Research On Vibration Fault Feature Extraction And Diagnosis Of Wind Turbine Rolling Bearing

Rolling bearings,as one of the key components of the transmission chain of wind turbines,are often difficult to detect in complex working conditions and can evolve into major accidents.In order to eliminate adverse safety hazards of wind turbines and effectively improve the reliability of wind turbine power supply,early fault feature extraction and diagnosis of rolling bearings has become a hot and important research direction.The main research content of this article is as follows:(1)A complete set empirical mode decomposition(CEEMDAN)method with adaptive noise is introduced to address the problem of modal aliasing and pseudo components in empirical mode decomposition(EMD).By utilizing the advantages of HHT time-frequency analysis and comparing simulated signals,it is found that CEEMDAN has a more significant effect in combating mode aliasing compared to EMD and EEMD,making it a basic method for signal decomposition in the following text.(2)A method based on CEEMDAN decomposition,autoregressive minimum entropy deconvolution(AR MED)filtering,and 1.5 dimensional spectral demodulation is proposed to address the issues of weak vibration signals and low signal-to-noise ratio in the early stage of rolling bearings.Use AR-MED to filter and denoise the intrinsic mode function reconstructed from CEEMDAN decomposition based on the kurtosis correlation coefficient criterion,enhance its impact component,and finally perform a 1.5-dimensional spectral analysis on the filtered signal.Compared to traditional Fourier transform,envelope spectrum,TEO energy spectrum simulation and measured signals,the method used in this article can effectively identify the type of bearing fault and has better accuracy.(3)A novel swarm intelligence biomimetic algorithm based fault diagnosis method for parameter optimization is proposed to address the problem of poor fitting performance caused by the difficulty in selecting important parameters [C,G] in support vector machines(SVM).Firstly,a series of natural mode components of bearing vibration signals are decomposed using CEEMDAN,and then the sample entropy of each component is calculated and merged into eigenvectors,which are then sent to the SVM network optimized by Sparrow Search Algorithm(SSA)for training and testing.Finally,it was verified through experiments that the diagnostic model proposed in this article has good classification performance.