Roller bearing fault diagnosis based on wavelet packet decomposition and support vector machine

Motor-driven machines such as water pump, air compressor, and fans are often subject to fatigue failures after long operating hours and cause catastrophic breakdown. These failures are foreshadowed by incipient faults under which the machine is still functioning but with low efficiency. Most frequent failures of the motor-driven machines are the consequence of rolling bearing faults which could be detected by their noise and vibration in real time. The incipient faults, however, are difficult to identify because of their low signal-to-noise ratio, vulnerability to external disturbances, and non-stationarity. Because Fourier spectrum is insufficient to deal with transient and non-stationary signals, a novel approach based on wavelet packet decomposition and support vector machine was proposed to distinguish multiple types of bearing faults. By using wavelet and statistical method to extract bearing fault features, the proposed fault diagnosis method could achieve the maximum accuracy while reducing false alarms.