Denoising vibration signals of rotating machines using probability density function, similarity measure and improved thresholding function

In this paper, a new method for removing the noise from the vibration signals acquired from the rotating machinery for its condition monitoring is presented. Firstly, each signal is decomposed into its modes using the empirical mode decomposition method. Then for distinguishing the noisy modes from the noise-free modes, the similarity measure between the probability density function of the raw signal and its modes is calculated. Finally, the noise-dominate modes are denoised by the improved thresholding function, and the denoised signal is reconstructed. In this study, the proposed method is implemented for denoising the simulated signal and real data corresponding to different bearing conditions. Finally, the kurtosis and the envelope spectrum of the denoised signal are calculated for detecting the fault presence and its type. The results show that the proposed technique can improve the quality of the reconstructed signals so that the sensitivity of the kurtosis factor to the presence of the defect in the inner and outer rings is increased. Also, the defects frequencies appear in the spectrum of the signals denoised, and the fault type can easily be detected. The results indicate that the proposed denoising technique is superior to the conventional empirical mode decomposition-based denoising method.