Prediction of Unknown Fault of Induction Motor using SVM following Decision-Directed Acyclic Graph

Abstract

Prediction of unknown fault of induction motor is an important task to prevent it from unscheduled shutdown. Here, an unknown fault of induction motor has been classified and authenticated from other type of faults using multiclass support vector machine (SVM) following decision-directed acyclic graph (DDAG). Three phase current data samples are collected from induction motors with different known type of fault conditions and one induction motor with unknown type fault condition. Experiment has been performed for motor fault current signature analysis (MFCSA) to identify the type of unknown fault among the mixture of various type of faults. A feature extraction and dimensions reduction process called principal component analysis (PCA) is used to extract the information from fault current signature of each faulty motor and two eigenvalues of stator currents which are called principal components are effective fault features of the motors are captured with the help of PCA transformation. One vs one (OVO) SVM algorithm is applied to separate each pair of classes out of six classes nonlinearly by RBF type kernel assigning the unknown test sample to the class. The multiple PC values of each phase of each faulty induction motor are considered as one class. The OVO-SVM constructs n (n-1)/2 no of classifiers for n class problem of each phase and DDAG technique is used to create a directed acyclic graph using the classifiers to take accurate decision about the classification of the unknown fault. The unknown fault is classified for each phase among different type of faults depending on maximum membership count generated by classifiers and the fault is also authenticated from the results of DDAG of three phases.