Drive-by Bridge Health Monitoring Using Multiple Passes and Machine Learning

This paper studies a machine learning algorithm for bridge damage detection using the responses measured on a passing vehicle. A finite element (FE) model of vehicle bridge interaction (VBI) is employed for simulating the vehicle responses. Several vehicle passes are simulated over a healthy bridge using random vehicle speeds. An artificial neural network (ANN) is trained using the frequency spectrum of the responses measured on multiple vehicle passes over a healthy bridge where the vehicle speed is available. The ANN can predict the frequency spectrum of any passes using the vehicle speed. The prediction error is then calculated using the differences between the predicated and measured spectrums for each passage. Finally, a damage indicator is defined using the changes in the distribution of the prediction errors versus vehicle speeds. It is shown that the distribution of the prediction errors is low when the bridge condition is healthy. However, in presence of a damage on the bridge, a recognisable change in the distribution will be observed. Several data sets are generated using the healthy and damaged bridges to evaluate the performance of the algorithm in presence of road roughness profile and measurement noise. In addition, the impacts of the training set size and frequency range to the performance of the algorithm are investigated.