Research On Fault Monitoring And Health Assessment Of Wind Turbine Gearbox Based On Data Driven

With the intensification of the energy crisis,wind energy,as an important green renewable energy,has gradually occupied an important position in the global energy structure.After more than 10 years of rapid development,China’s wind power industry has gradually played a leading role in the development of wind energy in the world.As a green energy base in China,Inner Mongolia has excellent wind energy resources and its installed wind power capacity ranks first in the country.However,the operating environment of wind turbines in wind farms is harsh,and the working conditions are complex and changeable.The load of wind turbines presents typical randomness and non-stationarity.With the continuous expansion of wind power installed capacity,a large number of wind turbines in our district are facing the expiration of the warranty period or close to the reliable operating life of wind turbines,and the failure rate of the wind turbines is increasing day by day.Studies have shown that the operation and maintenance costs of wind turbines are as high as 30% to 35% of the total cost of wind power.Therefore,the study of fault monitoring and health assessment of wind turbines is of great significance to ensure the safe operation of wind turbines,improve the availability,and reduce the operation and maintenance costs of wind farms.Aiming at the problems of low accuracy of fault monitoring and health assessment of wind turbine gearbox and insufficient ability of feature extraction,this paper studies data-driven fault monitoring and health assessment algorithm based on historical big data provided by SCADA system.The research contents mainly include:(1)Feature parameter selection of wind turbines and SCADA data cleaningThere are a lot of abnormal data in SCADA data of wind turbine,which seriously affects the prognostic and assessment of the health status of wind turbine.Therefore,a new algorithm based on Copula mutual information and local outlier factor(CMI-LOF)is proposed to select the characteristic parameters and clean the abnormal data.Firstly,the key SCADA parameters are selected as the data cleaning object based on Copula function and mutual information characteristics.Secondly,a local outlier factor algorithm(LOF)is used to clean the abnormal data,and the outliers and stacking points are filtered out.Finally,the algorithm is verified by the actual wind turbine operation data.The experimental results show that the method can effectively identify all kinds of abnormal conditions and has a good cleaning effect on abnormal data.(2)A new kernel entropy PLS algorithm is proposed and applied to the condition monitoring of wind turbine gearboxAiming at the non-linear and non-stationary characteristics of SCADA data,a new kernel entropy PLS monitoring algorithm is proposed.The algorithm uses Renyi entropy to arrange and reduce the dimension of feature vectors,which can better characterize the angle information between different nonlinear features.In addition,KEPLS can extract high-order statistics effectively,which can overcome the non-stationary problem of data to a certain extent,and solve the problem that traditional KPLS can only represent second-order statistics,and often ignore the fault information hidden in high-order statistics.Then,the validity of the algorithm is verified by theoretical derivation.Finally,the algorithm is applied to the fault condition monitoring of gearbox of wind turbine.The experimental results show that the algorithm has better monitoring effect than the traditional PCA and KPCA method.(3)Health assessment of wind turbine gearbox based on kernel entropy partial least squares and fuzzy membershipIn order to predict the early fault degree of gearbox and evaluate the health status of gearbox,a fault residual prognostic and health status assessment method based on kernel entropy PLS-fuzzy membership is proposed.Firstly,the key variables representing the health status of the gearbox are selected according to CMI.Then,KEPLS is used to predict the normal trend of the key variables,and the residual is calculated with the current state value.According to the residual error of fusion prediction,fuzzy membership is used to evaluate the health status of gearbox.Multivariate predictive residuals are used in the fusion evaluation,which enhances the stability and reliability of the algorithm to some extent.Finally,the algorithm is verified by the actual wind farm operation data.The experimental results show that the proposed prognostic model and assessment method are accurate,simple and intuitive,and can be used to analyze and evaluate the health status of wind turbine gearbox,which has a certain engineering application value.