Field Test Of A Machine Learning Technique For Safeguards Assessment At A Fuel Fabrication Facility

Many safeguards systems used by the IAEA are extremely expensive because of the need to authenticate the data produced by those systems. Their prohibitive costs limit both the number of systems available to support effective safeguards as well as the current scope of unattended monitoring. Therefore, as an increasing number of facilities require safeguards, inexpensive unattended systems will become necessary. To help pursue this end, we are conducting an ongoing field trial of a machine learning algorithm to draw safeguards conclusions from an array of inexpensive sensors deployed at a fuel fabrication site in the United States. Specifically, we are testing the machine learning method’s ability to identify activities at a fuel fabricator facility and unusual movements that are inconsistent with their expected pattern of behaviors. This small-scale trial is a follow-up to a proof of concept previously conducted on this method. If successful, the conclusion of this test will encourage a follow-up, larger-scale trial in which additional sensor types (e.g., radiation detection devices, scales, etc.) are incorporated into the machine learning method for a full-scale, “real world” application of this method. This trial will help us assess the potential for machine learning technological advancements to advance nuclear safeguards in an inexpensive, accessible way. Findings will have direct and practical implications for international safeguards technology and IAEA operations.