Identification of spent nuclear fuel with multivariate analysis based on database

Abstract

In recent years, as the international situation about nuclear arms control changes, the multilateral international arms control cooperation including non-proliferation, nuclear terrorism and nuclear security has drawn more and more attention. As a key technology, nuclear forensics plays a significant role in the deterrence, prevention and response to illegitimate nuclear activities, for which it needs studying in depth. At present, there are plenty of researches into nuclear forensics, mostly focusing on the characterization of materials and the interpretation of data. As one of the most important aspects of nuclear forensics research, the interpretation is faced with a variety of different objects, including uranium ore, yellow cake, nuclear fuel, spent nuclear fuel and so on, among which spent nuclear fuel has attracted more and more attention due to its potential threats. In this paper, we primarily focus on the development of multivariate statistical analysis with an aim to interpret the comparative signatures of spent nuclear fuel. A database is established with uranium and plutonium isotopic compositions of spent nuclear fuel samples through simulation. These samples are of different reactor types, initial fuel enrichments and burn-ups. Subsequently, multivariate analysis, including factor analysis, discriminant analysis and regression analysis, are used to the database to validate the feasibility of the identification work. First of all, dimension reduction and visualization work is carried out to determine the possibility for classification by factor analysis. Afterwards, some known samples are assumed to be unknown to further study the possible capabilities of quantitative attribution by conducting factor analysis, including the determination of initial fuel enrichment and burn-up. To eliminate the errors in the identification work and to achieve better outcomes, the discriminant analysis and regression analysis are used to the database to assist with the identification of the reactor type, initial fuel enrichment and burn-up. As revealed by the study, factor analysis is more suitable for the dimension reduction and visualization work, disciminant analysis is more suitable for the identification of reactor type, and regression analysis is more suitable for the identification of initial fuel enrichment and burn-up. Upon the comparison drawn of the three different multivariate analysis methods, a framework for identification process is established to provide a theoretical basis and feasible scheme for the possible identification work of spent nuclear fuel with database, and facilitate the related nuclear forensics work.