Multi-domain Bearing Fault Diagnosis using Support Vector Machine

Faults in rolling element bearings are the main cause of rotating machine failure. Locating and isolating the faults has become a critical concern for the stable operation of rotating machinery. This paper puts forward a methodology to derive optimum fault indicators from vibration signature and to make a robust model using Support Vector Machine (SVM) for bearing fault diagnosis. The vibration signatures are collected at three speeds at a constant loading condition. Thereafter, optimal statistical features are extracted in the time, frequency, and time-frequency domain. The proposed technique involves the performance comparison of SVM models trained with optimal features. Results show that the multi-domain time-frequency features gave a better performance as compared to the individual domain signals.