Blind Separation Of Convolutive Mixtures Of Non-stationary Vibration Signals

Rotating machinery is widely used in industrial sector. They are often out of order because different kinds of faults, which would affect their normal work and bring about serious economic losses. Then, it becomes necessary that how to monitor effectively and analyze by rotating machinery vibration signal is one of the key concepts.The appearance and development of blind source separation make the separation of mixed-signal becoming possible. This paper makes the summary of the use of blind source separation in the analysis of rotating machinery vibration signals. At present, blind source separation is rarely used in the separation of convolutive mixtures of mechanical vibration signal. This paper will make some exploratory research in this direction.The existing methods, which used in the separation of convolution, are mostly based on time and frequency domain. In terms of the combination of time domain and frequency domain, this paper proposes a new method to solve the problem of the convolutivly mixed non-stationary signals. According to the characteristics of non-stationary signals, firstly the method sorts out the time interval in which only one source signal is working, so that the multi-input multi-output system is transformed into a number of single-input multiple-output system. Then, the difficulty to solve the problem is decreased. The second step is to solve the single-input multiple-output system. Compared with existing methods, this method avoids the problems of uncertainties and reduces the amount of data needed.In this paper, the simulation experiment was made by using this method, and the actual vibration data were analyzed to verify the feasibility and good result of the method.