This study aims to achieve rapid and economical damage diagnosis of small- and medium-span bridges under moving loads. To this end, a new method is proposed based on synchrosqueezed wavelet transform (SWT) and multi-label convolutional neural network (CNN).
Single-label CNN (SL-CNN) and multi-label CNN (ML-CNN) models were developed for damage identification. To acquire plenty of damage samples, vibration responses of the bridge when the vehicle passes through were simulated by element finite analysis under various damage cases. Then, the time domain response signals were transformed to time-frequency wavelet scalograms by SWT, which were fed into the constructed CNN model for training and testing its ability to identify structural damages.
The efficiency and accuracy of the proposed method were validated through numerical and experimental studies. Results showed that the training of ML-CNN was more efficient than the SL-CNN model, and that the ML-CNN model behaved with superior identification accuracy and generalization performance.
The proposed SWT and ML-CNN based method is more suitable for damage identification under complex multi-damage cases. It requires only a few sensors to adapt to diverse structural scenarios and diagnostic needs, and is thus highly practical for engineering applications.
