Abstract:To investigate the seismic vulnerability of in-service bridges, considering the uncertainty of system parameters in finite element modeling, this study presents a bridge model modified by a back-propagation (BP) neural network. A variable cross-section continuous girder bridge in East China was taken as an example. Utilizing Midas Civil software, a refined finite element model was established, where the bridge's dynamic responses served as inputs and structural parameters as outputs. The original finite element model was modified using the measured modal data of the bridge. The modification results showed that the BP neural network reduced model error from 22.92% to 4.58%, enhancing computational accuracy. Following the seismic isolation design theory in China's "Code for Seismic Design of Highway Bridges", lead rubber bearings were installed. The incremental dynamic analysis (IDA) method was employed to conduct nonlinear time-history analyses on three models: original model, modified model, and isolation-optimized model. Structural responses under different seismic waves were extracted to develop vulnerability curves. The data results indicated that the modified model had slightly lower damage probability than the original one. The use of seismic isolation bearings effectively reduced the probability of structural failure under seismic load, with the maximum damage probability decreasing by approximately 42%, demonstrating significant isolation effectiveness.
