Smart multichannel mode extraction for enhanced bearing fault diagnosis

In bearing fault diagnosis, multichannel data can contain more abundant and complete fault information to alleviate the influence of accidental factors in a single channel. To fully employ the fault information concealed in the multichannel data, this paper proposes a smart multichannel mode extraction (SMME) for enhanced bearing fault diagnosis. The SMME method based on multivariate variational mode decomposition (MVMD) and manifold learning overcomes the problems of predefined model parameters in MVMD and shows good performance in mining the intrinsic nonlinear and nonstationary features from multichannel modes of different quality. First, inspired by the convergence property of MVMD, a smart multichannel spectral structure scanner with solid mathematical theory is constructed to adaptively detect the latent center frequencies (CFs) in the multichannel bearing signals without prior knowledge. Second, multichannel single-step decomposition induced by the detected CFs is established to obtain corresponding multichannel modes through only single-step calculation instead of considerable iterations. Third, a fault feature enhancement strategy is designed for locating and fusing the aligned multichannel sensitive modes with different qualities of fault information to highlight the inherent fault features. The superiority of the SMME method for enhanced bearing fault diagnosis in effectiveness and efficiency is proven through simulation and two experiments.