Multi-class Fault Diagnosis in Gears Using Machine Learning Algorithms Based on Time Domain Data
Keywords: Vibration signature, Support vector machines, Gear faults, Fault diagnostics
AbstractThe support vector machine, as a powerful machine learning algorithm, is recognized to have good generalization ability in its application to multi class machine fault classification problems. In this paper, an application of the SVM in the multi class gear fault diagnosis has been performed based on the gear vibration data in time domain. From time domain data the statistical features are extracted and fed to the SVM for the training/testing. When the training and testing data are at the same running speed, it is found that the SVM classifier has excellent multiclass classification accuracy. Though, this approach relies on the availability of both the training and testing data at that particular speed of the machine operation; it is expected that features of time domain data would change with operating speeds. Moreover, the training data may not always be available continuously at all operating speeds of gears, especially for variable drive systems. Thus, a novel technique of interpolation/extrapolation has been proposed in the present work that helps the SVM classifier to carryout multi class gear fault diagnosis with appreciable accuracy even in the absence of the training data at a given running speed. It also investigates the speed bandwidth of the training data for which interpolation/extrapolation predictions are reasonably good. The similar analysis is also extended to artificial neural networks (ANN) as the classifier and predictions are noted, and compared with the SVM.
Number of References cited: 21