Presentation Information
[WBP1-24]Analysis of Current Blocking Obstacles in REBCO Coated Conductors Based on Reel-to-Reel Magnetic Microscopy and Object Detection
*Zeyu Wu1, Kohei Higashikawa1, Kazutaka Imamura1, Takanobu Kiss1 (1. Kyushu University (Japan))
Keywords:
Scanning Hall-probe Microscopy,REBCO,Spatial inhomogeneity,Object Detection
[Purpose]
We have developed a method to automatically detect local current-blocking obstacles in long-length REBCO coated conductors (CCs) and to analyze their position and domain size. Spatial inhomogeneity is critical for practical applications of CCs, and longitudinal critical current (Ic) uniformity is usually tested prior to shipment by using such methods as TapestarTM and transport measurement by manufacturers. Significant Ic drop can be screened out by these inspections, however, the behavior of localized Ic deterioration at the same level of Ic fluctuation in the tape has not yet fully understood. Such localized Ic deterioration possibly be an origin of instability or limit the coil performance under practical operation conditions at low temperature and high magnetic fields. To overcome this limitation, we demonstrated a reel-to-reel scanning Hall-probe microscope to evaluate the two-dimensional sheet current density (2D J) distribution [1] and analyzed the local obstacles by machine learning-based classification method [2] in our previous study. In this study, we further explore this approach to make it possible to identify the localized Ic deterioration more directly by adopting a machine learning–based object detection.
[Method]
Reel-to-reel scanning Hall-probe microscopy and machine learning–based object detection
[Results]
A 200-m-long pulse laser deposition (PLD)-processed CC was analyzed. The detected obstacles were defined into two categories: isolated-type and cluster-type domains, respectively. The latter being a more complex structure composed of multiple isolated-type obstacles. An object detection model was trained with these two categories. As a result, local obstacles were successfully detected in the 2D J maps, and their position and size distribution were quantitatively evaluated. We compared it with the previous classification approach and found quantitative agreements of its domain size and frequency.
[Consideration]
[Conclusion]
This approach provides detailed insights into local obstacles and holds strong potential to be refined into a robust evaluation indicator for the spatial inhomogeneity of CCs.
We have developed a method to automatically detect local current-blocking obstacles in long-length REBCO coated conductors (CCs) and to analyze their position and domain size. Spatial inhomogeneity is critical for practical applications of CCs, and longitudinal critical current (Ic) uniformity is usually tested prior to shipment by using such methods as TapestarTM and transport measurement by manufacturers. Significant Ic drop can be screened out by these inspections, however, the behavior of localized Ic deterioration at the same level of Ic fluctuation in the tape has not yet fully understood. Such localized Ic deterioration possibly be an origin of instability or limit the coil performance under practical operation conditions at low temperature and high magnetic fields. To overcome this limitation, we demonstrated a reel-to-reel scanning Hall-probe microscope to evaluate the two-dimensional sheet current density (2D J) distribution [1] and analyzed the local obstacles by machine learning-based classification method [2] in our previous study. In this study, we further explore this approach to make it possible to identify the localized Ic deterioration more directly by adopting a machine learning–based object detection.
[Method]
Reel-to-reel scanning Hall-probe microscopy and machine learning–based object detection
[Results]
A 200-m-long pulse laser deposition (PLD)-processed CC was analyzed. The detected obstacles were defined into two categories: isolated-type and cluster-type domains, respectively. The latter being a more complex structure composed of multiple isolated-type obstacles. An object detection model was trained with these two categories. As a result, local obstacles were successfully detected in the 2D J maps, and their position and size distribution were quantitatively evaluated. We compared it with the previous classification approach and found quantitative agreements of its domain size and frequency.
[Consideration]
[Conclusion]
This approach provides detailed insights into local obstacles and holds strong potential to be refined into a robust evaluation indicator for the spatial inhomogeneity of CCs.