Stator Winding Faults Monitoring using Advanced Classification Algorithm

Abstract

Objective: The classical support vector machine is not appropriate to classify large scale datasets. In this context, this paper presents a novel method utilizing large-scale active support vector machine (LASVM) to identify stator related defects of three-phase induction motor.

Methodology: In this work, an experimental study has been performed to record the IM parameters as stator currents, and phase voltages at different health conditions of stator winding such as healthy stator (HS), voltage unbalance fault (VUF), open winding fault (OWF), and stator inter-turn short circuit fault (SISCF). Voltage unbalance index (VUI) parameter has been calculated using standard formula and utilized along with stator currents as features to train and validate the results using proposed method LASVM technique to identify stator defects. A comprehensive performance has been evaluated against other state-of-the-art sequential learning algorithms including online sequential extreme learning machine (OS-ELM), Least Square Support Vector Machine (LS-SVM) and Budgeted Stochastic Gradient Descent Support Vector Machine (BSGD) using popular time series regression and classification benchmarks.

Results: Various performance parameters such as accuracy (ACC), sensitivity (SE), specificity (SP), and positive prediction value (PPV) have been determined to evaluate the effectiveness of proposed method. The simulated results have depicted that this method achieved classification ACC, 98.11%, 98.45%, and 96.35% for datasets HS-VUF, HS-OWF and HS-SISCF for radial basis function (RBF) kernel. The performance of this method was also found better in comparison to BPNN, online sequential extreme learning machine (OS-ELM), support vector machine (SVM), Budgeted Stochastic Gradient Descent Support Vector Machine (BSGD-SVM), and least square support vector machine (LS-SVM) methods.