters. The calculated features were used for a supervised classification approach, where damage location and severity was known. Also an unsupervised anomaly detection was considered, where only the deviation to a reference state is analysed. Evaluation was based on the data of a two-span beam bridge, where numerical data and experimental data were considered for a step-by-step damage of the bridge. The results show that the ratio-based features provide a high degree of accuracy for classifying the bridge damage and also allow the calculation of a distance to the reference state.
