Research On Fault Diagnosis Method Of Wind Turbine

Wind energy is an important kind of clean and renewable energy developed all around the world.It is important to the amount of wind energy whether wind turbines can run safely or not.Wind turbines often run in the poor environment,so it is easy for them to malfunction.Maintenance crews will overhaul them regularly and judge the faults through their experience.With the spread of computers,through setting up sensors on the important parts of wind turbines,the data of wind turbines will be collected in the database when they are running.Studying these data and finding useful information in them can give a hand to maintenance crews and reduce losses caused by wind turbines faults.This paper firstly proposes a method of fault diagnosis of wind turbines based on rough sets and BP neural nets through studying of data of wind turbines.It firstly reduces dimensionality of attributes based on discernibly matrix.Then use equivalence partitioning to get the rules in the neural nets and diagnose the faults.Results of experiments shows that the method of fault diagnosis of wind turbines based on rough sets and BP neural nets can effectively avoid the noise jamming made by redundancy attributes.It can also improve the accuracy of diagnosis.But the study also found that the failure of this method for a wind turbine gearbox and pitch system occurring in the troubleshooting process,convergence is slow,and diagnostic accuracy relative to other failure diagnosis is significantly lower.In order to solve the above problems,this paper proposes a method of fault diagnosis based on wavelet neural network.In this method,the unsupervised learning based on principal component analysis to preprocess the data,and then use wavelet neural network approach to mine the knowledge in data.For the difficult of distinguishing the parameters,this paper propose the concept of fault offset vector group,not simply divided the parameters into two kinds of normal and abnormal.This paper think the parameter of abnormal also divided into a variety,which can be used to distinguish between the scope section.These different abnormalities may lead to different failure to provide a guarantee for a more accurate diagnosis of faults.This paper shows that the proposed method of diagnosis,for maintenance staff,as long as the input data of wind turbines into the neural network.They could diagnose faults without any prior knowledge.The method can effectively improve the wind fault diagnosis efficiency.