Study Of Wind Turbine Generator And Gearbox Fault Diagnosis Methods

At the same time of the rapid progress of wind power, it also brings a series of challenges which is highlighted in the frequent quality problems of wind turbine, which seriously affect the normal production and electricity generation, causing huge economic losses. Therefore, the strengthening wind turbine fault diagnosis to reduce wind farm maintenance costs and improve economic efficiency wind farm operation has important significance. Wind turbines mainly include gearbox, generator, blades, hydraulic system and components such as yaw system.Gearbox and generator are the key components, which are also the highest incidence of failure parts whose operation affects the stability of the whole machine performance, wind turbine generator and gearbox are taken as research object, the fault diagnosis methods are studied.Firstly, taking into account the weak vibration response of generator bearing failure, fault characteristic extraction method of generator bearing based domain coherent analysis was presented, coherent analysis method was used to reduce the measurement signal noise and highlight the fault information, and then did the pour frequency domain of coherence function, feature sideband was extracted, according to the characteristics of non-stationary and uncertainty of turbine gearbox vibration signal, diagnosis method of turbine gearbox crack based on wavelet packet and cepstrum analysis was presented. The wind turbine generator bearings and gearbox gear fault diagnosis can achieve through simple spectral analysis.Secondly, the homogeneous information feature fusion method was proposed. Kurtosis and peak were selected as characteristic values in time domain in this algorithm, while in time-frequency domain, frequency band energy and2-norm calculated by wavelet packet algorithm were extracted. Considering the correlation between the characteristic values, PCA (principal component analysis) was applied to determine the principal component, thus decreasing the input variables of neural network. By means of genetic algorithm, the weights and bias of BP neural network were optimized and the fault diagnosis model for genetic neural network was built. Simulation test indicated the effectiveness of this algorithm.Thirdly, the heterogeneous information feature fusion method was proposed, aiming at the limitation of single fault signal and strong nonlinear relation of fault features in Gearbox, the method take the acquired vibration signal、temperature signal and lubricating oil signal as original signals, Kurtosis and wavelet packet band energy, temperature of gearbox bearing and oil, lubricating oil viscosity were respectively extracted. Considering the correlation between the characteristic values, PCA was applied to carry out dimensionality reduction fusion of the combination of original features and obtain information complementary features. The fused features were sent to BP neural network that was optimized by means of genetic algorithm for fault pattern recognition. Compared with homogeneous information feature fusion method, Simulation test indicate the method proposed in this paper can obtain higher diagnosis accuracy.