| The confluence of a debilitated, often obsolete bridge population nearing the end of its design life with a massive shortfall of funding for rehabilitation and replacement has led to an unsustainable balancing act for bridge owners. One tool for addressing this issue is Structural Identification (St-Id), a six-step process for effectively integrating subjective information from visual inspections, with analytical modeling and experimental investigations to provide a more robust and defendable foundation for decision-making regarding constructed systems.;In an effort to understand and document St-Id from a global perspective, the International Bridge Study (IBS) brought researchers from all over the world to demonstrate various techniques and technologies on a single test specimen in New Jersey. Through this application a comprehensive documentation of a best-practices application of St-Id was developed and numerous advances to each step of the process and their integration were achieved. These advances included formalization of the criteria for bridge selection as a candidate for St-Id, development of a framework for instrumentation design considering safety and test objectives, design and implementation of a distributed data acquisition system with a real-time visualization interface and spatially-correlated data interpretation.;As part of this application some shortcomings associated with conventional model-experiment correlation approaches were identified. The second phase of the research aimed to develop and validate rigorous methods to mitigate these shortcomings. This research examined multiple-model approaches such Markov Chain Monte Carlo (MCMC) and multi-dimensional MCMC methods through both a benchmark numerical problem and the IBS Bridge. The primary findings in this phase included the difficulty in framing the problem to fully map the model space using Reversible Jump MCMC, an alternative approach through a multi-dimensional MCMC method, the difficulty in applying MCMC to actual bridges due to the complexity of the likelihood space and that MCMC can implicitly account for interaction between parameters through covariance and correlation coefficients. In general, the response predictions from the MCMC chains were more robust and informative than the results from the single model correlation. |
