Application of Haar Wavelet De-noising with Cross Correlation and Neighboring Coefficients to the Bearing Faults Prognosis

Rolling bearings are important moving parts of the transmission system in a helicopter. Rolling bearings are prone to failure due to poor working conditions, and the fault diagnosis of these parts is difficult. For rolling bearings, periodic impulses indicate that faults exist in the components. These impulsive features at the early stage are weak and immersed in heavy noise. The features can be distinguished through wavelet de-noising. However, existing wavelet threshold de-noising methods may not extract the feature available due to the limitation in adjacent scale information, and conventional term-by-term thresholding ignores the effect of neighboring coefficients. A new approach for wavelet threshold de-noising is presented in this paper. First, a wavelet transform domain cross-correlation filter is constructed following the theories of wavelet correlation analysis. Then, a novel method for de-noising is proposed, which uses not only the Haar wavelet cross-correlation as the basic estimation to match the impulse but also the neighboring coefficient shrinkage strategy to further enhance the impulsive features and suppress residual noise. Analysis results show that the proposed approach outperforms the soft- and hard-thresholding methods in terms of the combined criterion. It also compares favorably to the other wavelet method for signals with low kurtosis or low signal-to-noise ratio. Bearing fault prognosis results verify the effectiveness of the proposed approach in extracting impulsive features from noisy signals.