Optimized feature selection for condition-based monitoring of cylindrical bearing using wavelet transform and ANN

Abstract

This chapter presents the fault diagnosis methodology based on the advanced signal processing technique along with the data mining method. The wavelet analysis defines the basic mathematical function of the vibration signal by comparing it with the predefined wavelet. After analyzing the signals with the use of the wavelet concept, data acquisition helps in the detection of the bearing defects. This chapter presents the effective selection of the attributes of signal for data mining methods. The training of data in the artificial neural network using different algorithms and with various combinations gives different results. The result for the cylindrical bearing with four conditions shows that out of 12 different features, the best result of 97.6% comes from the combination of the top ten ranked features. The advanced method of ranking the features according to their accuracy in the ANN is adopted, and a combination of top ranked features shows better results in comparison with the conventional methodology. The experimental setup gives the time-frequency data for four bearing conditions such as healthy bearing, bearing with outer race defects, bearing with inner race defect, and bearing with ball defect. The result proves that the proposed method has the advantage of the fault diagnosis of the cylindrical bearing with the most effective feature selection method.