A combined spectral subtraction and wavelet de-noising method for bearing fault diagnosis

In this paper, the Gabor wavelet is used for wavelet filter based de-noising the vibration signal measured from faulty bearings. In this approach the parameters of the daughter wavelet corresponding to center frequency and bandwidth namely scale and shape-factor should be selected properly. The ratio of the geometric mean to the arithmetic mean of the wavelet coefficient moduli called smoothness index is used as a measure for the selection of these parameters. As bandpass filtering does not eliminate the in-band noise with frequency content on the range covered by the daughter wavelet, spectral subtraction technique is applied prior to wavelet transforming the signal. This has significantly improved the performance of the wavelet filter based de-noising method. Results are presented for both simulated and experimental data.