BAYESIAN MODEL UPDATING FOR EXISTING SEISMI-ISOLATED BRIDGES USING OBSERVED ACCELERATION RESPONSE DATA

摘要

In ageing existing bridges, the deterioration of the members cannot be ignored and the model parameters should be carefully estimated in time through observation data from the structures. On seismic-isolated bridges, the post-yield stiffness of rubber bearings is a particularly critical parameter, since it affects the structural behavior against huge earthquakes. Additionally, its characteristic can be captured using available seismic response data. In this study, a seismic-isolated bridge pier is modelled as a two degree of freedom lumped mass system and its uncertain model parameters are estimated based on time-histories of the acceleration response through Bayesian model updating. The Bhattacharyya distance is used as a comprehensive uncertainty quantification (UQ) metric to capture multiple uncertainty sources from both model predictions and observation data. A sequential procedure is proposed to prevent the so-called “dimensional curse” due to very high dimensional time-series data in model updating. In this procedure, the Bhattacharyya distance is calculated for each predefined time segment and their root mean square is adopted as the UQ metric. The results demonstrate that the proposed framework can successfully update model parameters including the post-yield stiffness of the rubber bearing through seismic response data. Furthermore, the calibrated model can represent wholly the probability distribution of the target time-series observation.