The PNNL scientists can trace the origins of a sample by comparing it to a database of electron images from authorized sources.

Researchers at Pacific Northwest National Laboratory (PNNL) in the United States are using machine learning techniques to identify rogue nuclear weapons. The analysis of the chemicals and infrastructure required to produce these weapons is key to detecting undesirable nuclear activity.

PNNL is developing an autoencoder model that processes images of radioactive material to determine the source and production method of nuclear samples. The software creates a unique signature of the sample and compares it to a database of electron microscope images from universities and other national laboratories. The model can help estimate the purity of the unknown sample and trace the source materials to the laboratories manufacturing the nuclear products. The method can help identify whether the material is of sufficient quality to produce a viable nuclear weapon and who is behind it. (The Register)

PNNL’s work has helped law enforcement to target threats and expedite investigations. The International Atomic Energy Agency monitors nuclear reprocessing facilities in non-nuclear-armed states to ensure they dispose of plutonium generated in nuclear power plants correctly and do not stow it away surreptitiously to produce weapons.

PNNL is also developing a technique to train transformer-based software to monitor nuclear reprocessing labs and detect suspicious activity automatically. The software creates a virtual replica of a reprocessing facility and uses data generated by the model tracking temporal patterns to predict what patterns should be observed in various areas within a plant if it is being used for peaceful purposes.

If the data collected from a facility does not match the model’s predictions, experts are called in to investigate further.

However, the use of these automated methods only detects possible illegal nuclear activity, and human experts still need to verify and confirm reports.

Related Story: Nuclear Fusion Startup Zap Energy Achieves First Plasmas; Nabs $160M