AI brings a new tool to nuclear hide-and-seek

Published in

Purdue Engineering Review

4 min readApr 7, 2022

Sometimes you can’t get somewhere, but you still have to see inside. That’s as true for the hidden chambers of the Egyptian pyramids as it is for the handling and non-proliferation of nuclear fuel. A little bit of artificial intelligence (AI) can come in handy.

Our research focuses on developing new monitoring, imaging, and AI algorithms to optimize cosmic ray muon tomography for nuclear engineering and national security applications. The idea is that better inspection of the heavily shielded containers that store spent (used) nuclear fuel could provide efficient, inexpensive, remote assessment of the contents.

Cosmic ray muon tomography is an imaging technique that uses naturally-occurring background radiation to generate three-dimensional images of complex structures that could not be produced using conventional methods like X-rays. Muons are heavy particles that are created when atoms from outer space collide in Earth’s atmosphere, and rain down on the Earth with very high energies that can penetrate dense materials.

Muon imaging can provide critical knowledge about what’s inside the dense, sealed containers, called dry casks. Nondisclosure of missing spent nuclear fuel — which contains uranium and plutonium that could be used to build nuclear weapons — poses a direct threat to public safety and national security. For safety concerns, international safeguard inspectors responsible for verifying the contents of these casks cannot open the casks.

Our Radiation Imaging and Nuclear Sensing Laboratory is developing real-time monitoring algorithms derived from machine learning and Bayesian (statistical) principles to minimize the risk of incorrectly determining how much spent nuclear fuel is inside a cask. Optimal algorithms could verify the contents of casks of spent nuclear fuel in minutes, whereas conventional imaging algorithms typically require hours to days.

In addition, we conduct robust evaluations of an algorithm’s performance and determine the lower detection limit — an important parameter for the minimum number of cosmic ray muons needed to identify missing spent nuclear fuel. Essentially, we calculate the minimum muons required to indicate whether spent nuclear fuel casks are full.

Dry cask storage area for spent nuclear fuel (Photo credit: Nuclear Regulatory Commission)

Using this information, safeguard inspectors can plan ahead for the resources needed to ensure that missing spent nuclear fuel will not go undetected. This approach could be extended to other applications related to national security, including identification of hidden nuclear materials in cargo containers and characterization of high-density materials in waste packages.

Our innovations also can be used in the development of advanced muon detectors, and new monitoring and imaging reconstruction algorithms for reducing national security threats and lessening the time needed to identify nuclear materials. Improving our capability to secure nuclear materials will help decrease uncertainty in spent fuel transportation, and eventually may lead to the operation of underground repositories that securely dispose of nuclear waste.

Maintaining continuity of knowledge of nuclear materials is a top priority for agencies, such as the International Atomic Energy Agency (IAEA), that safeguard and protect nuclear materials around the world. Preventing nuclear terror has been recognized by the National Academy of Engineering as one of the Grand Challenges for Engineering in the 21st century.

Cosmic ray muons have been used successfully to search for hidden compartments in the Great Pyramid of Giza, built as a tomb for an Egyptian pharaoh more than 4,000 years ago. People still try to hide things today, and we’re applying AI to the hunt.