Deep learning-based detection of hinge joint and girder damage in prefabricated multi-girder bridges using vehicle-bridge interaction responses

Accurate detection of damage in hinge joints and girders is crucial for ensuring the safe operation of prefabricated multi-girder bridges. This paper presents a deep learning-based approach for identifying damage in hinge joints and girders of prefabricated multi-girder bridges, leveraging the dynamic response of the vehicle-bridge interaction (VBI) system. The proposed method integrates a hybrid deep learning model that can simultaneously capture local features and global dependencies within the VBI response time series. This model extracts feature representations that are highly sensitive to hinge joint and girder damage, using responses from either the girder or the passing vehicle, thereby enabling precise classification of multiple damage types. The vehicle speed and bridge surface roughness that affect VBI vibration are considered key operational and environmental variables, and the effectiveness of the damage classification model is validated through numerical examples of various damage scenarios, including hinge joint and girder damage at different locations and severities. Additionally, further robustness evaluations demonstrate that the model consistently maintains high classification accuracy and exhibits strong environmental adaptability across varying VBI response types, vehicle speeds, noise levels, and bridge surface roughness levels.