Detection of Turn-to-Turn Short Circuit Fault of Permanent Magnet Synchronous Motor Based on Wavelet Packet Transform and KNN

A fault feature extraction and fault diagnosis method based on wavelet packet transform and KNN classification algorithm is proposed for the problems of difficult diagnosis, hard detection, and inaccurate feature extraction in permanent magnet synchronous motor. The method decomposes the torque and current signals collected under the same fault level of a turn-to-turn short circuit by different wavelet bases and decomposition layers. The decomposed coefficients of each layer are used to calculate the energy value as the original features. A new feature extraction and optimization method (I_w method) is proposed to handle the energy-valued features of both signals. The KNN algorithm uses the extracted features to classify and diagnose turn-to-turn short circuit faults. Our experiments show that the accuracy of our method is 100% for turn-to-turn short circuit fault detection, which is improved compared with other methods, and the time consumption is shorter.