Detecting railway bridge scour using in-service train signals and machine learning tools

Instrumenting a train crossing over a bridge with a scour problem and acquiring signals presents an alternative approach to traditional scour detection techniques. The study utilizes displacement measurements obtained via a conveniently integrated system on a 6-axle vehicle crossing a railway bridge. Given that the measurements provided correspond to the healthy state of the bridge, scoured state vehicle crossings are synthetically generated with a vehicle-bridge interaction model. The bridge's pier stiffness is determined through a Cross-Entropy optimization algorithm that minimizes the sum of squared differences between measured track irregularities and Finite Element model-calculated displacements combined with rail irregularities. The variation between the healthy and scoured states of the bridge is utilized as a damage indicator. The case study is extended to the network level through Machine Learning (ML) algorithms.