MULTIPLE BAND-PASS AUTOREGRESSIVE DEMODULATION FOR ROLLING-ELEMENT BEARING FAULT DIAGNOSIS

This paper presents a novel method to enhance the detection and diagnosis of low-speed rolling-element bearing faults based on discrete wavelet packet analysis (DWPA). The method involves the automatic extraction of wavelet packets containing bearing fault-related features from the discrete wavelet packet analysis representation of machine vibrations. Automated selection of the wavelet packets of interest is achieved via an adaptive network-based fuzzy inference system (ANFIS), which can be implemented on-line. The resultant signal extracted by this technique is essentially an optimal multiple band-pass filter of the high-frequency bearing impact transients. Used in conjunction with the autoregressive (AR) spectrum of the envelope signal, a sensitive diagnosis of the bearing condition can be made. The discrete wavelet packet analysis multiple band-pass filtering of the signal results in a significantly improved signal-to-noise ratio compared to its high-pass counterpart, with an exceptional capacity to exclude contaminating sources of vibration. A more modest increase in the signal-to-noise ratio is achieved when compared to digital band-pass filtering, with the filter range adjusted to obtain the best possible isolation of the bearing transients.