Internal Defect Identification of Arc Magnets Based on a Deep Residual Network Combined with GRU and SqueezeNet

Arc magnet is the critical component of permanent magnet DC motors that produce a constant magnetic field. Internal defects are the primary problem affecting the quality of arc magnets. However, traditional acoustic detection methods for detecting internal defects of arc magnets usually have low accuracy and low efficiency. This article develops an improved SqueezeNet model that combines deep residual network and GRU network to identify the acoustic signals and realizes the internal defect detection of arc magnets. This method presents a deep residual network to solve the problem of gradient disappearance. The GRU neural network can remember essential features in the past to ensure the stability of the model, and the improved SqueezeNet model is used to reduce the calculation parameters. The accuracy rate for internal defect identification of three types of commonly used arc magnets has reached 100%. The experimental results show that the SqueezeNet-GRU model dramatically reduces the number of parameters and enhances convergence and stability. The proposed model provides better performance than other network algorithms.