Diagnosis of bearing faults using multi fusion signal processing techniques and mutual information

Bearing is a widely used rotating component in most of the industrial machinery. Failure of bearings can incur substantial losses in the industries. During operation, to prohibit failure in bearing, it becomes necessary to identify faults that occur in bearings. In the present work, bearing vibration signals have been taken for the detection of faults in bearings. In the next step, features obtained from various signal processing techniques such as ensemble empirical mode decomposition (EEMD), walsh hadamard transform (WHT) and discrete wavelet transform (DWT) have been used to detect bearing faults (inner race defect, outer race defect, and ball defects). To select the mother wavelet, the maximum energy to entropy ration criteria has been used. Mutual Information feature ranking algorithm is used to select the relevant features. Machine learning techniques such as Random Forest, Support Vector Machine, Artificial Neural Network, and IBK are used. Training and tenfold cross-validation procedures applied to all ranked features. Results reveal that random forest gives 100 % training accuracy with one ranked feature and 98.43 % ten-fold cross-validation accuracy with seven features. From the results, it is observed that the proposed methodology can be reliable and it may serve as an effective tool for fault diagnosis of bearing.