François M. Hemez

The figure illustrates the validation of a family of numerical simulations developed to predict the propagation of a mechanical impact through a layer of crushable foam material. The trade-offs between three attributes of prediction accuracy are shown: fidelity-to-data (R), robustness-to-uncertainty (α*), and confidence in prediction (CF).Fidelity assesses the correlation between measurements and predictions. Robustness indicates the effect of uncertainty on prediction accuracy, using a family of convex models of uncertainty to represent the lack-of-knowledge about the crushable foam constitutive behavior. Confidence derives from the consistency of predictions made by multiple numerical simulations. Confidence is estimated from a total uncertainty metric that collects multiple sources of non-probabilistic uncertainty. The figure indicates that calibration leads to zero robustness and over-confidence. Fidelity-to-data cannot be improved without reducing robustness to lack-of-knowledge. Likewise, increasing robustness comes at the cost of decreasing confidence. Examining these trade-offs is essential to assess the prediction accuracy and adequacy of the numerical simulation. To learn more, visit Sample of V&V Projects or Downloads.