Finite element model updating of existing bridge structures based on measured data and deep learning method

Due to the influence of design parameter errors, construction inaccuracies, and time-dependent material degradation, such as shrinkage and creep, it is challenging to directly predict the bearing capacity of existing concrete girder bridge structures using numerical simulations. To address this issue, a finite element model (FEM) updating method based on a proxy model (PM) is proposed and applied to a real-life plate girder bridge, aiming to enhance the accuracy of bearing capacity evaluation compared to traditional estimation methods. Firstly, considering the critical mechanical parameters that significantly influence structural performance, such as deflection and fundamental frequency, a bridge structure in service was simulated using the FEM and subjected to random static vehicle loads. This generated a large dataset of deflection and fundamental frequency samples. Based on these samples, a high-precision proxy model (PM) was constructed using a deep neural network. Additionally, through sensitivity analysis, key mechanical parameters of the bridge structure—such as boundary restraint stiffness, compressive strength of concrete, and elastic modulus of reinforcement—were determined. The finite element model was then updated using these parameters. The reliability and accuracy of the proposed method were validated against in-situ measured data. The difference between the measured deflection and the predicted deflection of the updated model was only 0.4 %. Among three proxy models, the deep learning network (DLN) showed the best prediction performance with sufficient samples, while XGBoost proved more effective in cases with missing samples. Furthermore, the updated FEM was used to predict the bearing capacity of the original bridge. The results demonstrated that the proposed proxy model (PM) method provided high accuracy in assessing the bearing capacity, offering a reliable approach for the safety evaluation of existing bridge structures and the residual bearing capacity assessment of damaged bridges, even with limited measured data.