Machine Learning-Based Fault Diagnosis for Single- and Multi-Faults in Induction Motors Using Measured Stator Currents and Vibration Signals

In this paper, a practical machine learning-based fault diagnosis method is proposed for induction motors using experimental data. Various single- and multi-electrical and/or mechanical faults are applied to two identical induction motors in lab experiments. Stator currents and vibration signals of the motors are measured simultaneously during experiments and are used in developing the fault diagnosis method. Two signal processing techniques, matching pursuit, and discrete wavelet transform, are chosen for feature extraction. Three classification algorithms, support vector machine (SVM), K-nearest neighbors (KNN), and ensemble, with 17 different classifiers offered in MATLAB Classification Learner toolbox are used in the study to evaluate the performance and suitability of different classifiers for induction motor fault diagnosis. It is found that five classifiers (fine Gaussian SVM, fine KNN, weighted KNN, bagged trees, and subspace KNN) can provide near 100% classification accuracy for all faults applied to each motor, but the remaining 12 classifiers do not perform well. A novel curve fitting technique is developed to calculate features for the motors that stator currents or vibration signals under certain loadings are not tested for a particular fault. The proposed fault diagnosis method can accurately detect single- or multi-electrical and mechanical faults in induction motors.