The Fault Feature Extraction Techniques Based On Single Channel Blind Source Separation Theory

Mechanical fault feature extraction technology is important to the condition monitoring and troubleshooting of the mechanical system. The observed signal obtained directly from the mechanical system is essentially a mixed signal produced by each components. The effective signal characteristics which can reflect the health of critical components are often submerged in the mixed signal.In order to accurately monitor and diagnosis the key mechanical components, the information which can reflect the state of equipment must be effectively estimated and separated from the complex observed signal. It is also the important task of the fault feature extraction techniques based on single channel blind source separation theory.Single channel blind source separation(SCBSS) is to extract the key feature information only using single-channel mixed signal received by the sensor.In actual situation, due to limitations of mechanical structure, installation conditions and costs, there is often only a single sensor.It’s just same with the application environment of SCBSS.This makes SCBSS having far-reaching scientific significance and value in fault diagnosis of mechanical vibration, attracting a growing number of researchers to study.This paper discuss the separation algorithm of single-channel blind source and its application on the background of the machinery fault feature extraction. The main research work and achievements are as follows.(1) Independent component analysis theory and algorithms are researched and analyzed.Experimental results show that there are several advantages to the algorithm of Fast ICA, such as iterative speed, stable and reliable output and its algorithm is easy to implement, which can be used as a core tools in single channel blind source separation.(2) The single channel blind source separation algorithm based on virtual multi-channel is studied. The algorithm conversion the single-channel issue into virtual multi-channel issue through the single-channel mixed signal preprocessing.The key is to choose preprocessing algorithm. The performance, advantage and shortcoming of preprocessing algorithms such as Delay Act, DWT, EMD and EEMD are analyzed and contrasted through experiments.(3) The principle and characteristics of UWT is studied. Because UWT does not require sampling operation, high pass signal and low pass signal of each decomposition’s output have the same length as the original signal. Information is redundant in the time domain and frequency domain, and the decomposition band are rarely overlap. Its application to fault diagnosis of generator set can effectively extract bearing looseness features.(4) A method in single channel blind source separation is proposed based on the UWTICA. Experiments show that the method has good performance and operation efficiency. The isolated component clearly reveals the characteristics associated with the bearing failure while applying the method in the bearing fault characteristic frequency extraction.