In this study, a comparison is presented between two distinct approaches for interpreting validation results for light water reactor (LWR) fuel criticality safety assessments: one based on frequentist tolerance limits and the other on a Bayesian framework. In general, both the frequentist statistical methods and Bayesian models have inherent advantages and disadvantages, making it valuable to compare the results of the criticality safety evaluations (CSEs) obtained using both approaches. Of particular interest in this context is the application of CSE in conjunction with the burnup credit concept for LWR used nuclear fuel (UNF), whose composition differs significantly from that of fresh fuel, which is primarily used in validation studies worldwide.

This paper aims to illustrate a comparative analysis of different CSE methodologies applied to a model of a UNF disposal canister filled with identical fuel assemblies as a function of burnup. The study found that the Bayesian approach yielded less penalizing results, leading, in the analyzed case, to a relaxation of the burnup requirement for UNF criticality safety by approximately ~2 to 3 GWd/tonnes heavy metal for pressurized water reactor fuel with an initial 235U enrichment of 5 wt%.

However, an interesting and somewhat counterintuitive behavior was observed in that the Bayesian-based results indicated a reduction in the safety margins as burnup increased, despite the absence of benchmarks with UNF in the employed validation suite. In any case, the observations and discussions presented suggest that the performance of both the frequentist and Bayesian methodologies, as applied in the context of the postulated CSE task and the employed nuclear data with associated uncertainties, requires further investigation before these approaches can be routinely and effectively adopted for licensing applications involving UNF.