Spline Wavelet based Filtering for Denoising Vibration Signals Generated by Rolling Element Bearings

In this paper, we study biorthogonal spline wavelet decomposition to extract fault features from vibration signals generated by rolling element bearings. Common challenges in the analysis of vibration measurements taken from a real industrial environment is that non-stationary components, generated by other machine components, disturb the analysis. Vibration signals generated by non-faulty and faulty rolling element bearings are studied. As known, the Fourier transformation does not work very well on non-stationary signals because their spectral content changes over time. In the time-frequency domain methods, signal decomposition is performed to split spectrum into sequential sub-spectral components that are processed individually. The weakness of the short-time Fourier transform is that the constant window size does not provide sufficient frequency and time resolution at the same time. Lately, the wavelet transform has been applied on signal demodulation and optimal band-pass filter design. More flexible than basic wavelet basis are spline wavelets that are constructed with a spline function. Spline wavelets are linear combination of B-splines and they can be defined explicitly. Biorthogonal spline wavelets are regular, compactly supported and have finite impulse response implementation. Computer simulated vibration signals and vibration signals acquired from a realworld application are used in our study.