Rolling Bearing Fault Diagnosis via ConceFT-Based Time-Frequency Reconfiguration Order Spectrum Analysis

Bearing vibration signals under nonstationary conditions exhibits time-varying instantaneous frequency (IF) feature, resulting in difficulty in fault characteristic frequency identification. Order tracking (OT) is one of the most prevalent techniques to remove the influence of speed fluctuation. However, it produces different spectrums guided by different IFs, which hinders the fault diagnosis. Generalized demodulation (GD) is another effective method newly proposed to process nonstationary signals. Nevertheless, in the demodulation operation, all phase functions of target frequency components must be required. In fact, it is hard to detect one consistent IF to guide OT, and even harder to estimate all IFs to facilitate GD without tachometer. As such, a novel method is proposed, which can pinpoint all frequency components of interest guided by only one IF, and the result spectrum does not change with the variation of references, i.e., arbitrary IF extracted from the signal can guide the reconfiguration. First, the Hilbert transform is applied to bearing signal to highlight the impulsive components and obtain the envelope signal. Second, the Chirplet path pursuit is adopted to extract one IF from the envelope. Then, the concentration of frequency and time (ConceFT) algorithm is exploited to generate the time–frequency representation (TFR) with sharp time–frequency ridges. Next, the ConceFT-based TFR is reconfigured guided by the extracted IF. Finally, the reconfigured TFR is mapped to the 2-D representation, yielding time–frequency reconfiguration order spectrum. The performance is validated by both simulated and experimental data.