Vibration Analysis of Heterogeneous Gearbox Faults using EMD Features and SVM Classifier

Abstract

Gearbox is one of the important mechanical power transmission device most commonly used in automobiles and industries to get the desired change in speed and torque. The gearbox fault diagnosis has given utmost importance for its significance in preventing halts of a mechanical system and guaranteeing an advantage of sufficient maintenance. This paper presents the vibration analysis of heterogeneous gearbox faults using EMD features and SVM classifier. The vibration signal is converted into intrinsic mode functions (IMF) with decreasing order of frequencies using empirical mode decomposition (EMD) method. Feature vector consisting of information theoretic features have been computed for each IMF and concatenated to form a feature set. By using random permutations, the feature set has been divided into training and testing sets. The support vector machine (SVM) algorithm has been used as a classification technique to diagnose the gearbox faults, which consists of five-class classification. The accuracy of the developed algorithm has been validated using 100 Monte Carlo runs. A comparative study has been carried between computed features and varying IMF components. The observations made were-clear discrimination of the gearbox faults and improved classification accuracy, which contain-chipped tooth, missing tooth, root fault, surface fault and healthy working state of the gear.