A new algorithm could enable faster, cheaper detection of weapons-grade nuclear materials at borders, quickly differentiating between benign and illicit radiation signatures in the same cargo. The development is a collaboration among researchers at the University of Illinois at Urbana-Champaign, the University of Michigan, Los Alamos National Laboratory, Heriot-Watt University (Edinburgh, UK), and the University of Edinburgh.
The growing terrorism threat based on the use of special nuclear materials—highly enriched uranium, weapons-grade plutonium, or materials that produce a lot of radiation—has reinforced the need for improved population protection mechanisms. Nuclear security aims to deter and detect the smuggling of these materials across national borders.
The researchers developed an algorithm capable of identifying weak radiation signals, such as might be seen from plutonium encased materials that absorb radiation. It works even in the presence of a high radiation background, including everyday sources such as cosmic rays from space and radon from the rock underfoot.
Based on their results, they believe the use of their algorithm could improve the ability of radiation portal monitors at national borders to tell the difference between potential smuggling activity and benign radiation sources. For instance, naturally occurring radioactive materials such as ceramics and fertilizers, or radionuclides in recently treated nuclear medicine patients, can set off “nuisance” alarms at radiation scanning facilities.
“There’s also the concern that somebody might want to mask a radioactive source, or special nuclear material, by using naturally occurring radioactive materials such as granite or kitty litter,” said Sara Pozzi, a professor of nuclear engineering and radiological sciences at the University of Michigan and senior author on the paper. “As vehicles or boxes are scanned, the data from the detector can be put through these algorithms that unmix the different sources. The algorithms can quickly identify whether special nuclear materials are present.”
Unmixing sources so that smuggling activity can’t be hidden among benign radiation sources is hard to do quickly. For that, the team turned to specialists in machine learning, who could use data collected by Pozzi’s group to “train” algorithms to look for the signatures of materials that could be used to make a nuclear bomb.
“The algorithm's unprecedented accuracy was not attained using off-the shelf machine learning methods,” said co-author Alfred Hero, John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science at the University of Michigan. “Rather, we crafted an unmixing model that both reflects the basic physics of the problem and was also amenable to fast computation.”
DiFulvio said this research began at Michigan within the framework of the Consortium for Verification Technology, led by Prof. Pozzi at NERS (Nuclear Engineering and Radiological Sciences), and continued when she moved to UIUC in 2018.
“This work is a powerful example of the benefit of close and sustained collaboration between computational data scientists and nuclear engineers, resulting in major improvement of nuclear radiation detection and identification,” Hero said.
Click here to read the full article.
Special thanks to Katherine McAlpine and Catharine June for their contributions.