Research On Early Warning Method For Self-excited Vibration Of Turbo-generator Units

This paper analyzes the two main kinds of failure modes (instability of the oil film and steam-excited vibration) of turbo-generator units shaft self-excited vibration using failure modes and effects analysis method (FMEA) and fault tree analysis method (FTA). This analysis provides theoretical support to the research on fault early warning of the turbo-generator units shaft self-excited vibration. This paper presents a combination method of ensemble empirical mode decomposition (EEMD) and BP neural network to predict time series based on the shaft vibration data of operating turbo-generator unit in plants. And this combination method considers the nonlinear and nonstationary characteristics of the data. After failure modes for self-excited vibration of turbo-generator units having been analyzed and prediction model of vibrational trend having been constructed, the paper takes the fault of a power plant in actual production as example and analyzes the early warning method for the self-excited vibration of turbo-generator units based on EEMD-BP model. And it can provide a useful reference for the safe operation of power plants.In the research of turbo-generator units shaft vibrational trend forecast and early warning method, the shaft vibration signals is an important state parameter which is nonlinear and nonstationary. The traditional methods are difficult to find the variation rule and can’t predict it perfectly. But EEMD method has a good applicability for the nonlinear and nonstationary signals. So it can be used to analyze the shaft vibration time series of turbo-generator units. It is a powerful research tool for power plant equipment condition monitoring. This paper uses ensemble empirical mode decomposition (EEMD) method to decompose the original signal of turbo-generator units shaft vibration amplitude (micro m pp) into several intrinsic mode functions and the residue. Take the various components as input factors to BP neural network for one step prediction. Then reconstruct the forecast results of each component, we can obtain predicted data for the original time series using the method of EEMD-BP. To measure the effect of EEMD-BP prediction method, the paper uses BP neural network method and the EMD-BP prediction model which is the combination of the empirical mode decomposition and BP neural network to predict the same original time series respectively. The comparison of the forecast results by three kinds of method verifies the fact that the EEMD-BP and EMD-BP method can achieve better prediction effect for turbo-generator units shaft vibration original time series than using BP neural network method. And EEMD-BP method is the best. The preprocessing for the original signals using the EEMD method has a positive meaning for improving forecasting accuracy of shaft vibration time series.