A Hybrid Model For Steam Turbine Rotor Vibration State Forecasting

In this paper, the development of steam turbine condition monitoring and fault diagnosis at home and abroad have been introduced. The turbine rotor vibration state forcasting was put forward based on the steam turbine safely working and changing from periodic repair to corrective maintenance. In the second chapter, the common faults of rotor imbalance, rubbing, rotor misalignment, oscillation and components loose etc. have been analyzed in theory and their main feature have been given out. All of these provide the guidance of turbine state prediction. The time series model such as AR (p), MA (q), and ARMA (p, q) and the artificial neural network model have been analysed in detail in the chapter three and they are the basic principle of each unique forecasting model and the hybrid forecasting model.As well known, the system of rotor vibration is typically nonlinear and time-dependent and the forecasting model for rotor vibration should satisfy the requirement of the accuracy of prediction and the speed of calculation. In order to solve the contradiction between forecasting accuracy and computing speed, a new hybrid model based on AIRMA(p,d,q) and BP model for rotor vibration online forecasting has been proposed in the paper. In this new method, the time series of vibration have been considered containing both linear and non-linear structures and the ARIMA(p,d,q) model and BP model are used to predict the linear and non-linear part respectively. By this way we can take advantage of the unique model and forecast the turbine state more accurately. A new forecasting model structure have been give out in the paper. In this method, the structure was divided into two parts which are the model training part and the online forecasting part. New forecasting model is trained continuously by the model training part based on the real-time data. Once the new model trained, it is shifted to the online forecasting part and used to forecast the rotor vibration. In this new method, the online forecasting part has avoided the time-consuming model training process and the forecasting model can be renewed in time without interrupting the prediction.The steam turbine rotor experiment set has established to simulate the steam turboset’s different working state and used the DASP software to collect the vibration data. The ARIMA(p,d,q), BP model and hybrid model have established and used to forecast the vibration of different state in horizontal and vertical direction. Experimental results show that the method proposed is valid for rotor vibration state forecasting. Finally, a MATLAB GUI program containing each unique forecasting model and the hybrid forecasting model was biult and then changed to the C program. By this way, the program with a friendly interface can work under the MATLAB and DOS environment and be widely used.