Fault classification in gears using support vector machines (SVMs) and signal processing

This study presents a procedure for gear fault identification based on vibration signal processing techniques and support vector machines (SVMs). The required feature vector is extracted from vibration signals by time, frequency and time-frequency analysis. A feature selection technique based on Euclidian distance is utilized and five salient features are selected from the original feature set. These features are fed into the classification algorithm. Gear conditions considered were healthy, slightly worn, medium worn and broken-teeth gears. The output of classifier algorithm indicates the status of the gearbox by four labels. The results show that the developed SVM-based procedure is able to discriminate the faults clearly. The effectiveness of the feature selection method is demonstrated by experiments.