Fault detection in a ball bearing system using minimum variance cepstrum

The signals that can be obtained from rotating machines convey information on a machine operating condition. For example, if the machine has faults, it generates a signal that is usually composed of pulse signals. This paper addresses the way in which we can find the faults for periodic pulse signals. Specifically, we have an interest in the case that it is embedded in noise. How well we can detect the fault signal in noise directly determines the quality of fault diagnosis of rotating machines. We propose a signal processing method to detect fault signals in noisy environments. The proposed method is 'minimum variance cepstrum' because it minimizes the variance of the signal power in its cepstrum representation. To test the performance of this technique, various experiments have been performed for ball bearing elements that have man-made faults. Results show that the proposed technique is quite powerful in the detection of faults in noisy environments. In other words, it is possible to detect faults earlier than with conventional methods (McFadden and Smith 1984 J. Sound Vib. 96 69–82, Ho and Randall 1999 6th Int. Congress on Sound and Vibration pp 2943–50, Lee and White 1998 J. Sound Vib. 217 485–505, Kim et al 1991 Mech. Syst. Signal Process. 5 461–73, Staszewski and Tomlinson 1997 Mech. Syst. Signal Process. 11 331–50).