Research On Fault Diagnosis Technology Of Wind Power Transmission Machinery Based On Image Feature Extraction

With the development of the information theory and image analysis theory, in recent years fault diagnosis methods based on image processing research have become a hot research field of fault diagnosis. Time-frequency image compared to the time domain and frequency domain graph contains more abundant wind power transmission machinery running status information, more suitable for wind power transmission machinery failure analysis, but the time-frequency image analysis generally need professional personnel discriminant analysis, and general parameters in time domain or frequency domain analysis, did not take full advantage of image information, the low degree of automation. Therefore, this article research wind power transmission machinery fault diagnosis method based on image feature extraction, to improve the efficiency and accuracy of fault diagnosis. The main research work includes:(1) In order to make image processing technology applied to fault diagnosis, firstly, research how to use the vibration signal to build suitable parametric image. In this paper, short-time Fourier transform, Wigner-Ville distribution and wavelet transform based on Threshold De-noise is the three time-frequency analysis methods we have studied, and analysis the analog signal with the time-frequency methods to obtained the two-dimensional frequency images,the comparative analysis results showed that the time-frequency images obtained by wavelet transform based on Threshold De-noise has the best time-frequency resolution and noise immunity, can more clearly reflect the wind turbine failure information.(2) In order to extraction the feature of the time-frequency image and get the statistics feature, we studied how to extract image feature with GLCM method or Hu invariant moments method,and then combined the two methods utilizing the correlation between the eigenvalue extracted by the two methods to have a better performance of the timefrequency image. GLCM accurately reflect the texture of space complexity, roughness and repeat directions, Hu invariant moments can reflect the characteristics of the image area fully, can more efficient extracted the wind turbine drive mechanical failure information.(3) For the image feature vectors we have extracted before, we applied the negative selection algorithm of the artificial immune algorithm to the wind power transmission machinery fault diagnosis, and then studied the variable real threshold immune negative selection algorithm and the process of the fault diagnosis method used to the Image Identification.(4) To verify the study of the wind power transmission machinery fault diagnosis method based on extract the image characteristics, we use the wind turbine gearboxes normal vibration signals and failure vibration signals by the wind farm in Hebei. After the data processed, build the time- frequency images, image feature extracted and applied the artificial immune algorithm for fault analysis. Then we improve the accuracy of fault diagnosis successfully.