Bridge damage classification using multiple responses of vehicles and 1-D convolutional neural networks

Bridges exposed to extreme environmental conditions are susceptible to damage and even failure during service life. Traditional monitoring techniques may necessitate the installation of numerous sensors on the bridge, which can be time-consuming and costly. Instead, the indirect method typically employs several accelerometers attached to the passing vehicle, which is more economical and more accessible to operate. To promote the development of the indirect method, this paper proposes a novel vehicle vibration-based method for classifying bridge damage of varying severity using cutting-edge deep learning techniques. Initially, the framework for damage classification based on the responses of a single vehicle and 1-dimensional convolutional neural networks (1-D CNNs) is appropriately designed and introduced. Then, the proposed approach is evaluated using a steel continuous beam and a model truck in the laboratory, which is utilized to simulate a vehicle-bridge interaction (VBI) system in engineering applications. The experimental results indicate that the bridge’s damage severity can be predicted by the CNN with high accuracy, thereby validating the inclusion of bridge damage information in the passing vehicle’s responses. Furthermore, it is determined that employing multiple responses from the vehicle facilitates the improvement of damage classification accuracy. Heavier vehicles are conducive to the transfer of more bridge-damaged information and are therefore recommended in engineering.