System identification of constructed civil engineering structures and uncertainty

Characterization of constructed civil engineering structures through system identification has gained increasing attention in recent years due to its tremendous potential for optimum infrastructure asset management and performance-based engineering. However, the lack of reliability in system identification, especially when applied to large-scale complex constructed systems, poses a major challenge for its widespread implementation. It is believed that this primarily stems from epistemic uncertainty associated with identification processes, due to unknown or less understood structural behaviors as well as the interaction of the system with its environment. The objective of this thesis is to investigate the effects of epistemic uncertainty on the reliability of identification and to develop solutions to recognize and mitigate these uncertainties. The research which was undertaken included laboratory and field investigation as the primary components. First, a cantilever beam with two test configurations was designed and constructed in the laboratory as a test bed. By comparing different identification scenarios, the impact of modeling uncertainty with epistemic mechanism on the field-calibrated analytical model was evaluated. Feasible techniques were developed to recognize and mitigate significant epistemic modeling uncertainty which controls the test-analysis discrepancy. In applications of system identification on real-life structural systems, the tempo, frequency and spatial incompatibility between detailed finite element model and information contained in test measurements often further complicates the identification process. It was demonstrated through the Henry Hudson Bridge that it was possible to characterize the fundamental behaviors of large-scale complex structures by integrating heuristics and conventional techniques. Measurements to assess the adequacy of the field-calibrated models were proposed to ensure that significant epistemic modeling uncertainty was efficiently reduced and critical physical mechanisms was properly conceptualized.