UC San Diego

Aging, man-made and natural hazards (e.g., earthquakes and hurricanes) may induce significant damage or even cause the collapse of civil structures. Such damage and failures imply life and economic losses, and function disruption of critical facilities; however, these devastating consequences can be reduced by means of accurate and timely risk mitigation decisions taken before and after the damage-inducing event. Structural health monitoring (SHM) has emerged as an attractive technology for the research and engineering communities to provide tools and protocols aiming to inform and prioritize the decision- making process and, therefore, has been accepted as a critical tool to achieve sustainable and resilient communities. Condition-based inspection and monitoring strategies to assess the residual life, detect any damage and safety threat at the earliest possible stage, and prioritize the repair or replacement of critical infrastructure are crucial preventive and proactive actions that can be facilitated by the use of advanced SHM methodologies. Because of recent advances in computational resources and cost reductions in sensor technologies, nowadays, dense and sophisticated sensor networks have been deployed and are collecting data for different types of civil structures throughout the world. However, current methods and practices in SHM are not achieving the goal of supporting the decision-making process. In this regard, two major hurdles are : (1) there is still a need to validate current state-of-the-art system identification (SID) and damage identification (DID) methods using data recorded from large and complex civil structures subjected to real or realistic damage-inducing events (e.g., man- made or natural hazards such as earthquakes), and (2) there is a disconnect between the advances made in the subfields of SHM and mechanics-based modeling and simulation of structures. This dissertation contributes to overcome these two hurdles by (1) analyzing vibration data recorded from a full-scale five-story reinforced concrete (RC) building fully outfitted with nonstructural components and systems, which was seismically tested and subjected to progressive damage on the NEES@UCSD shake table, and (2) developing and validating a novel and advanced SHM and DID framework that integrates high- fidelity mechanics-based nonlinear finite element (FE) structural modeling and analysis with state-of-the-art Bayesian inference methods. The first part of this dissertation focuses on SID and dynamic characterization of the full-scale five-story RC building specimen. Dynamic data for different sources of excitation, including ambient vibration as well as free and forced vibration tests, are used to investigate the evolution of the modal properties during construction of the building. Variations of the modal properties of the building, under both fixed- base and base-isolated configurations, due to the effects of nonstructural components, seismic-induced damage, and environmental conditions are explored comprehensively. In the second part of the dissertation, a novel framework is developed for system and damage identification of nonlinear structural systems subjected to known or unknown inputs. The proposed framework is validated using homogeneous and heterogeneous sensor data simulated from realistic nonlinear FE models of structures of increasing complexity, including 2D and 3D steel and RC frame structures, subjected to seismic excitation. Stochastic (Bayesian) filtering methods, including the Unscented Kalman filter and the Extended Kalman filter, are used to estimate unknown parameters of the FE model, unknown input base excitations, and their estimation uncertainties using spatially-sparse noisy measurements. By employing the estimated model parameters and input excitations, the updated nonlinear FE model can be interrogated to detect, localize, classify, and assess the damage in the structure, and can also be used for damage prognosis purposes