Magnetization Estimation for Permanent Magnet Using Convolutional Neural Network

A new method for the nondestructive estimation of the magnetization distribution (MD) inside a permanent magnet, using a convolutional neural network, is proposed herein. The proposed method improves the estimation accuracy by 15.1% compared to the conventional method that uses a multi-layer perceptron. It achieves high accuracy in estimating the MD. Furthermore, the proposed method accurately predicts the MD from the measured magnetic flux density distribution.