Simulating nuclear fuel inspections: Enhancing reliability through synthetic data

Visual inspection of nuclear fuel assemblies is critical for assessing fuel reliability and ensuring safe operation. However, the sensitivity of real inspection data, along with its inflexibility and high collection costs, limits its use for research and development (R&D) tasks. These challenges hinder the ability to test and validate new inspection methodologies, making innovation slow and expensive. To address these limitations, we propose the development of synthetic nuclear fuel datasets that simulate fuel assembly inspections. These data sets replicate various defects and degradations in fuel assemblies, providing a controlled environment for hypothesis testing, operator training, and the evaluation of automated inspection techniques. Unlike real-world data, synthetic data offers the advantage of known ground-truth parameters, allowing for rigorous testing and validation. This approach enables the continuous development of inspection technologies, regardless of hardware availability and operational outages in nuclear facilities. By reducing the reliance on costly real-world experiments, synthetic data offers a scalable and flexible solution for the advancement of nuclear fuel inspection methods. • Scalable and cost-efficient datasets reduce costly hardware dependency, risks. • Anonymized datasets enable broader collaboration across the nuclear industry. • Provides training scenarios to operators without risking sensitive real-world data. • Rigorous algorithm validation enhances visual inspection for deformation, oxidation. • Additional validation supports fuel assembly bowing detection improvements.