Deep learning-based damage assessment of hinge joints for multi-girder bridges utilizing vehicle-induced bridge responses

Hinge joint damage is a typical defect in multi-girder bridges, and its effective evaluation is crucial for ensuring bridge safety and sustainable operation. However, directly identifying this type of damage from girder responses is challenging. This study presents a novel framework for hinge joint damage detection in multi-girder bridges, by leveraging a deep learning model to reconstruct the responses of the two adjacent girders on both sides of a hinge joint. Through analysis of reconstruction errors of batch samples, the results demonstrate a significant disparity between the damage and baseline distributions, enabling effective evaluation of hinge joint damage. The proposed methodology is numerically validated through various damage scenarios, including cases of single and multiple damages with different severities. Additionally, performance evaluations involving different vehicle speeds and levels of measurement noise indicate that higher vehicle speeds enable better detection performance, and meanwhile the proposed method demonstrates remarkable noise robustness.