Abstract:

In order to extract incipient features caused by bladed machinery in the presence of coherent noises, a novel diagnostic approach using sparse demodulation operator is proposed. First, the recorded vibration signal from the bladed machinery is decomposed by the centralized multiresolution analyzing method and each subspace is reconstructed in the time domain. Second, the Hilbert demodulation method is performed on the reconstructed signals and some specific subspaces within which the harmonic tones of fault frequencies are dominant are selected. Third, comb filters are employed to separate the harmonic tones of fault frequencies such that a referential model for the fault features can be obtained. Finally, the reconstructed signals of the selected subspaces are denoised, via the wavelet threshold strategy combined with the referential model, to retrieve fault induced incipient features. The proposed method is applied to a fault diagnosis case study of a booster fan with blade cracks. It is found that the periodic impulsive features cannot be directly extracted in the time domain by merely using multiscale decomposition. However, with the proposed method, the actual fault features can be significantly enhanced after suppressing noises of strong coherence.

Key words: vibration measurement, bladed machinery, sparse representation, centralized multiresolution