ABSTRACT
CFM-5 bridge in Barkston Ash, UK presents an asset monitoring and maintenance challenge. Since its construction in 1868, the masonry arch bridge has suffered damage and undergone a variety of repairs. The uncertainty of the repairs’ effectiveness motivates the need for automated monitoring. As such, Fiber Bragg Grating fiber optic sensors have been installed along the underside of the masonry arch and dynamic strain data have been collected during train passages over the last 2 years. Each FBG measures average strain over its gauge length, and 2 lines of 10 grating rosettes were installed along the arch soffit. These provide distributed strain data over the extent of the arch, yielding more global information compared to typical point sensors. In this context, this paper presents a framework for post-processing and interpreting the monitoring data from bridge CFM-5. The objectives are to investigate long-term and future structural performance. To achieve this, data must first be normalized against train type; using features extracted from the dynamic response to a train passing overhead, the train type is classified via XGBoost, a framework for gradient-boosted decision trees. Then, the dynamic responses are analyzed to evaluate the sensitivity to different train types by observing peak dynamic strains and their correlation with temperature.
