Speaker
Description
REBCO conductors are emerging as a prominent choice among high-temperature superconductors due to their superior superconducting properties at high magnetic fields. However, depending on the fabrication route, there can be microstructural heterogeneities such as porosity, CuO precipitates, and a-axis grains. Such defects at a larger scale can disrupt current transport and limit the critical current density. This study focuses on the investigation and characterization of defects in commercial REBCO conductors of varying characteristics. Top-view images of the REBCO layer are captured post-etching using scanning electron microscopy (SEM). Traditional image analysis and segmentation are limited and inadequate for quantifying such defects in SEM images, while manual identification is labor-intensive and time-consuming. To address this, the application of machine learning (ML) techniques is explored for rapid, automated and accurate defect detection. Specifically, different open-source foundational computer vision models are fine-tuned to analyze the SEM images. The results from the ML models are compared and utilized for statistical analysis with the hope of establishing correlations between these microscale defects and measured in-field transport critical current.
This work was supported in part by the U.S. Department of Energy, Office of Science, Office of High Energy Physics, through the Magnet Development Program, under Contract DE-AC02 05CH11231, which is performed at Lawrence Berkeley National Laboratory.
