The Method Of Diagnosis And Prediction Of Bearing Shaft Current Fault Of Wind Turbine Based On Time Frequency Manifold

Bearing has always been an indispensable core component of wind power generator, which plays an important role in the operation of wind turbine generator. With the extensive use of clean energy, wind power generator toward the direction of large-scale, complex. When the electric insulation layer of the bearing of the wind turbine is worn, the load is abrupt or suffer from lightning and other extreme weather, generator axial current problem becomes very prominent, and influence caused by wind power generator normal operation, if not promptly treated, may cause failure of wind turbine, causing huge economic losses. Compared to the ordinary bearing fault vibration signal, shaft bearing current fault vibration signal with multiple modulation characteristics, traditional frequency analysis method, it is difficult to identify the fault characteristics and fault warning. As the result, the author multiscale subband manifold preserving algorithm for fault diagnosis and sub with manifold secondary were square root bearing damage prediction method. The experimental results show that the proposed method has a good effect. The main research work is as follows:(1) The current damage process of the bearing of the wind turbine is discussed, analysis of the bearing shaft current damage of bearing vibration signal in time domain and frequency domain characteristics, found that the relative rolling bearing fault vibration signal features, bearing shaft current injury due to the inner ring and the outer ring and rolling bodies have been damaged, fault vibration signal spectrum is difficult to find a single fault characteristic frequency, with characteristics of multiple modulation;(2) Proposed a multiscale subband manifold preserving algorithm for fault diagnosis, first of all the fault signal wavelet packet decomposition, to get multi-scale signal, of multiscale signal subband decomposition, to get multi-scale subbands signal, and extract multiscale subband signal sample entropy, preliminary fault signal characteristic value, smoothed pseudo Wigner Ville distribution characteristic of signal values, then the of Locality Preserving Projection(LPP) manifold learning and nonlinear dimensionality reduction, fault signal the final time-frequency manifold fault feature extraction, bearing fault signal and used to verify the proposed method. The results show that the method of bearing of multi class fault recognition has a very good effect.(3) Proposed a subband manifold secondary were square root bearing damage prediction method based on, first of all the fault signal wavelet packet decomposition, extraction of multi-scale signal, again for each a multiscale signal manifold learning and nonlinear dimensionality reduction, manifold subband signals is extracted, again for each flow shaped sub band signal secondary root mean square value extraction, fault prognostics feature values, and the use of adaptive Back Propagation(BP) neural network to predict the fault, the the method has been verified by the bearing fault signal simulation. The results show that the method of complex bearing fault has good prediction effect;(4) Build shaft bearing current damage simulation test bench, extraction at different stages of the bearing shaft current damage fault vibration signal, the frequency manifold fault diagnosis and condition prediction method of verification, results show that based on multi-resolution sub with manifold maintain fault diagnosis calculation method can effectively identify the bearing shaft current damage fault. The accurate rate of recognition can reach 96%, based on subband manifold secondary were square root bearing damage prediction method can effectively predict the bearing current damage state prediction errors lower than 4.33%.