Advancing Nuclear Forensics, Machine Learning methods for Nuclear Crime Scene Investgation, ADVANFO (IAEA)

With the increasing global emphasis on nuclear security and non-proliferation, the detection and identification of nuclear and radioactive materials at the radiological crime scene are of paramount importance. Traditional techniques with portable detectors often suffer from false positives, limited specificity, and challenges in real-time detection, especially in complex environments (e.g. high radiation background) or with shielded materials. The primary problem this research addresses is the inherent difficulty in accurately detecting and identifying nuclear and radioactive materials in diverse environments, amidst various forms of interference with different radiation detectors. Current methodologies often lack the granularity and adaptability needed to discern between materials, especially in scenarios with overlapping signatures or in close proximity to benign radioactive materials.

This research aims to bring to the crime scene the confidence achievable with thorough investigation in the laboratory, exploiting artificial intelligence (AI) algorithms on the in-field spectroscopic measurements conducted with portable detectors. Specifically, the GEANT4 or MCNP simulation tools will be utilized to generate detailed and comprehensive signatures of various nuclear and radioactive materials in various environmental scenarios. These simulated signatures will then serve as the foundation to train advanced machine learning algorithms, providing them with a rich dataset that is otherwise challenging and risky to obtain in real-world conditions.