Machine Learning Based Fault Diagnosis for Single-and Multi-Faults for Induction Motors Fed by Variable Frequency Drives

Abstract

In this paper, an effective machine learning based fault diagnosis method is developed for induction motors fed by variable frequency drives (VFDs). Two identical 0.25 HP induction motors under healthy, single-and multi-fault conditions were tested in the lab with different VFD output frequencies and motor loadings. The stator current and the vibration signals of the motors were recorded simultaneously under steady-state for each test, and both signals are evaluated for their suitability for fault diagnosis. The signal processing technique, Discrete Wavelet Transform (DWT), is chosen in this paper to extract features for machine learning. Four families of machine learning algorithms in the MATLAB Classification Learner Toolbox, decision trees, support vector machines (SVM), k-nearest neighbors (KNN), and ensemble, with twenty classifiers are evaluated for their classification accuracy when used for fault diagnosis of induction motors fed by VFDs. To allow fault diagnosis for untested motor operating conditions, the feature calculation formulas are developed through surface fitting using experimental data of a range of tested frequencies and loadings of the motor in order to provide training data for untested conditions.