BWIM aided damage detection in bridges using machine learning

In this study, a new, model-free damage detection method is proposed and validated on a simple numerical experiment. The proposed algorithm used vibration data (deck accelerations) and bridge weigh-inmotion data (load magnitude and position) to train a twostage machine learning setup to classify the data into healthy or damaged. The proposed method is composed in its first stage of an artificial neural network and on the second stage of a gaussian process. The proposed method is applicable to railway bridges, since it takes advantage of the fact that vehicles of known axle configuration cross the bridge regularly, that normally only one train is on the bridge at a time and that the lateral positioning of the loads does not change. The novelty of the proposed algorithm is that it makes use of the data on the load's position, magnitude and speed that can be obtained from a Bridge Weigh-in-Motion system to improve the accuracy of the damage detection algorithm.