Development of Novel Methodologies for Better Understanding and Manufacturing of REBCO Coated Conductors by Integrating High Throughput Magnetic Microscopy and Machine Learning

Recent advances in manufacturing techniques have made it possible to obtain commercial RE₁Ba₂Cu₃O_7-δ (REBCO, RE: Rare Earth) coated conductor (CC) tapes having a length of several 100 m. Along with this trend, research on the development of CC applied equipment is also being vigorously promoted in various countries of the world. However, in developing REBCO CC based large-size superconducting equipment, unexpected quenching and lowering of critical current (<i>I</i>_c) and yield at coil-forming are noticed as problem. Technology development for obtaining more stable performance with good reproducibility becomes an urgent issue. The author considers that, as the cause, 1) the behavior of the long CC tape under the practical operation condition is not sufficiently understood, and 2) the problem of the complexity of the manufacturing process. In this study, the local nonuniformity of critical current density (<i>J</i>_c) in the tape-plane of the long CC is measured nondestructive and noncontact manner at high spatial resolution using a reel-type high-speed magnetic microscope. Integrating with image analysis by machine-learning (ML), we succeeded in the automated analysis of several thousands of magnetizing current images and clarified the presence of domains of the local nonuniformity in the long tape, which is difficult to be detected by the conventional inspection by local <i>I</i>_c criterion, and obtained detailed information on the<i> J</i>_c lowering domain such as the size, position, and statistical distribution. To improve the process conditions, we also developed a ML regression model which can quantitatively estimate <i>I</i>_c obtained by the manufacturing condition through a coupled analysis by high-speed <i>I</i>_c measuring and ML regression using a combinatorial sample in which the manufacturing condition was systematically changed in the longitudinal direction of the tape. Furthermore, using the Genetic Algorithm on the basis of this ML regression model, we have also succeeded in deriving the combination of control variables, i.e., the production conditions, to maximize the <i>I</i>_c as output as the solution of the inverse problem. These new methodologies enable us to elucidate the reason for local non-uniformity of long CC tapes, to develop CC tapes with better spatial uniformity, to derive process parameters quickly on PC to maximize the tape performance and are expected to greatly contribute to drastic shortening of lead times for CC tape development and realizing stable tape performance.<br/><br/>Acknowledgements: This work was supported by JSPS KAKENHI Grant Number JP19H05617.