Research On Prognostic Approach Of Degradation State Of Large-scale Wind Turbine Main Bearing

In recent years,due to the emergence of environmental issues such as global warming and haze,people’s awareness of environmental protection has been raised.Therefore,wind energy,as the most mature and renewable clean energy,is gradually replacing traditional energy technologies such as coal and petroleum.As the size of wind turbine is increasing year by year,and the service is subjected to variable loads in extreme environments such as wind,sand,rain,and snow,the performance of the main transmission components of wind turbine will be seriously degraded.Moreover,the main bearing is used as a key component of wind turbine,once it fails,which will cause equipment downtime and bring huge economic losses.Therefore,it is possible to timely detect the early failure point of wind turbine main bearing and accurately predict the future degradation state,and then the maintenance of the main bearing at the proper time has very important significance for ensuring the normal operation of wind turbine.Therefore,this paper studies the four aspects of de-nosing preprocessing of vibration signal,the extraction of fault eigenvector,the establishment of the eigenvector regression performance indexes,the establishment and prediction of prediction model.The main research content includes the following aspects:Firstly,aiming at the complicated and changeable extreme environment of wind turbine,the vibration signal is non-stationary and non-linear,and is mixed with many noise signals.In order to filter out the noise signal,the original fault signal is used to wavelet packet de-composition with appropriate wavelet base and reconstructed according to the calculated optimal wavelet packet tree to realize de-noising for the original signal and improve the signal-noise ratio,which can provide a good guarantee to extract the fault eigenvector of the vibration signal accurately.Secondly,aiming at weak endpoint effect problem of the traditional local mean decomposition,a fault feature extraction method is proposed based on improved LMD.The reconstructed signal is de-composed by using LMD method,and then the correlation coefficients between product function components and the reconstructed signal and kurtosis of PF components are calculated in order to eliminate the false component and enhance the amplitude of fault signal.Then envelope spectrum analysis of real PF component is carried out,and the fault feature of fault signal is extracted.The experimental results show that the improved LMD method can effectively extract the early fault eigenvector and successfully diagnose the fault type of the roller bearing.Finally,aiming at the problem of degraded state prediction of wind turbine main bearing,a prediction method is proposed based on improved LMD and gray model.The kurtosis value and RMS of the PF components that can best reflect the frequency of fault feature in each stage of the full life cycle are used as regression performance indexes to train the GM prediction model.Then,the trained GM model is used to predict the trend of regression performance indexes in the future period of time,and the degradation state is determined according to the curve of the two regression performance indexes.Through the experimental data of full life cycle of wind turbine main bearing,the results show that the trend of the two regression performance indexes predicted by this method can successfully determine the deteriorating trend of the future running state of wind turbine main bearing.