Data-based Fault Diagnosis Of Wind Turbine

Wind power is a clean and renewable energy technology with good prospects.Under the current trend of energy saving and reducing pollution,the wind turbine has been used in a large scale.Most of the wind turbine work in a relatively harsh environment,resulting in a higher failure rate.The rapid and accurate fault diagnosis of the whole machine and the effective vulnerability analysis have the significance of reducing the maintenance cost and reducing the loss of power generation.In this paper,the problem of whole machine fault diagnosis in wind turbines is studied,and the two-layer PSDG(Probabilistic Signed Directed Graph)based fault diagnosis method and the improved PageRank based fault ranking method are proposed.These methods have realized the function of the fault diagnosis after the fault and the importance analysis of the fault before the fault,and overcome the shortcomings of the existing fault diagnosis method only for single component and lack of integrity analysis.First,this paper studies the problem of whole machine fault diagnosis of wind turbines after the fault.In order to locate the main fault quickly and accurately,this paper proposes a fault diagnosis method based on two-layer PSDG,which transforms the fault classification problem into a sorting problem.According to the fault list,candidate faults are extracted.By calculating the partial correlation coefficients,a complex network is established,and the kNN algorithm is used to rank the candidate faults.The fault sequence including the main fault is extracted through the fault list,and a Bayesian network is established.Bayes method is used to sort the candidate faults.With the sorting of sensor layer and fault list layer,the comprehensive ranking index is determined to obtain the fault search sequence and the fault diagnosis result.The method ensures quickness and accuracy of fault diagnosis.Then,this paper studies the vulnerability analysis and fault sorting of the wind turbine before the fault.Vulnerability analysis and fault sorting analysis have two levels.For single machine level,this paper chooses the time range of the fault data,and uses naive Bayes method to predict the possible faults of single machine.For multiple machine level,the PageRank algorithm based on forgetting factor and directed edge weight is applied.Considering the weighted edges and the timeliness of faults,a universal importance ranking is achieved.The vulnerability analysis and fault sorting are helpful to the maintenance of the normal state of wind turbines,and to reduce the number of shutdown.Finally,based on the existing platform,a big data platform based on Spark computing framework is built,and the client program is made.On the big data platform,it not only realizes the function of whole machine fault diagnosis,vulnerability analysis and fault ranking,but also achieves the functions of distributed storage and fault message warning.