Research On Wind Turbine Key Mechanical Components Fault Diagnosis Method

With the exhaustion of mineral resources such as oil, coal and other mineral resources, the global ecological environment is becoming worse. And human beings are increasing aware of the significance of renewable wind energy. Hence the wind power industry received widespread concern and attention. Due to the influence of the uncertainty of wind and wind turbine, the maintenance cost of wind turbine is relatively high. The maintenance is mainly derived from the mechanical parts of the wind turbine drive system. So making some research on fault diagnosis method for gear and bearing in the wind turbine.First of all, the vibration mechanism of key parts of the gear box of the wind turbine is studied. The common types and the main transmission system components are analyzed as well as the typical fault and fault characteristics of gear and bearing. At the same time, the paper analyzed the popular signal processing methods in the time domain, frequency domain and time-frequency domain. These studies provide the theoretical basis for the following work.Secondly, the paper studied fault feature selection and extraction method of key components of the wind turbine gearbox and bearing. In view of the signal of the vibration is non-Gauss, non-stationary and non-linear When the key parts of the transmission system of the wind turbine failing,. A method was proposed based on the intrinsic mode function(IMF) and kernel principal component(KPC) analysis to solve the problem that the signal characteristic parameters are not easy to be selected. Before the empirical mode decomposition(EMD) is performed to construct the IMF, the signal was decomposed by soft threshold Sym8 wavelet. In order to solve the high dimension of nonlinear state feature, the method of KPC is used to reduce the dimension of the feature vector.Then, the paper presented a novel fault identification method that is an improved k nearest neighbor classifier(IkNN). Owning to the traditional k nearest neighbor classifier k value is difficult to determine, processing large sample data is very difficult too. The Euclidean distance is used in the selection of neighbors. However, the Euclidean distance is equal to each characteristic parameter so that it similarity calculation of the feature vector is not accurate enough, Whose classification accuracy remain to be further improved. IkNN classification diagnosis model was set to realize fault classification rapidly and precisely for wind turbine key components.Finally, this paper designed fault diagnosis experiment scheme and verified the validity of fault diagnosis classification model. Using the fault simulation test platform of the transmission system of the wind turbine, the failure experiment of the key mechanical components such as gear box gear and bearing are accomplished. 8 kinds of fault data(including normal operation) of the gears and bearings were collected, moreover the effectiveness of the soft threshold Sym8 wavelet is validated by the signal processing. The fault features are extracted by the IMF and the KPC analysis, and the comparison in classification accuracy between IkNN, SVM, ELM, Knn, Fuzzyk NN were made where good results were obtained to verify the effectiveness of improved k nearest neighbor classifier.