Experimental investigation of cylindrical bearing fault diagnosis with SVM

Abstract

Abstract Cylindrical roller bearing is a basic principal component of so many rotating machineries. Continuous rotation of the various parts may lead to failure and identification of that in an early stage can prevent failure. Some of the Soft computing techniques like Classifier algorithms or neural networks are always helpful in finding out the various defects occurring when bearings are in operation. Here in this research, Wavelet packet decomposition with the use of mother wavelet ‘sym2′is used for finding out the most useful features for auto defining the condition of bearing. Experimental data are taken for the cylindrical bearing NJ305 and four different conditions like inner race defect, outer race defect, roller defect, and healthy bearing are taken for the consideration. Features like kurtosis, crest factor, energy, skewness, etc. are calculated for all the database of 900 signals. For auto-identification of the condition, initially, ANN is trained and tested, and after that various classifiers are also analyzed here for finding out the best method. It was observed that the Support Vector Machine technique as classifier stands the best among all with almost 96% efficiency.