Abstract: A denoising method of block thresholding based on DT-CWT (Dual-tree complex wavelet transform) is proposed, aiming at the difficulty of extracting weak failure information in machinery vibration signals. The presented method can obtain higher signal-to-noise ratio (SNR) than NeighBlock and other common methods based on orthonormal discrete wavelet trans-form, because of improving asymptotic optimality of block thresholding estimator and shift invariance of DT-CWT. It can not only suppress white Gaussian noise, but also remove spike pulse noise in impulsive signal effectively. Experiments indi-cate that the method can extract incipient fault feature in gear-box vibration signals and cryptic information submerged in a heavy noise, and has an excellent effect on extracting weak periodical impulse components to diagnose fault.

Key words: Block thresholding, Denoising, Dual-tree complex wavelet transform, Fault diagnosis