This study puts forth a methodology to discern structural damage in bridges that employs two-dimensional convolutional neural network (2D-CNN), which is rooted in the principles of continuous wavelet transform (CWT) theory. The method combines the vehicle–bridge coupled vibration response with deep learning models to extend the application of indirect bridge damage identification methods. To test the proposed method, a spatial vehicle and bridge computational model is established for a three-span continuous beam bridge, and bridge damage is simulated by reducing the stiffness of the unit under different damage conditions. Considering the stochastic nature of road roughness, a self-developed vehicle–bridge coupled vibration analysis program is utilized to acquire the vehicle acceleration response signal and construct the dataset. The 2D-CNN model, with its high sensitivity to 2D data features, is used to extract features from the vehicle vertical acceleration vibration signal. The signal undergoes transformation via CWT, resulting in a 2D grayscale time-frequency image. This image is subsequently utilised as input to construct the 2D-CNN model. Results demonstrate that this method performs well in the identification of bridge structural damage, exhibiting high accuracy in identifying the location and severity of such damage. Thus, a novel avenue is provided for the identification and assessment of bridge structural damage.

Based on the displacement-induced linearity error curve and the theory of wavelet packet analysis, a bridge structural damage identification method is proposed, integrating two damage indicators: Quasi-Static Displacement-Induced Linearity Error Curve (QSDIL) and Relative Energy Rate of Wavelet Packet Energy Spectrum (RES). This method first constructs the QSDIL damage indicator based on a quasi-static displacement-induced linearity error, which is used for the preliminary localization of the bridge structural damage. Subsequently, relying on the principles of wavelet packet analysis, the method constructs the RES damage indicator for accurate positioning of the damage location in the bridge structure. The proposed method is experimentally validated, and the impact of factors such as single-point damage, multi-point damage, signal noise, lane position, and vehicle weight on the experimental results is investigated. The results indicate that the proposed method exhibits excellent identification performance for the location of structural damage in both single-point and multi-point damage scenarios, with good agreement between experimental and theoretical values. As the signal-to-noise ratio decreases, the accuracy and precision of the RES curve in locating the bridge structural damage position exhibit a nonlinear decreasing trend, with relatively small identification errors observed at noise levels of 90 dB to 100 dB. Different lane positions have a minimal impact on the damage identification effectiveness. With increasing vehicle weight, both QSDIL and RES curves show an increasing trend in peak values, facilitating the localization of bridge structural damage positions.

A finite element analysis (FEA) model was developed for Euler and Timoshenko curved beam elements by incorporating trigonometric displacement distribution patterns. Local polar coordinate stiffness matrices were derived based on force-displacement relations and static equilibrium conditions. By employing the kinetic energy theorem and triangular displacement functions, an expression for the consistent mass matrix of a curved beam element was obtained. A coordinate transformation matrix for the curved beam element was established by relating the local polar coordinate system to the global polar coordinate system. Calculation programs were implemented in the Fortran language to evaluate the static–dynamic performance and natural frequency characteristics of curved beam bridges. The obtained results were then compared with those obtained using ANSYS solid models and “replace curve with straight” beam element models. The comparison demonstrated a strong agreement between the results of the Euler and Timoshenko curved beam element models and those of the ANSYS solid models. However, discrepancies were observed when comparing with the results of the “replace curve with straight” beam element model, particularly in terms of lateral static displacement. This discrepancy suggests that the characteristic matrix derived in this study accurately represents the stiffness and mass distribution of the curved beam, making it suitable for mechanical performance analysis of curved beam bridges. It should be noted that the “replace curve with straight” method overlooks the initial curvature and the bending–torsion coupling effects of a curved beam, resulting in calculation deviations. On the other hand, the use of curved beam elements in numerical analysis provides a simple and practical approach, which facilitates further research in areas such as vehicle–bridge coupling vibrations and seismic analysis of curved beam bridges.

To overcome the limitations of existing indirect bridge damage identification methods based on vehicle-induced responses, this study presents an integrated approach combining vehicle-induced responses with a one-dimensional deep neural network (DNN) model for bridge damage detection. A laboratory truss bridge model was constructed, accompanied by an experimental system consisting of the truss bridge, a moving vehicle, and a signal acquisition unit, along with its corresponding finite element model. By varying truss member cross-sectional areas and reducing element stiffness, acceleration time-history responses were obtained under different damage scenarios, with vehicle load acting as the excitation source. A total of 3,600 single-damage and 3,600 multiple-damage condition datasets were generated. Two deep learning architectures, 1D-CNN and CNN-LSTM, were developed and optimized with respect to network structure, learning rate, and convolution kernel size.The performance of each model in terms of damage identification accuracy was assessed. For singledamage scenarios, the CNN-LSTM model outperformed the 1D-CNN model in both accuracy and training efficiency, highlighting the advantage of the LSTM module in capturing temporal features.In the case of multiple-damage conditions, the CNN-LSTM model was applied for multi-label classification and regression tasks. The model successfully identified damage locations and severities, achieving a damage location recognition accuracy of 92.20% and an R² value of 0.9168 on the test set, demonstrating the feasibility and robustness of the proposed approach for practical bridge health monitoring.