NE Seminar: “Experimental and Computational Methods to Discriminate Chemically Separated Plutonium”


1:55 pm-2:55 pm
Add to Outlook/iCal
Add to Google Calendar


Rhines Hall Room 125
549 Gale Lemerand Drive
Gainesville, FL 32611


Sunil S. Chirayath, Ph.D.

Associate Professor, Nuclear Engineering
Texas A&M University



Efforts have been ongoing at Texas A&M University to develop a robust nuclear forensics methodology capable of attributing the conditions of origin for separated plutonium samples. This methodology should be able to accurately identify the reactor-type that produced the plutonium sample, as well as the sample’s burnup and time since irradiation (TSI).

An initial approach to this problem yielded a methodology that found reactor-type, burnup, and TSI using a maximum likelihood method to compare the intra-element isotope ratios of an unknown sample to those in a library of simulated. This approach could accurately attribute plutonium sourced from a single reactor-type but was unable to attribute a sample that was a mixture of plutonium from multiple reactor sources.

A new methodology is under development that can improve in this area. The library of intra-element isotope ratio library of the first methodology was leveraged to produce predictive models trained using machine learning to perform the attribution rather than the maximum likelihood calculation. The machine learning methodology uses a classifier to determine the reactor-type of origin, a regression model to determine burnup, and the TSI is predicted analytically.

After verifying that the machine learning approach could replicate the success of the maximum likelihood approach for single reactor source plutonium samples, work is now underway to produce new models that have been trained using mixed source plutonium data. This is accomplished by training a new classifier that contains data from the original reactor-type classes, but also new classes with isotope ratios that are characteristic of plutonium produced by mixing two of the original classes. This necessitated large changes to how the training data set was produced and the classifier training regime. Future work aims to produce a regression model capable of finding mixture ratios for the multiple reactor source plutonium.


Hosted by

Department of Materials Science & Engineering