UC San Diego

Civil structures are prone to deterioration due to aging and exposure to man-made and natural hazards, such as explosions, earthquakes, and hurricanes. These events can result in significant consequences, including loss of life and compromised functionality of critical infrastructure like power plants and hospitals. Structural health monitoring (SHM) and damage prognosis (DP) are essential frameworks for guiding decision-making before and after such events, to mitigate their impacts by enabling early damage detection, structural integrity assessment, and residual service life estimation. However, challenges persist in implementing SHM and DP for civil infrastructure. Two primary challenges are: (1) applying system identification (SID) methods to large, complex civil structures, while quantifying the influence of environmental/operational conditions on SID outcomes; and (2) utilizing and calibrating mechanics-based nonlinear finite element (FE) models as effective tools for SHM/DP.This dissertation addresses the two above primary challenges by performing: (1) linear SID of the UCSD Geisel Library using seismic and ambient vibration (AV) accelerometer data, and (2) Bayesian nonlinear FE model updating for a reinforced concrete (RC) structural wall subjected to quasi-static cyclic loading. State-space-based SID methods are employed to identify modal properties of the library using seismic data from low-intensity earthquakes. Additionally, higher modal properties are identified using AV data. The modal properties derived from eigenvalue analysis of a FE model of the Geisel Library are compared to the identified modal properties. This study also examines the influence of atmospheric conditions and the ambient vibrations’ amplitude on the identified modal properties. A Bayesian updating methodology is used to calibrate a nonlinear FE model of a RC structural wall, incorporating both the recursive unscented Kalman filter (UKF) and batch transitional Markov Chain Monte Carlo (TMCMC) methods. A novel approach for constraining estimation parameters within the UKF is developed and validated. A global sensitivity analysis of the nonlinear FE model parameters is conducted for screening prior to Bayesian inference. The parameter estimation results from the UKF and TMCMC methods are compared and analyzed. The Bayesian-updated nonlinear FE model can be used to: (1) assess the presence, location, classification, and evaluation of structural damage for a given input loading, and (2) conduct damage prognosis for performance under future loads.