Finite element model updating seeks to modify a structural model to reduce discrepancies between predicted and measured data, often from vibration studies. An updated model provides more accurate prediction of structural behavior in future analyses. Sensitivity-based parameter clustering and regularization are two techniques used to improve model updating solutions, particularly for high-dimensional parameter spaces and ill-posed updating problems. In this paper, a novel parameter clustering scheme is proposed which considers the structure of the objective function to facilitate simultaneous updating of disparate data, such as natural frequencies and mode shapes. In a small-scale updating example with simulated data, the proposed clustering scheme is shown to provide moderate to excellent improvement over existing parameter clustering methods, depending on the accuracy of initial model. A full-scale updating example on a large suspension bridge shows similar improvement using the proposed parametrization scheme. Levenberg-Marquardt minimization with Bayesian regularization is also implemented, providing an optimal regularized solution and insight into parametrization efficiency.
Summary
Model updating is used to reduce error between measured structural responses and corresponding finite element (FE) model outputs, which allows accurate prediction of structural behavior in future analyses. In this work, reduced‐order parametrizations of an underlying FE model are developed from singular value decomposition (SVD) of the sensitivity matrix, thereby improving efficiency and posedness in model updating. A deterministic error minimization scheme is combined with asymptotic Bayesian inference to provide optimal regularization with estimates for model evidence and parameter efficiency. Natural frequencies and mode shapes are targeted for updating in a small‐scale example with simulated data and a full‐scale example with real data. In both cases, SVD‐based parametrization is shown to have good or better results than subset selection with very strong results on the full‐scale model, as assessed by Bayes factor.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.