A Bayesian Approach for Designing Experiments Based on Information Criteria to Reduce Epistemic Uncertainty of Fuel Fracture During Loss-of-Coolant Accidents

In probabilistic risk assessment (PRA), the fracture limit of fuel cladding tubes under loss-of-coolant accident conditions plays a critical role in determining the core damage, highlighting the need for accurate modeling of cladding tube fracture behavior. However, for high-burnup cladding tubes, it is often infeasible to conduct extensive experiments due to limited material availability, high costs, and technical constraints. These limitations make it difficult to acquire sufficient data, leading to substantial epistemic uncertainty in fracture modeling. To enhance the realism of PRA results under such constraints, it is essential to develop methods that can effectively reduce epistemic uncertainty using limited experimental data. In this study, we propose a Bayesian approach for designing experimental conditions based on a widely applicable information criterion (WAIC) in order to effectively reduce the uncertainty in the prediction of fuel cladding tube fracture with limited data. We conduct numerical experiments to evaluate the effectiveness of the proposed method in comparison with conventional approaches based on empirical loss and functional variance. Two cases are considered: one where the true and predictive models share the same mathematical structure (Case 1) and one where they differ (Case 2). In Case 1, the empirical loss-based design performs best when the number of added data points is fewer than approximately 10. In Case 2, the WAIC-based design consistently achieves the lowest Bayes generalization loss, demonstrating superior robustness in situations where the true model is unknown. These results indicate that the proposed method enables more informative experimental designs on average and contributes to the effective reduction in epistemic uncertainty in practical applications.