Rolling element bearing fault diagnosis using supervised learning methods- artificial neural network and discriminant classifier

Bearings are the principal component in the induction motor responsible for 50–60% of faults in an induction motor. Hence, detecting and diagnosing bearing faults in an induction motor is essential for reliable operation. Some soft computing techniques like artificial intelligence-based classifiers are always useful in fault diagnosis. This research diagnoses the bearing fault under three vibration signal conditions: raw vibration signal, filtered vibration signal, and wavelet-based denoised vibration signal. The statistical features such as RMS, kurtosis, standard deviation, variance, etc., are extracted from each condition. The db2 wavelet is selected based on the minimum Shannon entropy criteria for the wavelet denoising. Vibration signal data is collected from the experimental setup for four bearing conditions: healthy, outer race defect, ball defect, and cage defect. Total 1600 samples are collected from 2,000,000 data points for each condition. An artificial neural network and discriminant classifier are trained and tested for fault identification. Two other classifiers from each pedigree, i.e., support vector machine and radial basis function neural network, are also analyzed to compare the classification performance. It is observed that the ANN classifier stands the best among all, with a classification accuracy of 99.58% and a minimum computational time of 1.62 s.