Abstract
Motor-driven machines, such as water pumps, air compressors, and fans, are prone to fatigue failures after long operating hours, resulting in catastrophic breakdown. The failures are preceded by faults under which the machines continue to function, but with low efficiency. Most failures that occur frequently in the motor-driven machines are caused by rolling bearing faults, which could be detected by the noise and vibrations during operation. The incipient faults, however, are difficult to identify because of their low signal-to-noise ratio, vulnerability to external disturbances, and non-stationarity. The conventional Fourier spectrum is insufficient for analyzing the transient and non-stationary signals generated by these faults, and hence a novel approach based on wavelet packet decomposition and support vector machine is proposed to distinguish between various types of bearing faults. By using wavelet and statistical methods to extract the features of bearing faults based on time-frequency analysis, the proposed fault diagnosis procedure could identify ball bearing faults successfully.