Multiclass Fault Taxonomy in Rolling Bearings at Interpolated and Extrapolated Speeds Based on Time Domain Vibration Data by SVM Algorithms

The multiclass fault taxonomy of rolling bearings based on vibrations through the support vector machine (SVM) learning technique has been presented in this paper. The main focus of this article is the prediction and taxonomy of bearing faults at the angular speed of measurement as well as innovatively at the interpolated and extrapolated angular speeds. Five different bearing fault conditions, i.e., the inner race fault, outer race fault, bearing element fault, combination of all faults, and a healthy bearing have been considered. Three different statistical feature parameters, namely, the standard deviation, the skewness, and the kurtosis have been obtained from time domain vibration data for bearing fault predictions. The Gaussian RBF kernel and one-against-one multiclass fault classification technique has been used for the taxonomy of bearing fault by the SVM. Also the study of the selection of SVM parameters, like gamma (RBF kernel parameter), best datasets, and the best training and testing percentages have been presented in this paper. The present work observes a near perfect prediction accuracy of the SVM prediction performance when the training and testing are done at a higher rotational speed. It shows a better fault prediction accuracy at the same rotational speed than that of measurement as compared to the interpolated and extrapolated rotational speeds. Also the SVM capability of fault taxonomy decreases with increase in the range of interpolation and extrapolation speeds. © 2014, ASM International.