Research On Fault Diagnosis And Early Warning Analysis Of Wind Turbine Gearbox

With the rapid development of wind power industry in China,more and more wind turbines have been put into operation,the single machine capacity and total capacity increase year by year,and the safety and reliability of mechanical equipment are also gradually increased.The transmission system enters the peak period of fault with the cumulation of operation time,equipment damages often occur,and effect the normal operation of the wind farm,even endanger the personal safety when seriously.The wind turbine fault warning can help to detect fault symptom earlier,and make fault nipped in the bud;The wind turbine fault diagnosis can capture the fault information accurately and quickly,to prevent major accidents and facilitate staff to maintain to shorten the maintenance time and cost.Fault warning and diagnosis of wind turbine generator has become one of the important research directions of wind power development.In this paper,taking the gear box of wind turbine as the research object,the research and experiment on the two aspects of fault warning and fault diagnosis are carried out,the main contents are as follows:(1)Taking the Datang Gansu wind farm goldwind S50/750 wind turbine as the research object,a fault early warning method based on support vector regression and normal distribution theory is proposed.Support vector regression model of wind turbine gearbox is built based on SCADA system history data.Based on the normal distribution theory,a comprehensive analysis is made on the residual error of the forecast and the actual value under the normal condition,and then the fault warning threshold value is set to realize the fault warning of the gear box of the wind turbine.(2)A wavelet threshold de-noising method based on adaptive CEEMD and permutation entropy is proposed.The amplitude standard deviation and the average number of CEEMD added gauss white noise based on signal itself is select adaptively,after the CEEMD decomposition,permutation entropy is used to select the IMFs which have a bigger randomness,then the wavelet threshold denoising is used to reconstruct the signal.Taking simulated fault signal,time domain signal of wind farm and Laboratory of angular domain signal as examples,this method and integral wavelet threshold denoising,CEEMD decomposition abandoned the IMF1,CEEMD decomposition abandoned the IMF1 and IMF2 methods are compared on the aspect of noise reduction results,the resultshows that this method has a good denoising effect,it can effectively separate different frequency mode of the actual field signal,and can find gearbox missed early slight fault.