Abstract:

The vibration system of the steam turbine rotor is a definite complex system ,the vibration sequence of which consists of many kinds of frequency components.An integrated and accurate mathematical model is important to the extraction of fault premonition information and the forecasting of faults.In this paper,according to the data in both the horizontal direction and the vertical direction of four typical faults,such as rubbing,loosing,uncountershaft and mass unbalance collected from the Bently experimental set,a ARMA(AutoRegression Moving Average)
model is built up for the vibration fault series after the stable examination of the random stationary noise item in which both the tendency and periodicity items ale removed.The calculated results indicate that the mean errors of 8 forecasted vibration faults of turbine rotor in half period obtained by the ARMA model are all less than 0.55 ixm,and the determ inistic factors r are all more than 0.99 1 5.This means that the proposed method is of great forecas.ting precision and is effective in the further extraction of fault symptom inform ation and the further forecasting of the developing trend of vibration faults.

Key words: stream turbine rotor, vibration fault, forecasting, ARMA model