Classification of spent reactor fuel for nuclear forensics

Area of Science:

• Nuclear Forensics
• Applied Machine Learning
• Isotopic Analysis

Background:

• Nuclear forensics relies on analyzing spent nuclear fuel's isotopic and elemental composition to determine its origin and processing history.
• Traditional methods focus on individual material properties, which can be challenging due to the complexity and numerous variables in nuclear materials.
• Multivariate statistical analysis and dimensionality reduction techniques have shown promise in extracting more information from isotopic data for nuclear forensics.

Purpose of the Study:

• To apply advanced pattern recognition and machine learning techniques for determining the reactor type of origin for spent nuclear fuel.
• To explore robust dimensionality reduction techniques, specifically manifold embedding, for enhanced analysis of intrinsic dataset information.
• To implement and evaluate novel classification algorithms for reliable spent fuel classification within the field of nuclear forensics.

Main Methods:

• Utilized isotopic and elemental measurements of spent nuclear fuel samples.
• Employed advanced dimensionality reduction techniques, including manifold embedding, to capture intrinsic data characteristics.
• Applied a range of classification algorithms, including novel approaches for nuclear forensics, to categorize fuel by reactor type.

Main Results:

• Demonstrated the successful application of pattern recognition and machine learning for accurate spent fuel reactor type determination.
• Showcased the effectiveness of manifold embedding in capturing complex relationships within the isotopic and elemental data.
• Achieved reliable classification of spent fuel, confirming the robustness of the developed models.

Conclusions:

• The developed machine learning and pattern recognition techniques provide a powerful and robust tool for nuclear forensics.
• These methods significantly enhance the ability to determine the origin of spent nuclear fuel by analyzing multiple material parameters simultaneously.
• The study confirms the excellent potential of these advanced analytical techniques for improving nuclear forensics capabilities, particularly for spent reactor fuel.