Extraction of sparse equalized signals in recovery of potential cyclic impulses from a multi-fault signal mixture

Abstract

Up to now, many advanced signal processing methods have been developed for machine condition monitoring and fault diagnosis. One assumption for the use of these methods is that one vibration source is isolated from other irrelevant sources in advance. In some practical cases, when a transducer is installed in the vicinity of closely arranged components of a machine, it is inevitable to obtain a vibration signal mixture generated by multiple sources. As a result, the isolation of the desired vibration source from other vibration sources becomes a challenging problem if only one transducer is employed to sample the single-channel signal mixture. Blind equalization based methods, such as the eigenvector algorithm (EVA), are potentially capable of recovering each of the vibration sources through setting different equalizer lengths. However, selection of an appropriate equalizer length is rarely reported by a systematical method. To determine the appropriate equalizer length, an improved EVA for extracting sparse equalized signals, such as a cyclic impulsive signal, is developed in this paper. The improved EVA is able to automatically select an appropriate equalizer length for the EVA and adaptively recover the cyclic impulsive signal from multiple vibration sources. Two multi-fault signal mixtures, including a simulated signal and a real vibration signal collected from an industrial machine, are employed to verify the effectiveness of the improved EVA. Comparisons between the original EVA and the improved EVA are done. The results demonstrate that the improved EVA is effective on automatic selection of the appropriate equalizer length and adaptive recovery of the cyclic impulsive signal of interest from the single-channel multi-fault signal mixture. Finally, the improved EVA is generalized to extract different kinds of signals.