Multiscale envelope manifold for enhanced fault diagnosis of rotating machines

The wavelet transform has been widely used in the field of machinery fault diagnosis for its good property of band-pass filtering. However, the filtered signal still faces the contamination of in-band noise. This paper focuses on wavelet enveloping, and proposes a new method, called multiscale envelope manifold (MEM), to extract the envelope information of fault impacts with in-band noise suppression. The MEM addresses manifold learning on the wavelet envelopes at multiple scales. Specifically, the proposed method is conducted by three following steps. First, the continuous wavelet transform (CWT) with complex Morlet wavelet base is introduced to obtain the wavelet envelopes at all scales. Second, the wavelet envelopes are restricted in one or more narrow scale bands to simply include the envelope information of fault impacts. The scale band is determined through a smoothness index-based (SI-based) selection method by considering the impulsiveness inside the power spectrum. Third, the manifold learning algorithm is conducted on the wavelet envelopes at selected scales to extract the intrinsic envelope manifold of fault-related impulses. The MEM combines the envelope information at multiple scales in a nonlinear approach, and may thus preserve the factual envelope structure of machinery fault. Simulation studies and experimental verifications confirm that the new method is effective for enhanced fault diagnosis of rotating machines.