Performance Analysis of Support Vector Machine and Wavelet Packet Transform Based Fault Diagnostics of Induction Motor at Various Operating Conditions

This paper analyzes the performance of wavelet packet transform (WPT) and support vector machine (SVM) based fault diagnostics of induction motors (IMs) at various operating conditions. Four mechanical faults (namely, bearing fault, bowed rotor, unbalanced rotor, and misaligned rotor) and three electrical faults (namely, stator winding fault, broken rotor bar and phase unbalance) are considered for the diagnosis. In addition, two levels of severity of stator winding fault and phase unbalance are also considered. In order to develop the present fault diagnostics, firstly the vibration and current signals acquired from laboratory experiments are decomposed by the WPT via Haar wavelet. A number of useful wavelet features are then extracted from the decomposed signals of different IM faults. For estimating the correct fault type, the one-versus-one multiclass method of the SVM is finally applied by inputting the most suitable features. Here the most suitable features are chosen using the wrapper model of feature selection. The diagnostics is executed and checked for various operational conditions (i.e., the load and the speed) of IM to test the robustness of developed diagnostics. This work is of practical significance as training or testing data are not always available at all motor operational conditions.