De-noising mechanical signals by hybrid thresholding

This paper presents a hybrid wavelet thresholding approach for reducing white Gaussian noise in mechanical fault signals to offset the deficiencies of hard and soft thresholding. We observed that it is not appropriate to use the mean squared error (MSE) as the only criterion in the evaluation of the de-noising results of mechanical signals. As such, we proposed a combined criterion incorporating both MSE and false identification energy (E_false) to evaluate the de-noising results. In our simulation studies, the proposed hybrid thresholding approach outperforms both the soft- and hard-thresholding methods in terms of the combined criterion. The proposed approach is then successfully applied to noise reduction and fault feature extraction of bearing signals