Estimation Method for Magnetization Distribution in Permanent Magnet Using Deep Neural Network

A new estimation method for magnetization distribution (MD) in permanent magnets using a deep neural network (DNN) is proposed. It estimates the physical MD from the measured magnetic flux density distribution. The proposed method overcomes problems of indefiniteness and using a new method of constructing training data that models the realistic MD. The DNN trained on the data calculated by the Biot-Savart method can accurately estimate the MD for measured and untrained data.