Blind equalization based eigenvector algorithm for the recovery of mechanical vibrations

Many advanced techniques have been developed for the analysis of mechanical vibrations. It is one of the prerequisites to vibration-based machine fault diagnosis that the vibration signal measured from a machine component must be well isolated from other vibration signals that are generated by adjacent components. Due to the physical constraints of installing sensors in the machine, sometimes only one sensor can be installed. Hence, the sensor will collect an aggregated source of vibrations rather than just the vibration generated from the inspected component. Manufacturing machines are prone to such interference of multiple vibrations. Thus the fault-related vibration must be recovered from the aggregated sources for accurate fault diagnosis. In this paper, the eigenvector algorithm (EVA) of blind equalization (BE) is applied to the recovery of mechanical vibration signals. The conventional EVA can extract only one dominant source from the collected data at a time. In this paper, we propose an enhance EVA that is constructed with the method of channel extension and further post-processing algorithm to recover multiple sources of vibrations. That is, besides the dominant vibration, other less dominant vibrations but relevant to existing faults can also be recovered by using the analyses of correlation and kurtosis. The experiments performed on real vibrations that are generated by industrial machines present the effectiveness of the proposed methods.