Applied Machine Learning for Simulated Reprocessing Safeguards: Unsupervised Networks

Volume: 63, December 2021, pages 15-29,
Special Issue on Data Analytics for Safeguards and Non-Proliferation

Authors: Nathan Shoman¹, Benjamin Cipiti¹, Thomas Grimes¹, Ben Wilson², Randall Gladen²

¹_{Sandia National Laboratories,} ²_{Pacific Northwest National Laboratory}

Abstract:

A goal of the International Atomic Energy Agency (IAEA) is to deter the spread of nuclear weapons through detection of nuclear material and technology misuse. Detecting diversion of nuclear material from large bulk handling facilities, such as a reprocessing plant, is a goal that can prove to be both challenging and resource intensive as it often requires destructive analysis of numerous samples taken from various locations across the facility. The IAEA has sought out methods to develop an integrated system of instrumentation and data processing to reduce this burden. The goal of this work is to leverage machine learning (ML) methods to improve the effectiveness and efficiency of safeguards by utilizing higher uncertainty measurements, such as process monitoring and Non-Destructive Assay measurements, which are not extensively used in traditional safeguards methods. This work is part of a series of two documents that consider the use of ML to improve one aspect of safeguards, namely nuclear mater ial accountancy. This part considers unsupervised networks that are used to detect anomalous behavior that could be indicative of material loss. The unsupervised approach is shown to exceed traditional methodologies but only after several practical barriers have been accounted for and resolved.

Keywords: safeguards; data science; machine learning; nuclear material accountancy; reprocessing

Reference guideline:

Shoman, N., Cipiti, B., Grimes, T., Wilson, B., & Gladen, R. (2021). Applied Machine Learning for Simulated Reprocessing Safeguards: Unsupervised Networks. ESARDA Bulletin - The International Journal of Nuclear Safeguards and Non-proliferation, 63, 15-29.
https://doi.org/10.3011/ESARDA.IJNSNP.2021.8