Study On The Fault Diagnosis Of Turbine-generator Sets Based On Wavelet

With the development of productivity and the advancement of science and technology, it is very important to the national economy that diagnosing the fault of a complex equipment, that tracking its running state closely and that safeguarding it according to its actual running state.The abnormality or fault of mechanical equipment in running process will lead to the unstable dynamic signal. Although traditional Fourier transform(FFT) gets a widespread use in engineering, it only adapts to analyzing stable signal. To unstable signal, FFT averages its frequencies. So it can not reflect the feature of signal true.Wavelet is a method of time-frequency anslysis. It has a higher frequency distinguished-degree and a lower time-distinguished degree in low frequency part. As high frequency part, it is the other way around. So wavelet is adapt to probing the instantaneous abnormal phenomena of signal exactly and to show its components. In addition, in state monitor and fault diagnosis area, signal analysis of vibration and noise has already been in a critical position. It can separate the real signal from noise signal effectively and attains the purpose of de-noise by using time-frequency localizing anslysis, local feature abstract, time-change filter wave, restraining or attenuating some frequency range and other characters of wavelet. Furthermore, the ratio of signal to noise has been improved greatly and it is not affected by signal state.This thesis studies the law of affecting de-noise result and the selection of the threshold and the wavelet function, the combination of wavelet and FFT in the fault diagnosis of turbine-generator sets:By the de-noise anslysis of Blocks and Sin signals, concludes: To Blocks signals, usually adopts soft threshold; the law of affecting de-noise result is when use wavelet auto-de-noise, with the increasing of decomposed level, the de-noise result becomes worse while the level blow the 3, when the level above 3 and when uses wavelet packet, it is the other way round; the best de-noise methods of the signal is that uses "dbl "wavelet function, three level, soft and "rigrsure" threshold. To Sin signal, also adopts soft threshold; both by wavelet and by waveletpacket, with the increasing of decomposed level, the de-noise result becomes better; the best de-noise methods of the signal is that uses "sym8" wavelet packet, eight level, acquiesced or regulated threshold and "Shannon " entropy.The shape of turbine-generator vibration signal is similar to Sin signal. This thesis analyses the vibration signals of measured point 1 of 2#set in a generating plant, by applying the method of analyzing Sin signal which uses "sym8" wavelet packet, eight level and "Shannon" entropy to de-noise. This method eliminates the noise effectively and retains the main useful information, is not simple low or high pass-through filter. It shows the above conclusion is effect, especially the choice of wavelet function.This thesis measured the vibration signals of 1# and 2# running turbine-generator sets in a generating plant and their vibration signals that are indispensable to the study are gained.Firstly, the low frequency faults of machine stand in generator side, rotor unbalance, misalignment, oil whirl, oil whip are diagnosed by FFT analysis. Secondly, applying wavelet theory to probing the instantaneous abnormal phenomena of the signals and to show their components. The faults of serious friction is diagnosed precisely. The result is accurate, effective and conforms to the results of overhaul.