Age Dating of Repurified and Mixed Plutonium Using Machine Learning

Abstract

Age dating a sample of nuclear material is a key part of predetonation technical nuclear forensics. As plutonium stockpiles age, they are more likely to require repurification and mixing to remove in-grown daughter products and maintain a consistent product. Existing age-dating techniques do not adequately address this problem. Four models were trained using machine learning techniques to determine (1) if a sample of weapons-grade plutonium had been repurified, (2) the elapsed time after repurification, and (3) the minimum and maximum elapsed times between repurification and its initial separation/purification/fabrication. The trained models predicted the repurification status with 99% accuracy, the age after repurification with a root-mean-square error (RMSE) of 0.34 years, and the minimum and maximum ages before repurification with RMSEs of 4.66 and 9.34 years, respectively. Age dating plutonium provides valuable insight into the country and possibly the facility of origin of the material, which is one tool to deter state-sponsored nuclear terrorism.