Fault Detection of Permanent Magnet Synchronous Motor Based on SVD-RF

Abstract

In the traditional motor fault detection model, the algorithm has a single selection of classification features, weak generalization ability, and is prone to overfitting, under-learning and other shortcomings. This paper proposes a motor fault detection model combines the random forest and the improved singular value decomposition method: based on the characteristic frequency feature data of the motor fault, the singular value decomposition method is used to extract the maximum change axis and the smaller change axis. The optimal feature subspace affecting the performance of random forest classification detection is obtained by extracting features from these two parts through a stratified sampling method, and then selecting features and split points in this feature subspace to generate a decision tree to construct a random forest fault detection model. Finally, the fault detection model is used to predict the state of the motor. The experimental test shows that the fault detection method is more responsive and accurate than other commonly used fault detection models.