INCLUSION OF UNCERTAINTY QUANTIFICATION IN MODEL VERIFICATION AND VALIDATION

摘要

This paper proposes a Bayesian methodology to develop validation metrics that integrate uncertainties in both model prediction and experimental measurement. The simulation models that we address in this paper are primarily finite element-based structural analysis and limit state-based reliability prediction models. Several components of computational prediction error, such as discretization error, element error, and stochastic analysis error are included. Two types of measurement error are included, in the context of model validation: error in the measurement of input variables that affects the model prediction, and error in the measurement of output variables. These error estimates and experimental results are integrated to compute the statistics of model form error through a bootstrapping technique.