A Bayesian Framework of Transmissibility Model Selection and Updating

Abstract

For the purpose of system identification and structural health monitoring, transmissibility, as an output-output transfer function, plays a very important role because of its compatibility with input absent data and local parameter sensitivity. However, data-driven transmissibility estimations are always subject to uncertainties, and tremendously increasing the length of measurements and/or number of averages to suppress some types of uncertainty influence are often impractical. Based upon the likelihood function established in previous work, this paper adopts a Bayesian framework to probabilistically select the most plausible class of models and update parameters with a limited amount of test data. A structural computational model is considered as a test-bed on which the approximated transmissibility model is computed; consequently, under a quantified confidence, the quality of transmissibility estimations is enhanced via a lower order approximate model, with a much lower demand on data acquisition requirements.