Helium Bubble Evolution In Nano-Infiltrated Transient Eutectic SiC Utilizing Machine Learning Analysis

The nuclear energy industry is searching for materials that can be used in new reactor designs, such as fusion energy systems, and withstand the volatile conditions. Silicon carbide (SiC) has emerged as a versatile material that is already used in fission reactors and could be applied to the fusion reactor design due to its significant strength, thermal stability, and resistance to radiation damage. Characterization of materials results in massive amounts of data due to the ever-improving microscopy technology. Micrographs taken of samples implanted with helium bubbles can have thousands of microscopic features that need to be annotated. This large amount of data and human error are a limiting factor on data analysis in the nuclear materials science field. Many machine learning (ML) model approaches have been explored to assist human researchers process visual data faster than manual annotation. Models such as You Only Look Once (YOLO) have been used to identify features such as black dots, cavities, and grains. With the purpose of addressing the growing need for radiation stable materials, a comparison of two production methods of SiC, chemical vapor deposition (CVD) and nano-infiltration transient eutectic (NITE), under helium implantation at three different fluences (1 x 10¹⁴, 1 x 10¹⁵, and 1 x 10¹⁶ ions/cm²) was conducted. In addition, a ML model was made to evaluate the bubble size and density of the data. Nanoindentation was also performed to evaluate the mechanical stability of the materials. The insight gained into microstructural and nanomechanical properties during various helium irradiations helps to elucidate the structure-property relationship of NITE SiC during potential future fusion heat blanket applications.