Time-frequency Analysis Based on Vector Local Characteristic-scale Decomposition and Fast Kurtogram for Rolling Bearing Fault Diagnosis

A time–frequency analysis method based on vector local characteristic-scale decomposition (VLCD) and fast kurtogram (FK) is proposed for rolling bearing fault diagnosis. Local characteristic-scale decomposition (LCD), an adaptive method for non-stationary and nonlinear signal processing, decomposes vibration signals into a series of intrinsic scale components (ISCs). However, traditional LCD analyzes single-channel signals, which can limit its ability to fully capture the characteristic information of the rotor. Consequently, the analysis results based on signals from different channels in the same section may be inconsistent. To address this issue, VLCD, which employs full vector spectrum technology, was developed to provide a more comprehensive analysis. Still, the existing environmental noise in the raw signal remains in the corresponding ISC with the component of interest. ISCs have good performance in impulse detection with strong environmental noise but are susceptible to non-Gaussian noise. The proposed method combines the merits of VLCD and FK to detect rolling bearing faults. Firstly, by applying LCD, the raw signal is decomposed into a set of ISCs. Then, the ISC, which primarily characterizes fault information, is selected based on its cross-correlation coefficient with the original signal. Additionally, the selected ISCs are aggregated to construct the vector ISC signal. Finally, this signal is further filtered by an optimal band-pass filter based on FK to extract the impulse signal. Faults can be identified based on the appearance of characteristic frequencies in the Hilbert envelope spectrum of the filtered signal. The advantages of VLCD over conventional LCD are demonstrated through rolling bearing case analyses, showcasing its enhanced ability to extract meaningful information from multiple channels.