Time-frequency manifold for demodulation with application to gearbox fault detection

For rotating machines, the localized faults of key components generally represent as periodic transient impulses in vibrations. The existence of background noise will submerge the transient impulses in practice, and will thus increase the difficulty to identify specific faults. This paper proposes a novel fault demodulation method based on time-frequency manifold (TFM) to solve the aforementioned problem. This method uses the TFM base to demodulate the periodic impulses from raw signals. It mainly includes two following steps: first, the TFM is obtained by addressing manifold learning on the time-frequency distributions (TFD); second, a short TFM is adopted as a template to do correlation analysis with the original TFD. The proposed demodulation method can achieve a high resolution for identifying interesting impulse components. The novel method is verified to be superior to traditional enveloping demodulation method by means of simulation signal analysis and application to gearbox fault detection.