Advanced characterization-informed machine learning framework and quantitative insight to irradiated annular U-10Zr...

Area of Science:

• Nuclear Engineering
• Materials Science
• Computational Science

Background:

• U-10Zr metal fuel is a leading candidate for next-generation sodium-cooled fast spectrum reactors.
• Existing knowledge on fuel performance is largely at the engineering scale, lacking mechanistic understanding of microstructural evolution during irradiation.
• A gap exists in tools for rapid microstructure assessment and property prediction from post-irradiation examination data.

Purpose of the Study:

• To develop and validate a machine learning-enabled workflow for rapid, quantified assessment of irradiated nuclear fuel microstructure.
• To analyze microstructural evolution and property degradation in prototypical U-10Zr annular metal fuels.
• To reveal the distribution of secondary phases and compositional changes within the fuel.

Main Methods:

• Utilized a machine learning workflow integrated with domain knowledge.
• Employed a large dataset from advanced post-irradiation examination microscopies.
• Applied the workflow to assess microstructure in two reactor-irradiated U-10Zr annular metal fuels.

Main Results:

• Successfully provided rapid and quantified assessments of fuel microstructure.
• Revealed the distribution of zirconium-bearing secondary phases and constitutional redistribution across radial locations.
• Quantified the ratios of seven distinct microstructures along the temperature gradient and compared fission gas pore distributions.

Conclusions:

• The machine learning workflow enables efficient and accurate characterization of irradiated nuclear fuel microstructures.
• This approach addresses the need for advanced tools to understand fuel performance and degradation mechanisms.
• The findings contribute to the development and deployment of advanced nuclear fuel systems.