System Identification Methods Applied to Measured Seismic Response

Abstract

A unified Bayesian statistical framework is described for system identification which can be used to extract important information for earthquake-resistant design from the measured seismic response of structures. In this approach, the "best" (optimal) models within a chosen class of models are those which are locally most probable based on the available data. Using the methodology, one can determine the prediction accuracy of the optimal structural models, the precision of the parameter estimates of these models as well as the precision of the optimal prediction-error probability models, the updated predictive probability density function for the structural response even when there are multiple optimal models, and a principle of parsimony for comparing different classes of models on the same data. The application of modal identification to measured seismic response from some buildings, a bridge and an off-shore platform is reviewed. An example is also given of how the methodology can be used to handle the nonuniqueness in structural model identification. This often arises in practice because of the low density of sensors which is typical of structural seismic instrumentation arrays.