One-Dimensional Convolutional Neural Network for Detecting Internal Defects of Arc Magnets

Arc magnet is an essential part of a permanent magnet motor. It is a critical task to detect the internal defects of the arc magnet before installation and use. At present, this task is mainly completed by experienced workers in the industry, exposing the drawbacks of low efficiency and high cost. This paper proposes a convolutional neural network (CNN) detection method that combines batch normalization (BN) and global average pooling (GAP) in response to the detection requirements of the internal defects of the arc magnet. The model uses BN processing on the input data of the convolutional layer to speed up the network convergence speed while preventing overfitting and increasing the generalization ability of the model. GAP replaces the fully connected layer in the traditional convolution model, significantly reducing model calculation parameters by reducing feature values. Through the training of the original time-domain signal, the model can complete the intelligent diagnosis of the internal defects of the arc magnet with its powerful ability to extract features automatically. The experimental results show that the proposed method has high recognition accuracy for different types of arc magnets.