EACS 2016 paper - QUANTIFICATION OF UNCERTAINTY FOR EXPERIMENTALLY OBTAINED MODAL PARAMETERS IN THE CREATION OF A ROBUST DAMAGE MODEL
EACS 2016 Paper No. 191
Computational modelling is an important method in generating predictive models of engineering systems. These computational models are generally deterministic and therefore often ignore the inherent uncertainty in experimental results. Where the model predictions are to be used for damage identification this lack of uncertainty can lead to less robust classification, as damage states can appear more clearly separated than may be true for experimental data. The approach used in generating damaged state model predictions must therefore identify and quantify the main sources of uncertainties. By quantifying the main sources of uncertainties, a Naive Bayes approach can be used to define decision boundaries that incorporate this uncertainty, improving damage predictions. The combination of the computational model and a Naive Bayes approach will lead to a more detailed and realistic representation of the actual system. In this paper quantification of the uncertainties from modal tests for a prismatic metallic cantilever beam, with different levels of damage, is presented. The main sources of uncertainty are categorised and quantified before being applied to computational models using a Naïve Bayes approach. The probability of the likelihood of damage classification is then shown for the inclusion of uncertainty in the damage model, showing the improvement in decision bound and therefore the improvement in the damage model.