A Machine Learning Approach Toward Improving QA/QC of Coated Particle Fuels

Coated particle fuels, such as the tristructural isotropic (TRISO) fuel particle, have been studied and developed for decades for high-temperature gas reactor (HTGR) applications because of their efficiency and relative stability under off-normal conditions. Critical to this development is a strong understanding of the relationship between fuel fabrication, properties and performance, as is how the former can be adjusted and improved to optimize the latter. Accordingly, wide-scale implementation of coated nuclear particle fuels requires thorough and robust quality assurance/quality control (QA/QC) methods for fabrication, characterization, and deployment. Of the layers in a TRISO particle, the silicon carbide (SiC) layer serves as the main structural layer and barrier to release of non-gaseous fission products. The SiC layer microstructure has been shown to be correlated with the release of fission products, and thus microstructural metrics of this layer (i.e., grain size and shape) are an aspect of TRISO particle QA/QC. Previous work has shown significant twinning in the SiC microstructure, which significantly impacts the calculated grain metrics. This work presents progress toward the development of a machine learning-based algorithm for automating the detection of twin grain boundaries from grayscale images of the microstructure, which will allow for the refinement of microstructural QA/QC of TRISO particles. This data-driven approach to improve existing QA/QC capabilities will streamline coated particle fuel qualification at both the lab and industrial scales by facilitating grain boundary detection from easily obtainable, non-crystallographic microstructure images.