Machine Learning-Based Detection and Analysis of Current Blocking Local Obstacles in REBCO Coated Conductors Obtained from Different Manufacturing Processes

In this study, we successfully developed a method to analyze spatial inhomogeneity in long-length REBCO coated conductors (CCs) manufactured by different processes, by combining reel-to-reel scanning Hall-probe microscopy (RTR-SHPM) with a machine learning-based object detection algorithm. Spatial homogeneity of REBCO CCs is crucial for their practical applications. In this study, we employed RTR-SHPM to visualize two-dimensional (2-D) sheet current density (Js) maps and introduced a machine learning approach to detect local current blocking obstacles in long-length CCs fabricated via different manufacturing processes, including pulsed laser deposition (PLD), reactive co-evaporation by deposition and reaction (RCE-DR), and metal-organic chemical vapor deposition (MOCVD). The object detection model was trained usingJsmaps containing local obstacles from different processes. As a result, the Intersection over Union (IoU) reached 73%, and obstacles were successfully detected with a high F1-score of 89.8%, along with information on their associated manufacturing processes. These results indicate that obstacles features vary depending on the deposition method of the REBCO layers, and our approach shows strong potential to analyze local inhomogeneity independently of the critical current (Ic) value in long-length CCs.