Research On Fault Diagnosis Of Gearing Box Based On Analysis Of Wavelet And Support Vector Machine

With the depletion of traditional energy,the use of new energy development has become a national key research content.China's vast territory,especially in the Gobi beach area is rich in wind resources,and wind energy as a clean energy is more widespread concern,because the wind power required wind turbine units are generally working in the environment more harsh areas,prone to each Failure,and in the early stages of failure is generally difficult to find,and once the failure really occurs,it will lead to huge economic losses.Therefore,in order to reduce the occurrence of fan failure as much as possible,the establishment of a complete fault diagnosis system for wind turbine is very important for the stable operation of the whole wind power system.The traditional machine learning needs to use a large number of fault data to train the fault model,and the fault data of the wind turbine,especially the actual number of fault data of the gearbox,is limited.Therefore,according to the fault characteristics of the fan gear box,combined with the voting theory and decision tree theory,A support vector machine fault diagnosis classifier based on wavelet analysis is established.In order to better extract the training data,the wavelet signal analysis in the digital signal processing is used to extract and analyze the fault characteristic signal of the gearbox.In order to ensure the fault,the fault vector of the gearbox fault diagnosis classifier based on wavelet analysis is designed.The accuracy of the signal extraction is more accurate and the energy of the fault signal is more concentrated,and the fault signal with higher resolution is obtained.We use the wavelet packet analysis to extract the new fault signal,and then use the classical support vector machine theory to extract the gearbox fault The data were analyzed for fault classification.Because of the multi-classification problem of fault classification problem of gearbox,and the traditional support vector machine is for the second class,we introduce the voting theory and decision tree theory in support vector machine to establish a multi-fault classifier of gearbox.The results show that The multi-fault classifier for wind turbine gearbox has a high classification accuracy.The main content and innovation of this paper1.Firstly,this paper analyzes the characteristics of the fault signal which may occur due to the failure of the gearbox of the wind turbine,analyzes the various eigenvectors of the fault signal from the aspects of time domain and frequency domain,and makes the basis for the extraction of the signal The Secondly,according to the characteristics of various fault signals analyzed in the previous analysis,the difficulties that may exist in extracting the fault characteristic signals are pointed out.Then,several kinds of solutions are analyzed,and finally,The fault diagnosis process is analyzed and the correctness and superiority of wavelet analysis in fault extraction are tested by a series of experimental data.The foundation of fault diagnosis is established.Thirdly,a number of effective classification methods are summarized,and the advantages and disadvantages of each method are analyzed.Finally,a multi-fault classifier is established for the fault signal of gearbox by classical support vector machine classification.Because of the complexity of the fault signal of the gearbox and the difficulty of feature extraction,this paper adopts the wavelet packet analysis with higher signal time-frequency resolution to extract the fault signal.The wavelet packet can make the fault signal energy more concentrated and can analyze the multi-resolution In the high-frequency signal for further decomposition,and ultimately get more accurate fault signal feature vector.3.Because the wind turbine gearbox fault signal data is limited,and the traditional machine learning requires a lot of training data.In this paper,the theory of voting in statistics and decision tree theory are added to the training process of fault model.The experiment proves that this method can obtain high classification accuracy under the premise of less fault data.For practical engineering The application has great theoretical value.