Bayesian model updating of a 250 m super-tall building utilizing an enhanced Markov chain Monte Carlo simulation algorithm

Abstract

Finite element models (FEMs) are effective for predicting structural behaviors subjected to various excitations. However, modeling errors in FEMs always exist, especially for complex structural systems, and these affect the accuracy of FEMs. Model updating using measured data can significantly improve model accuracy as it can closely match the predicted and measured responses. In this work, field vibration tests were carried out on a 250-meter super-tall building, and model updating was conducted using an enhanced Markov chain Monte Carlo (MCMC), developed based on Bayesian theory. The sampling process was divided into multiple levels, with each level having a sampling level, generating the target PDF, which is regarded as the bridge PDF. Kernel density estimation is used to adaptively construct the proposal PDF in each level so that the generated samples can move to the region of high probability smoothly level by level. Sensitivity analysis was carried out to investigate the efficiency of the proposed model updating algorithm. Different algorithmic parameters, including the number of uncertain parameters, initial samples of each parameter, the ratio of error variances between two levels, and the number of sampling levels, are discussed to study the performance of the algorithm on the application of the super-tall building.

Keywords

• Model updating
• Markov chain Monte Carlo
• Multi-level sampling
• Bayesian inference
• Super-tall building