Time-Frequency Manifold for Machinery Fault Diagnosis

Abstract

In this chapter a new method called time-frequency manifold (TFM) is reported for signature enhancement and sparse representation of non-stationary signals for machinery fault diagnosis. In the framework of the TFM analysis, the phase space reconstruction is firstly employed to reconstruct the dynamic manifold embedded in an analysed signal, then the time-frequency distributions (TFDs) are generated in the reconstructed phase space to represent the non-stationary information, and manifold learning is finally addressed on the TFDs to discover intrinsic TFM structure. In this process, the TFM combines non-stationary information and nonlinear information simultaneously. This will provide a better time-frequency signature with the merits of noise suppression and resolution enhancement for machine health diagnosis. Furthermore, a TFM synthesis approach is further reported to explicitly recover the transient signal from the TFM signature by combining the sparse theory with the TFM structure. The objective of the introduced work is to exploit a TFM technology for enhancing the time-frequency signature and representing the transient feature with in-band noise suppression for machine fault signature analysis and transient feature extraction.