AVS 70 Session AC-ThP: Actinides and Rare Earths Poster Session

A novel approach for identifying star shapes in microscopic images via deep learning segmentation and classification.

Semi-automated data tagging: Introducing a new method for labeling data.

U-Net FCN model: Designed for segmenting various star-like shapes in single or multi-class scenarios.

Identifying shapes under a microscope involves separating the sampling paper into ~10,000 images, akin to finding a needle in a haystack due to its grueling nature and human eye limitations.

Star identification has evolved from manual eye scans to recent advancements in automatic

image processing tools.

This work employs a deep learning model, specifically the U-Net network, for segmentation and classification, integrating a 5-fold cross-validation analysis.

Work Content:

• A New image star database was created.
• Characterization of star types, model architecture, and optimization.
• Training for single and multi-class datasets.
• Automation of segmentation for batches of images.
• Research on optimizing frequency, epochs, and threshold settings,
• collaborating on simulated star classification.

Model achieved 92.04% accuracy for small single-class stars (<60µm), with 0.84 ROC area.

Segmentation of dual-class stars reached 86.3% validation accuracy.

Identification of simulated stars with varying leaf counts achieved 82.63% accuracy, while a computational model for higher magnitude stars achieved 0.90 ROC area.

Plutonium metal forms a passivated oxide layer upon exposure to air, with varying stoichiometry depending on oxygen conditions. At lower temperatures, interfaces in oxides can facilitate corrosion via fast-transport pathways. This can result in further growth of the oxide layers, or of corrosion via other species, such as hydriding from the absorption and transport of hydrogen throughout the material. Any of these forms of corrosion can have implications on the stability and longevity of material in storage. Structural, defect, and transport properties can vary across different grain boundaries and heterointerface orientations. This study presents structures and associated energies of selected Pu₂O₃-PuO₂ interfaces and selected grain boundaries, calculated with molecular statics, along with associated defect structures and formation energies. Comparisons of interfacial energies can suggest which interfaces might be more prevalent in these materials, which could then play a larger role in influencing material properties and behavior. Comparisons of defect energies, both across different interfaces and between interfaces and in bulk, can suggest influence of interfaces in corrosion and in transport phenomena more broadly.

The Cooper-Rushton-Grimes interatomic potential will be utilized for the molecular statics calculations as it has demonstrated comparable values to experimental lattice and elastic constants, and thermal properties [1]. Where applicable, the modification to this potential to include Pu³⁺ is also included, as developed by Takoukam-Takoundjou, et al [2].

[1] M W D Cooper et al 2014 J. Phys.: Condens. Matter 26 105401.

[2] C Takoukam-Takoundjou et al 2020 J. Phys.: Condens. Matter 32 505702.

Understanding the influence of alpha irradiation and dissolved hydrogen on the corrosion behavior of uranium dioxide (UO₂) is essential for evaluating the long-term impacts on storage environments.[1] Traditional experiments involving bulk SNF are typically costly due to stringent requirements for radiation shielding. To address these challenges, we have developed a novel particle-attached microfluidic electrochemical cell (PAMEC). This innovative microfluidic technique enables the multimodal analysis of UO₂ corrosion under simulated SNF storage conditions using minimal material quantities, significantly reducing both costs and risks of hazardous exposure.[2, 3]

Recently, we have incorporated palladium—a noble metal commonly found in irradiated nuclear fuel—into the UO₂ working electrode. This modification allows us to investigate how noble metals influence UO₂ degradation behavior in the presence of dissolved H₂ and alpha irradiation. In addition to utilizing electrochemical measurements such as open circuit potential, for qualitative insights into the degradation of the UO₂ matrix, PAMEC's unique design supports in situ chemical imaging of the UO₂ surface. In this study we employ in situ imaging of the PAMEC working electrode using scanning electron microscopy (SEM) combined with energy-dispersive X-ray spectroscopy (EDS). This integration of techniques offers a comprehensive view that enhances our understanding of the UO₂ corrosion mechanism. The potential applications of this technology are extensive, providing a safer and more effective alternative for conducting corrosion studies on materials that pose high exposure risks or are difficult to access due to their rarity.

References:

1. Long-term storage of spent nuclear fuel. Nature Materials, 2015. 14(3): p. 252-257.
2. A microfluidic electrochemical cell for studying the corrosion of uranium dioxide (UO(2)). RSC Adv, 2022. 12(30): p. 19350-19358.
3. Advancing radioactive material research method: the development of a novel in situ particle-attached microfluidic electrochemical cell. Frontiers in Nuclear Engineering, 2023. 2.