An improved Bayesian model updating framework by enhanced iterative model reduction technique in time domain

Abstract

An improved Bayesian model updating technique in the Transitional Markov Chain Monte Carlo framework in time domain is investigated. In doing so, a modified iterative model reduction technique is embedded within the improved Bayesian framework to address the limited availability of measurements. The transformation relation proposed includes the inertial effect in each term of the dynamic condensation equation making it less dependant on the selection of master degrees of freedom. This expects to predict responses efficiently from the reduced-order model for stiffness reductions involved in structural health monitoring problems. The proposed algorithm modifies the transition levels of sampling by performing weighted sampling at all levels, a modified choice of plausibility value and adaptively estimating the sample and covariance matrix for the proposal distribution with a tuning algorithm. Thereby, the proposed approach reduces the dependency on the determination of the statistical estimators in obtaining the transitional samples. The efficiency of the algorithm is demonstrated considering simulated data of a multi-storied frame and an available experimental dataset of a multiple spring-mass system. A comparative study of the accuracy and computational efficacy of the proposed approach is made with the existing approaches.