Enhanced eigenvector algorithm for recovering multiple sources of vibration signals in machine fault diagnosis

Many advanced techniques have been developed for vibration-based machine fault diagnosis. One of the prerequisites to use vibration for fault diagnosis is the vibration signal measured from a machine component must be well isolated from other vibrations that are generated by adjacent components. Many machines have numerous and small components that are closely packed together. Due to limited space or accessibility for installing sensors on the inspected machine component, sometimes only one sensor is allowed to be installed. An aggregated source of vibrations could be collected rather than just the vibration generated by the inspected component. Hence, an effective algorithm must be employed to recover the desired vibration out of the aggregated source of vibrations. The blind equalization-(BE)based eigenvector algorithm (EVA) has proven its effectiveness in recovering the overwhelmed vibration signal in the application of machine fault diagnosis. However, the conventional type of EVA can recover only one dominant source from the aggregated vibration. This dominant vibration may belong to the larger vibration generated by the inspected component or a nearby component. Hence, the ability of EVA in recovering signals besides the dominant signal is deemed necessary. In this paper, we proposed an enhanced EVA that consists of channel extension and a post-processing method to recover multiple sources of vibrations. The post-processing method includes the use of correlation and higher order statistics. With the help of these proposed algorithms, the enhanced EVA can recover other vibrations that are less dominant but highly relevant to existing faults. To verify its effectiveness, the ability of recovering the overwhelmed bearing faulty vibration is demonstrated. The results of the experiments using simulated signals and real machine vibrations have proven the effectiveness of the method. Hence, the enhanced EVA is suitable for vibration-based fault diagnosis on machines that have many closely packed components. (c) 2007 Elsevier Ltd. All rights reserved.