Machinery Fault Signal Reconstruction Using Time-Frequency Manifold

Machinery fault signals generally represent as periodic transient impulses, which often associate with important measurement information for machinery fault diagnosis. However, the existence of much background noise in practice makes it difficult to detect the transient impulses. Thus, it is very necessary to de-noise the measured signal and extract the intrinsic machinery fault signal for a reliable fault diagnosis. In this chapter, a novel de-noising method based on the time-frequency manifold (TFM) is proposed. This method mainly includes the following several steps. First, the phase space reconstruction (PSR) is employed to achieve a group of high-dimensional signals. For each dimensional signal, the short-time Fourier transform (STFT) is then conducted. Third, a suitable band carrying fault information is used for learning the TFM. Finally, the TFM is used to reconstruct the fault signal based on time-frequency synthesis and PSR synthesis. As the TFM has the merits of noise suppression and resolution enhancement to represent the inherent time-frequency structure, the reconstructed fault signal also has satisfactory de-noising effect, as well as good effect of inherent transient feature keeping. The proposed method has been employed to deal with a set of bearing data with rolling-element defect and outer-race defect, and the results show that the method is rather superior to two traditional methods in machinery fault signal de-noising.