AI-fueled US lab sniffs out rogue nuclear bombs and keeps us safe

Machine learning has been leveraged to accelerate analysis in nuclear processing facilities and investigations in the field.
Amal Jos Chacko
PNNL will use machine learning to detect nuclear threats quicker.
PNNL will use machine learning to detect nuclear threats quicker.

Devrimb/iStock 

Surprise nuclear attacks or threats will soon be a thing of the past. Researchers at the Department of Energy's Pacific Northwest National Laboratory (PNNL), U.S., have developed new techniques to accelerate the discovery and understanding of nuclear weapons by leveraging machine learning.

One enticing application of these new techniques in national security is to use data analytics and machine learning to monitor several ingredients used to produce nukes.

The International Atomic Energy Agency (IAEA) keeps a stern eye on nuclear processing facilities in non-nuclear weapon states to ensure that by-products of spent nuclear fuel, especially plutonium, are not diverted to producing nuclear warheads. These regular inspections safeguard nuclear materials in tandem with sample analysis and process monitoring.

However, with the massive rates of real-time data today, analyzing and processing data obtained from these facilities is arduous and labor-intensive. Analysts at PNNL have designed a system capable of automatically detecting suspicious activity. Steven Ashby, director of PNNL, describes the different techniques employed to flag such activities in a press release.

Predicting real-world results with virtual models

This design follows the "transformer-based model," a machine learning technique that finds its applications in use cases such as natural language processing, where the position of the data is key.

Researchers build a virtual replica model of these reprocessing facilities and then assign weights to the data obtained from the actual facility. Learning and training on this data enable the model to analyze patterns in how nuclear materials are being handled and then predict expectations for various locations within the facility.

These predictions are then held against actual observations to examine differences which are then flagged.

Although in the early days, this technique shows good promise.

Machine learning turns field agent

Another application PNNL and its collaborators identified to be suitable for machine learning is in the field when law enforcement agencies stumble upon nuclear materials. By leveraging a model called autoencoder, the software can quickly match the material's "fingerprint" against a library of electron microscope images of radioactive particles.

This could go a long way in speeding up investigations by determining the origin of these materials, guiding investigations, and preventing nukes from being produced by these materials accordingly.

Much exciting as it sounds, "Machine Learning algorithms and computers will not replace humans in deterring nuclear threats any time soon. But they may make it possible for people to discover important information and identify risks more quickly and easily," notes Ashby.

This new development sees yet another notch in favor of machine learning that makes the world safer for us to live in.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board