PNNL

Abstract

LiAlO2 is an important material that is used as a tritium producer for the Tritium Readiness Program. While LiAlO2 pellets have been employed in tritium-producing burnable absorber rods for years, tritium release from the material has not been well-characterized, nor has the role of microstructural features in that release. A full understanding of changes to the pellet microstructure as a result of irradiation is necessary to produce an integrated performance model to predict in-reactor behavior as well as target strategic experiments. This project aims at improving a predictive Deep Learning algorithm for the analysis of various microstructural features (grain boundaries, voids, precipitates, etc.) that are visualized by scanning electron microscopy of LiAlO2 pellets before and after irradiation. We use a Deep Convolutional Neural Network to obtain pixel-level classification for microstructural features, being one of either grain, grain boundary, void, precipitate, or zirconia artifact. Given classification results, we calculate statistical summaries based on standard aggregation and spatial point-process methodology to describe the material as a whole and quantify the performance of the classification. Our results show improved performance over a heuristic approach. Also, the computational efficiency of the methodology allows for characterization of many more SEM images than were previously possible.