An Analysis Approach for Multivariate Vibration Signals Integrate HIWO/BBO Optimized Blind Source Separation With NA-MEMD

Abstract

The feature extraction of multivariate vibration signals requires a good separation of signals from different sources so as to solve the problem of aliasing between different source signals and background noise. In order to avoid the loss of local information, multiple sensors may be used to collect signals at different locations on the equipment. In this paper, an analysis method combines blind source separation (BSS) and noise-assisted multivariate empirical mode decomposition (NA-MEMD) is proposed. The BSS algorithm optimized by hybrid invasive weed/biogeography-based optimization (HIWO/BBO) is used to separate the multi-component mixed signals with chaotic noise and the negative entropy of the separated signal as the objective function of HIWO/BBO. To reduce computational complexity, the separation matrix takes the form of a parametric representation. Afterward, the multiseparation signals without chaotic noise are decomposed with NA-MEMD, then the sensitive intrinsic mode function (IMF) is extracted by the improved correlation coefficient (ICC) analysis method. The advantage of the ICC is that the sensitive IMFs at different orders can be selected simultaneously in all channels. Finally, the characteristic frequency of the vibration signal can be obtained by analyzing the sensitive IMFs. The effectiveness of this proposed method is verified in the application of the synthetic signals and the actual bearing fault signals. It shows that this approach can play a role in filtering out the chaotic noise while solving the aliasing problem of mixed vibration signals. For another, it synchronously decomposes the multidimensional separated signals and extracts the characteristic frequency.