Characterization of plutonium for nuclear forensics using machine learning techniques

Traditional nuclear forensics methods aim to ascertain key pieces of information regarding material provenance using material characteristics. In 2018, a novel nuclear forensics methodology for source discrimination of separated plutonium was developed, which utilizes a maximum likelihood formulation and intra-element isotope ratios to identify the most likely source-reactor-type that produced plutonium, fuel burnup and time since irradiation simultaneously. In this research, the same intra-element isotope ratios along with four well established machine learning techniques for classification are used to determine the most likely source-reactor-type and multivariate linear regression to determine the fuel burnup. The results obtained on the test data set are promising and demonstrate the significant opportunity in the application of machine learning techniques (both for classification and parameter reconstruction) and separation-process agnostic forensic signatures to the nuclear forensics of separated plutonium, in case of an interdiction.