Uncertainty quantification of model-form and predictive uncertainties in nuclear codes using Bayesian framework

Bayesian-based model selection and UQ methodology is developed and applied to nuclear thermal-hydraulics codes and void fraction data in this paper. Uncertainties inherent in the experimental data along with the predictive and model-form uncertainty are quantified to construct a composite/hybrid model based on the competent models to predict the response with confidence. The predictive uncertainty or model discrepancy dominates the model-form uncertainty for the void fraction at low axial locations. Improvements in composite predictions are observed at higher axial locations at which model-form uncertainty plays a major role. It is found also that including the measured data uncertainty during the UQ process improves the model prediction performance, instead of assuming perfect data and penalizing the models. The proposed methodology is flexible and extendable to other types of physics, models, and data. Developing the underlying methodology of estimating the model weights will be the main focus of the subsequent studies.