Multiwavelet denoising with improved neighboring coefficients for application on rolling bearing fault diagnosis

The characteristic signal of a rolling bearing with a defect acts as a series of periodic impulses. These features are usually immersed in heavy noise and then difficult to extract. It is feasible to make the features distinct through wavelet denoising. Scalar wavelet thresholding has been used to extract features. However, scalar wavelet might not extract the feature available due to its limitation in some important properties, and conventional term-by-term thresholding does not consider the effect of neighboring coefficients. Since multiwavelets have been formulated recently and they might offer good properties in signal processing, a novel denoising method — multiwavelet denoising with improved neighboring coefficients (neighboring coefficients dependent on level, DLNeighCoeff for short) — is proposed in this article. The method proposed is applied to a simulated signal and fault diagnosis of locomotive rolling bearings, obtaining performance superior to conventional methods.