| The overarching aim of this thesis is to develop, validate and assess the value of a novel multiple model (MM) Structural Identification (St-Id) approach to inform decisions related to the preservation of critical infrastructures, such as long-span bridges. The approach developed employs: (a) heuristic knowledge along with various search and sampling techniques to generate populations of candidate models (composed of distinct parameter values, parameterization approaches and/or model forms), (b) various deterministic and probabilistic weighing approaches driven by a set of observed responses, and (c) simulations to estimate un-measureable attributes and their respective variability. To examine the value of this approach compared to conventional single-model methods, a multi-phase research program was carried out that included numerical applications, laboratory studies (using a physical grid model), and ultimately an application to an operating long-span bridge. In addition to numerous conclusions regarding the specific formulation and application of the approach, three principal conclusions were drawn. First, the proposed approach is capable of estimating the degree of non-uniqueness associated with the model-experiment correlation process, and thus provides important insight into the nature and reliability of the desired predictions. Second, the predictions produced by the method implicitly reflected the strength of the correlation between the observations and the desired responses. In addition, it was shown that the strength of this correlation could be amplified if proper mechanistic or heuristic-based weighing schemes were employed. Third, the method was capable of identifying member "elastic" redundancy within a complex structural system through the estimation of the variability of member forces (which is a direct measure of the presence of multiple load paths). Using this approach, it was shown that the level of redundancy not only varies significantly from member to member within a large constructed system, but that redundancy can vary for a single member depending on the direction of the forcing function. |
