Broken rotor bar diagnosis in induction machines through stationary wavelet packet transform and multiclass wavelet SVM

This paper presents the establishing of intelligent system for broken-rotor-bar (BRB) diagnosis based on a novel combination of both, stationary wavelet packet transform (SWPT) and multiclass wavelet support vector machines (MWSVM). The SWPT is used for feature extraction under lower sampling rate. In fact, it is demonstrated through experimental results that the use of the lower sampling rate does not affect the performance of SWPT to detect BRB, while requiring much less computation and low cost implementation. The multiclass SVM (MSVM) is used to automatically recognize the faults. Different MSVM strategies are compared with various kernel functions in terms of classification accuracy, training and testing complexity. The classification results show that the wavelet kernel function detects the faulty conditions with a higher accuracy.

► Broken rotor bar in induction motors are detected with different load levels.
► Stationary wavelet packet transform under lower sampling rate, 200 Hz, is used for feature extraction.
► Multiclass wavelet SVM is built to perform the faults recognition.
► Different binary multiclass SVM strategies are compared with various kernel functions.