Bayesian model updating of a full-scale finite element model with sensitivity-based clustering

Abstract

Model updating based on vibration response measurements is a technique for reducing inherent modeling errors in finite element (FE) models that arise from simplifications, idealized connections, and uncertainties with regard to material properties. Updated FE models, which have relatively fewer discrepancies with their real structural counterparts, provide more in-depth predictions of the dynamic behaviors of those structures for future analysis. In this study, we develop a full-scale FE model of a major long-span bridge and update the model to improve an agreement between the identified modal properties of the real measured data and those from the FE model using a Bayesian model updating scheme. Sensitivity-based cluster analysis is performed to determine robust and efficient updating parameters, which include physical parameters having similar effects on targeted natural frequencies. The hybrid Monte Carlo method, one of the Markov chain Monte Carlo sampling methods, is used to obtain the posterior probability distributions of the updating parameters. Finally, the uncertainties of the updated parameters and the variability of the FE model's modal properties are evaluated.