Bayesian Inference Based Model Calibration for the Dynamic Analysis of Seismic Isolated Bridges

Abstract

The model validation has a key role to obtain appropriate numerical analysis outputs that are used for the seismic strengthening or retrofit design of existing bridges in Japan. In this study, we verified the applicability of the Bayesian inference based model calibration for the numerical modeling of existing bridges especially focusing on the validity of boundary conditions, which were given by the bridge bearings. The target structure was an existing seismic isolated bridge with laminated rubber bearings; this bearing sometimes shows uncertain changes in the elastic property caused by the residual deformations due to previous earthquakes. The prior distributions of calibration parameters, i.e., material properties and spring stiffness for the rubber bearings in the FE model of the target bridge, were then updated to the Bayesian posterior distributions using the resonant frequencies acquired by the impact tests. The calibration model was obtained by assigning the mean values of the posterior distributions. The resonant frequencies of calibrated model then showed successful improvement in the agreement with the experimental result. Moreover, the possibility of identifying the uncertain changes of bearing parameters was also investigated by the experimental verification using a beam structure. The uncertainties of the rubber bearing parameters of the numerical model were successfully reduced, and furthermore, the change of parameter due to the residual deformation could be identified from obtained Bayesian posterior distributions.