Study on Multi-scale Adaptive Noise Cancellation Method for Gear Early Fault Detection

Adaptive noise cancellation (ANC) is a useful method to improve the signal-to-noise ratioin in early fault diagnosis. However, most of the latest ANC methods try to design a filter that adapts to the entire frequency band, which will weak the performance of certain narrow-band noise components. In order to settle this problem, a new multi-scale adaptive noise cancellation method is proposed in this paper based on the wavelet packet transformation (WPT) and evolutionary digital filter (EDF). In the proposed method, the analyzed signal and the noisy reference signal are firstly decomposed into several sub-bands through WPT with same parameters. Then, the analyzed signal in each sub-band is reconstructed and then adaptively de-noised via the corresponding sub-band noisy signal based on the EDF method, the minimum of the mean square error in each band is set as the target of the optimization. Finally, the de-noised signal is obtained from the sum of de-noised signals of each sub-band. The effectiveness of the proposed method is verified by an early gear crack fault diagnosis experiment in a gearbox. Results show that the proposed method yields a better performance in the cancellation of narrow-band noise components comparing to traditional EDF, the better SNR of the crack fault feature is obtained.