Fault Diagnosis of Ball Bearing with WPT and Supervised Machine Learning Techniques

In this paper, fault classification was done using RBIO 5.5 wavelet. Features were extracted at fifth level of decomposition with wavelet packet transform (WPT) where energy and kurtosis were extracted for both horizontal and vertical responses at all WPT nodes. Thus, total 400 samples were taken of defective bearing with reference to healthy bearing to minimize the experimental error. Multilayer perceptron of ANN with correlation-based feature selection has compared with sequential minimal optimization-based support vector method (SVM). Result shows that ANN with multilayer perceptron with CFS criteria has performed better than SVM for classification of ball bearing condition.