Bearing Fault Diagnosis Based on Wavelet Analysis

The de-noising principle using wavelet was discussed and the de-noising property was compared with that of using general threshold strategy. The steps of the per-level de-noising method were then given and the experimental study with a vibration model was carried out. The results prove that the method is advantageous to de-noising, which is more suitable to recover the interest mutations signal buried in intensive background noise. The application of the method in the bearing vibration was presented and the results show that it can inhibit the background noise effectively and recover the interest information satisfactorily. Rolling bearing is the most widely applied mechanical part in rotating machinery and also one of the most easily damaged components. Its running state often directly affects the overall equipment performance. But the early impact vibration signal of the fault rolling bearing is weaker, the fault signatures are often severely polluted by noise, and extracting noise signatures from the noise interference is the key ([1]) for fault diagnosis of the rolling bearing. When it is failed, the vibration often contains some certain abrupt signals, showing a certain non-stationary. The traditional Fourier transform can only make global transform for the signal, but can not depict the characteristics during specific time quantum or frequency band. Therefore, it is unsuitable for time varying signal analysis and has greater limitations. Wavelet Analysis has been highly concerned as frontier domain by scientific and technological circles both at home and abroad in recent years, and it is a new powerful time-frequency analysis tool. It overcomes the shortcomings ([2]) that the frequency domain analysis does not involve time information and the time domain analysis does not involve frequency domain information. In 1992, Donoho and Johnstone proposed the Wavelet Threshold Denoising Method and also gave the threshold value formula, which was applied to a variety of noise reductions and achieved great success. But this general threshold has serious "over-shrinked" tendency for wavelet coefficients, especially the higher frequency part in the useful signal is often treated as noise for removal ([3-4]). To solve this problem, this article proposes the Gradient Threshold Denoising Method based on the autocorrelation analysis of wavelet detail coefficient, and describes the determination of the wavelet decomposition layers and conducts experimental study. Finally, it applies this method to conduct denoising analysis on the engine bearing signal, and proves the validity of Wavelet Gradient Threshold Denoising Method in preprocessing of bearing fault diagnosis based on the analysis of the vibration signal.