Addressing practical aspects of the nuclear criticality safety evaluations with frequentist statistics

Abstract

General guidelines for Criticality Safety Evaluations (CSE) and regulatory requirements on the Nuclear Criticality Safety are well established nowadays. However, the expert knowledge is often still required in practice. Such situation is typical for the CSE methodologies grounded on the frequentist statistic inference and involving tolerance limits estimations and hypothesis testing. In this respect, various sensitivity tests are illustrated in the given paper to enhance understanding of the statistical assessments, in application to conventional CSE for Light Water Reactor (LWR) fuel configurations, when relatively small validation samples are generally available. It is further discussed that for rare situations with very small validation data samples, both parametric and non-parametric statistical methods have drawbacks: unreliability of normality tests for the first case and large non-parametric margins historically recommended for the second case. For such specific situations, an additional safety margin named “Normality Margin” is proposed in this work, towards compensation of the lack of knowledge of the underlying hypothetical probability distribution describing the criticality validation results.