Abstract:
Uranium ore concentrates(UOCs) are a kind of nuclear materials which can be traded legally in the international market. Uranium ore concentrates have become attractive targets for nuclear forensic because of the richness in characteristic signatures compared to other materials produced later in the fuel cycle. The rich trace impurities of UOCs is an ideal object for nuclear forensic analysis, in which it reflects the important information of uranium ore formation, mineral composition, process of uranium mining and geolocation. As one of the important characteristic fingerprints, the rare earth elements(REEs) distribution pattern of UOCs can be used as an attractive alternative to attribute the origin of uranium ore concentrates. In this paper, several multivariate statistical methods in pattern recognition, including principal component analysis(PCA), factor analysis(FA), cluster analysis(CA) and partial least squares(PLS) were compared and the advantages and disadvantages of various methods were summarized. The results show that it is a tough work to fully describe the characteristic fingerprint parameters of uranium ore concentrates by only one or two methods. The principal component analysis(PCA) method can reduce the dimension of the data and retain the original characteristics of the data. However, the classification of the data is not taken into account in the process of analysis and the results only show the difference between the samples. The factor analysis(FA) method is essentially a process of extracting potential factors from dominant variables and the form of factors is not unique. Cluster analysis(CA) can classify the origin of uranium ore concentrates. However, there is no quantitative data to illustrate the difference in the same classification of uranium ore concentrates. The partial least squares(PLS) method takes into account the difference between all the data points from the algorithm, and PLS can also perform multiple iterations to identify the subtle difference of uranium ore and uranium ore concentrates with the same origin, so that the precise geolocation of uranium ore concentrate can be realized. It is generally used in regression studies with less sample size, so PLS can show more detailed fingerprint features than PCA. To date, the scientists still face a great of difficulties to attribute the geolocation of uranium ore concentrates, especially to those uranium ore concentrates mined and milled in the countries or regions with vast territory and complex geological features. The combination of several multivariate statistical methods might be an attractive avenue to accurately and effectively trace the origin of uranium ore concentrates, but it still has a long way to go.