Artificial intelligence in non-medical and non-nuclear power applications of nuclear radiation: A review

Abstract

Artificial Intelligence (AI) and soft computing (SC) have transformed various fields, including nuclear radiation applications. These applications encompass a wide array of challenges, such as material discrimination in dual-energy X-ray radiography, neutron–gamma discrimination, automating complex tasks in particle accelerators, neutron spectrum unfolding, security applications, the imaging of dense structures using cosmic-ray muons, etc. Traditional techniques often face limitations in these areas due to high complexity, data variability, and the need for real-time processing. AI/SC techniques, including artificial neural networks (ANN), fuzzy systems (FS), and evolutionary algorithms (EA), offer novel approaches to overcome these challenges. Dual-energy X-ray radiography, for instance, utilizes modular neural networks to discriminate between different materials and their thicknesses quantitatively. Unlike traditional techniques, this approach ensures higher precision and adaptability. Similarly, in neutron–gamma discrimination, supervised learning methods such as multilayer perceptron (MLP) and clustering techniques like K-means improve the separation accuracy. Neutron spectrum unfolding, an essential process for extracting energy spectra from detectors, is another area where AI/SC demonstrates its strengths. The use of AI in encoding radiation signals and constructing energy spectra further expands its scope. Despite these advancements, challenges persist, particularly the dependency on extensive, high-quality datasets for training AI models, the computational demands of deep learning techniques, and the black-box nature of many AI algorithms that limit interpretability. Addressing these challenges requires collaborative efforts to create open-access datasets, develop transparent and interpretable algorithms, and optimize computational frameworks for real-time applications. In conclusion, integrating AI and SC into applications of nuclear radiation has paved the way for significant advancements in the future, enabling solutions to complex and traditionally unsolvable problems.