Research On Failure Process Modeling And Failure Monitoring Method Of Wind Turbine

Wind turbine is a complex mechanical and electrical system with complex structure and numerous components,its working environment is mostly located in remote areas,which is prone to extreme weather such as storm,lightning and icing,etc.In addition,with the complexity of wind turbine and the expansion of the wind field to more severe environment in recent years,wind turbine failures occur frequently,which brings severe challenges to the stable operation of wind turbine.It is of great significance to explore the evolution trend of wind turbine failure process and carry out failure monitoring research to reduce the failure rate of wind turbine and ensure the stable operation of the unit.This study is supported by the National Natural Science Foundation of China “Research on the Performance Degradation Mechanism of Variable Stiffness Wind Turbine Blades under Typical Northwest Wind Resource Environment(No.51965034)” and the Lanzhou Talent Innovation and Entrepreneurship Project “Health Assessment for Mechanical Equipment Operation and Maintenance And Intelligent Decision Optimization(No.2018-RC-25)”,this work takes 2.5MW direct drive wind turbine as the research object,combined with the wind turbine operating data collected by the SCADA system to carry out failure data classification and cleaning research.On this basis,a mathematical model of the wind turbine failure process is constructed according to the characteristics of wind turbine shutdown failure.Taking the main bearing of the wind turbine as an example,according to its temperature characteristics combined with the sliding window residual statistical principle,the failure monitoring of the key components of wind power units is expanded.The specific research content are as follows:(1)Classification and cleaning of abnormal data of wind turbine.To solve the problem that it is difficult to identify and clean the abnormal failure data of wind turbine,a data cleaning method based on Box-MDBSCAN is proposed.Firstly,combined with the wind speed-power data representing the operation status of wind turbine,the abnormal data are classified according to the distribution characteristics of the abnormal data.Then the Box-whisker Plot method,the Modified DBSCAN algorithm and the Box-MDBSCAN method are respectively applied to identify and clean the abnormal data.Finally,the Spearman coefficient is used to verify the effectiveness of the proposed method.The results show that there are four types of abnormal data of wind turbine based on wind speed-power data,the Box-MDBSCAN method is the best cleaning effect,and the Spearman coefficient of Box-MDBSCAN is 0.0105 and 0.0107 higher than that of Box-whisker Plot method and MDBSCAN method,respectively.(2)A mathematical model of the wind turbine failure process is established.Aiming at the problems of the traditional wind turbine failure modeling methods,such as inconsistent applicable conditions and large errors,the power law process is selected as the carrier to model the failure process of wind turbine.At first,the trend test of wind turbine shutdown failure data is carried out,and the wind turbine failure model based on power law process is established according to the Non-Homogeneous Poisson Process characteristics of shutdown failure data.Then,the Least Square(LS)method,Maximum Likelihood Estimation(MLE)and Levenberg-Marquardt(LM)method are used to estimate the model parameters.At last,the rationality of the proposed method and the advantages and disadvantages of different parameter estimation methods are verified by solving the goodness of fit index of the model.It shows that the goodness-of-fit indexes obtained by different parameter estimation methods are all close to 1,which indicates that the power law process model can accurately describe the failure process of wind turbine.Besides,the LM algorithm has higher precision in solving model parameters,which are 0.2% higher than that of MLE method and 4.2% higher than that of LS method.(3)Research on Failure Monitoring of Main Bearing of Key Components of Wind Turbine.According to the characteristics of abnormal temperature fluctuations caused by the failure of the main bearing of the wind turbine,a failure monitoring method is proposed based on the temperature model for failure monitoring of the wind turbine main bearing.First of all,the multiple linear regression model,grey model,support vector regression model and combined prediction model of the main bearing temperature are established under normal operating condition;Then the sliding window method is introduced on the basis of the best prediction model to study the statistical characteristics of the main bearing temperature prediction residuals;Eventually,it is judged whether the main bearing has a failure by comparing the confidence interval of the mean or standard deviation of the temperature residual with the set critical value.The research results show that the main bearing temperature combination prediction model has the best prediction effect,and its determination coefficients are respectively increased by 0.0493,0.0027 and 0.0002 compared with the multiple linear regression model,gray prediction model and support vector regression model;The statistical characteristics of the temperature prediction residuals solved by the sliding window method can reflect the operating status of the main bearing in time,which provides a theoretical basis for the failure monitoring of the main bearing and other key components of the wind turbine and the formulation of scientific and healthy maintenance strategies.