USE OF MACHINE LEARNING FOR CURRENT DENSITY DISTRIBUTION ESTIMATION OF REBCO COATED CONDUCTORS

UDC: [621.315.2.014.1:537.63.08]:004.85.032.26

Abstract

Rare-earth barium copper oxide (REBCO) coated conductors (CC) have the following advantages: high current density under high magnetic fields and mechanical strength. The REBCO CCs are expected to be applied for high field applications. However, the REBCO CCs have inhomogeneous critical current density due to local cracks or voids in the REBCO layer. It is known that these local defects might cause the normal-state transition, which can lead to significant damage to the magnet. It is necessary to perform non-destructive measurement of the REBCO conductor properties. A Hall Probe Measurement (HPM) is a non-destructive inspection method for critical current density distribution. In the HPM, a magnetic field is applied to the REBCO CC and the field induced by the shielding current is measured. Subsequently, the current density distribution is estimated from the measured field using inverse analysis, in common, based on the Biot-Savart Law or the Fourier transform. However, the resolution of the current density distribution by these methods is coarse and sometimes insufficient to detect the micro-scale defects. In this research, we have built a method to estimate the critical current distribution using a convolutional neural network (CNN). The CNN model is trained using a set of distributions of the critical current density and the field. The estimation accuracy of the CNN model is investigated, and we compare the estimation results with those using conventional methods.