A novel demodulation method with an improved dual-tree complex wavelet packet transform and a robust indicator for bearing fault diagnosis

To attain effective fault diagnosis and weak feature extraction in bearings, a preliminary selection process of the informative frequency band (IFB) is typically conducted. This paper introduces a novel automatic resonance demodulation method to tackle the pivotal challenges of frequency band leakage and high sensitivity of IFB selection indicators to high-impulse noise. Founded upon an improved dual-tree complex wavelet packet transform, the proposed method develops an innovative band division strategy. Through the sub-bands reconstructed and rearranged, the problem of frequency band disorder is resolved, providing full time-frequency resolution while preserving the benefits of approximate shift-invariance and frequency aliasing diminution. The utmost priority lies in enabling the direct capture of IFB in all divided frequency band regions without leakage of frequency bands. Afterward, the autocorrelation correntropy norm, as an IFB selection indicator for automatically determining the boundary range of the fault bands, demonstrates robust anti-strong impulse interference property, exceptional stability, and wide applicability in fault band extraction. The improvements and robustness of this method are corroborated through comparative analyses with three developed diagnostic methods using a simulated signal and two sets of actual data.