Bayesian Model-Updating Implementation in a Five-Story Building

In the realm of structural health monitoring (SHM) and structural engineering, numerical modeling and updating techniques strive to minimize disparities between experimental data and model predictions. Traditional methods seek to estimate an individual set of parameters, but they face challenges in capturing variations arising from the uncertainties inherent in structural models. In this context, Bayesian Inference (BI) has emerged. This study leverages BI to update parameters in a five-story building model and assess uncertainties. Modal frequencies and mode shapes are obtained from testing conducted on a full-scale reinforced concrete building affixed to the NEES-UCSD shake table at the University of California, San Diego, USA. This implementation incorporates an iterative updating approach for the covariance matrix, affording a deeper comprehension of correlations among modal parameters. A noteworthy advantage of this approach is its capacity to account for uncertainties inherent in experimental data and the correlation of each set of parameters and outputs, thereby offering a more robust representation of the building’s behavior. This approach is particularly valuable in cases with limited sample sizes, where experimental data variability is inherently low, facilitating precise posterior distribution estimation. Furthermore, this study explores the implications of utilizing a multivariate normal likelihood function as a likelihood function, potentially leading to better posterior distribution estimations by accommodating the unique characteristics of each parameter. The outcomes underscore the efficacy of the Bayesian model updating approach in achieving a statistically robust parameter updating, facilitating a thorough characterization of uncertainty, and elevating confidence levels in the model’s predictions.