ABSTRACT
In this paper, we propose an unsupervised data-driven damage detection strategy using acceleration responses from multiple passing vehicles. The strategy is evaluated numerically including practical operational variabilities, namely, vehicles speed, road roughness and signal noise. In the proposed method, acceleration responses from multiple vehicles crossing the healthy bridge are considered. The deep autoencoder based dimensionality reduction technique is applied to train a model for healthy bridge condition. The trained model is subsequently used for reconstructing measured vehicle responses. A Kullback–Leibler (KL) divergence-based damage index is applied on the reconstruction loss that allows for bridge damage detection and severity quantification. In this paper, the proposed method is validated numerically using a vehicle-bridge interaction model, where the vehicles consist of 5-axle trucks that traverse a finite element (FE) model of a multi-span continuous bridge. The performance of the proposed method is evaluated for different damage cases in the presence of road profile and random traffic on the bridge. The proposed method is also evaluated for the case of varying temperature conditions. The overall results suggest that the proposed method can detect damage with acceptable accuracy even in the presence of temperature variations, presenting a potentially useful approach for road network-wide bridge monitoring.
