Uncertainty quantification in vibration-based damage assessment by means of model updating

The success of vibration-based damage identification procedures depends significantly on the accuracy and completeness of the available identified modal parameters. This paper investigates the level of confidence in the damage identification results as a function of uncertainty in the identified modal parameters through a probabilistic damage identification strategy, i.e., Bayesian finite element (FE) model updating. This method allows accounting for pertinent sources of uncertainty and expresses the damage identification results in probabilistic terms. In the physical model, damage is represented by a local decrease in stiffness, so that the updating parameters correspond to the effective stiffness of a number of substructures. An accurate localization of damage is difficult, as the lowfrequency global modes of the structure are relatively insensitive to a local change in stiffness. Within the frame of the present paper, experimental modal data of a beam are used to identify the flexural stiffness distribution along the beam. The experimental data have been obtained from a test where a full scale subcomponent composite beam has been progressively damaged in several stages through quasi-static loading. A Bayesian inference scheme is used to quantify the uncertainties in the model updating procedure, based on an estimation of the combined measurement and modelling uncertainty. The results of the identification are compared to the observed damage (visual inspections) along the beam for validation purposes. This paper shows the possibilities of probabilistic damage assessment procedures based on real vibration data.