Abstract:

Induction motors are frequently used in the industrial drive systems due to their efficiency, robustness, power and size diversity. Diagnosis is becoming increasingly important as broken rotor bar failures in induction motors directly affect the efficiency of the system. Both the stator current signal and the motor vibration signal are used to diagnose broken rotor bar faults. In recent studies on this subject, machine-learning methods are used together with some signal processing techniques. In this study, determination of the number of broken rotor bars was performed using empirical mode decomposition (EMD) and machine learning methods. Firstly, a phase current and vibration signal taken from the faulty motor data set are filtered and enveloped. In the second step, these signals were decomposed into five intrinsic mode functions (IMF) using by the EMD method, and their spectral entropy and instantaneous frequency features were obtained. In the third step, these features are added end-to-end and a new feature vector is created. In the last step, feature vectors are classified by support vector machine (SVM), k nearest neighbor (KNN) and decision tree (DT) machine learning methods. Classification accuracy was used as the success parameter and the highest success in classification was obtained with SVM, with a classification accuracy of 93.9%. Final of the study, performance comparisons were made with the studies conducted for the same data set in the literature. As a result, it has been seen that the classification of the number of broken rotor bars can be done successfully with EMD and SVM.

Keywords: