A Rapid Reliability Evaluation Method for Beam Bridges Based on a PCA-GPR Surrogate Model

To address the susceptibility of deflection signals to noise and the difficulty of influence-line identification in finite element (FE) model updating of bridges, while overcoming the low efficiency associated with high-dimensional response modeling, this study proposes an FE model updating method for steel box girders using the error in the deflection influence line as the objective function. First, an attention-enhanced variational mode decomposition method is employed to denoise and reconstruct measured deflection time histories, and Tikhonov regularization is combined to extract the bridge deflection influence line. Next, influence-line samples are generated under different parameter combinations, and a PCA-GPR surrogate model is used to establish the mapping between updating parameters and influence-line features. Finally, a genetic algorithm is introduced to optimize the parameters to be updated. The updating results show that the overall relative error of the deflection influence line decreases from 56.02% to 17.3%, while the relative error of the peak value at the control section decreases from 49.44% to 0.6%, indicating that the proposed method effectively improves model updating accuracy and significantly reduces the discrepancy between the updated model and the measured response. On this basis, the updated model is used to perform a rapid reliability evaluation over a range of vehicle speeds and load levels. The results reveal systematic variations in failure probability and reliability index across the investigated operating domain, providing a useful basis for rapid assessment of bridge serviceability under varying traffic conditions.Abstract: To address the susceptibility of deflection signals to noise and the difficulty of influence-line identification in finite element (FE) model updating of bridges, while overcoming the low efficiency associated with high-dimensional response modeling, this study proposes an FE model updating method for steel box girders using the error in the deflection influence line as the objective function. First, an attention-enhanced variational mode decomposition method is employed to denoise and reconstruct measured deflection time histories, and Tikhonov regularization is combined to extract the bridge deflection influence line. Next, influence-line samples are generated under different parameter combinations, and a PCA-GPR surrogate model is used to establish the mapping between updating parameters and influence-line features. Finally, a genetic algorithm is introduced to optimize the parameters to be updated. The updating results show that the overall relative error of the deflection influence line decreases from 56.02% to 17.3%, while the relative error of the peak value at the control section decreases from 49.44% to 0.6%, indicating that the proposed method effectively improves model updating accuracy and significantly reduces the discrepancy between the updated model and the measured response. On this basis, the updated model is used to perform a rapid reliability evaluation over a range of vehicle speeds and load levels. The results reveal systematic variations in failure probability and reliability index across the investigated operating domain, providing a useful basis for rapid assessment of bridge serviceability under varying traffic conditions.