Influence of the prediction error correlation model on Bayesian FE model updating results

Abstract

Bayesian finite element (FE) model updating is a probabilistic method for uncertainty quantification in FE model updating, where the well-known Bayes' theorem is used to update probability density functions of model parameters, accounting both for the information contained in the data and for uncertainties present in the measurements and model predictions. In order to apply the Bayesian technique successfully, one has to adopt a probabilistic model for the error between predictions and observations. In most structural mechanics applications, however, the common assumption of a zero-mean uncorrelated (Gaussian) error is not always valid. In this paper, it is shown that accounting for prediction error correlation between sensors has significant influence on FE model updating results, and moreover that the choice of a suitable correlation model for the problem at hand is paramount. This is demonstrated by applying the Bayesian scheme to a number of simulated experiments.