Research On Fault Feature Extraction Of Wind Turbine Drive Trains Based On Vibration Analysis

Wind turbine drive train, which connect the wind rotor and the generator, is the key components to achieve energy conversion and transfer process in a wind turbine generator system. Compared with a drive train in general machinery equipments, the load state, operating conditions, environmental conditions, structural layout etc. of the wind turbine drive train are more complex, and that results in the particularity of failure mechanism and failure development mode for the gearbox, bearings and other major parts in a wind turbine drive train. So the failure rate is higher than similar equipment of other industries and the actual service life is much lower than the design life. Therefore aimed at the operational feature of wind turbine drive train and the common problems of fault diagnosis, the study on new theory and technique for improving the accuracy of fault diagnosis and prediction is of great significance for ensuring the health operation of the wind turbine equipment.Vibration monitoring is one of the main technologies for mechanical equipment condition monitoring and fault diagnosis and has been widely used in many industries. During the equipment procurement in the early wind farms in our country, vibration monitoring system is generally not equipped as the considerations for reducing the costs. With the investment of time for equipment, the problem of the failure rate and high maintenance costs is increasingly prominent. Wind power enterprises started to pay attention to running condition monitoring and fault diagnosis of wind turbines. Several power group companies project to carry out related research. And many wind companies has installed vibration monitoring and fault diagnosis system through equipment modification for running wind turbine. Especially, since the National Energy Board issued recommended national energy industry standard NB/T31004-2011’Wind Turbine Vibration Condition Monitoring Guidelines’, the development wind turbine vibration monitoring and fault diagnosis technology has been promoted powerfully.The structure of wind turbine drive train is complex and machine, electricity, liquid couples. Fault signal has the characteristics of high background noise, nonstationary and nonlinear. The route of transmission and the characteristic of attenuation is complex for fault signal and it is often a aliasing of multiple fault signals. It brings a problem to accurate analysis for the fault information. And it has an effect on the accurate fault diagnosis. Therefore, the key for achieving accurate fault diagnosis technology is to study the characteristic and the extraction technology of fault signal and obtain accurate fault characteristic information from monitored signal. In this paper, the research of vibration signal monitoring and new fault characteristic extraction technology for wind turbine drive train in complex operation environment is carried out in this context. The main research contents include three parts:1) on the basis of the classical cepstrum, two improved methods based on cepstrum analysis for feature extraction are carried out.2) on the basis of the classical envelope analysis, an improved method based on the Hilbert envelope analysis for fault feature extraction is carried out.3) take the average of low frequency part(demodulating component) in resonance region slice of spectrum correlation function as characteristic value.Specific study and main results of the various parts are as follows:1) A method which uses the homomorphic filtering of cepstrum for fault characteristic extraction of wind turbine drive train is proposed. The monitored gearbox vibration signal is filtered in the quefrency domain. As the vibration signal is the convolutional results with incentive source and structure natural vibration, through filtering in the quefrency domain, incentive source and structure natural vibration are separated. Then spectrum reconstruction is performanced by the resonant component in low quefrency range and used for reflecting the changes of structure inherent characteristics which is caused by the fault.2) Amplitude envelope information that reflects the failure is acquired by choosing vibration signal band which reflect structure resonance. This is narrow-band envelope analysis based on the Hilbert transform. The main problem of this method is that the resonant frequency band of narrow-band filter is difficult to determine, especially for wind turbine drive train of this kind of complex mechanical equipment. By choosing different filter bands, the results of envelope spectrum vary widely. This brings problems to the follow-up analysis, even wrong results. Aiming at this problem, a method called "spectral envelop" is proposed. The main idea of this method is similar to "narrow-band shifting cepstrum". Using the narrow-band filter which the center frequency shifts, the vibration signal is filtered. Then amplitude envelope spectrum for each section of the filtered signal is calculated. If the center frequency of narrow-band filter changes from low to high frequency, an envelope-spectrum array of "frequency-frequency" domain is got. It is called "spectrum envelope". Normal and failure state can be distinguished directly.3) Based on cyclic spectrum density analysis, the average of low frequency part(demodulating component) in resonance region slice of spectrum correlation function as characteristic value is proposed. The calculation amount of cyclostationary analysis is reduced and the analysis efficiency is improved.