Abstract: In response to the shortcomings of existing methods for indirectly identifying bridge damage based on vehicle response,this study addresses the issue by integrating the coupling vibration response of the vehicle-bridge system with various types of deep learning models,employing the theory of highway bridge-vehicle coupling vibration.The application of indirect bridge damage identification methods is expanded using a three-span continuous beam bridge as an example.Carrying out vehicle and bridge analysis models,bridge damage is simulated by reducing the unit stiffness,and different damage scenarios are established.Considering the randomness of road surface roughness,vehicle-bridge coupling vibration analysis is conducted under various road surface roughness conditions to obtain the vertical acceleration vibration signals of the vehicle.Applying an end-to-end damage identification method with vehicle acceleration response as the network input,two models,namely,one-dimensional convolutional neural network (1D-CNN) and convolutional long short-term memory neural network (CNN-LSTM),are constructed.A comparative analysis of the identification performance of both models is carried out.The results indicate that both networks achieve satisfactory identification results,with CNN-LSTM having advantages in terms of identification accuracy and computational efficiency over 1D-CNN.