Research On The Fault Feature Extraction Methods Of Wind Turbine Bearings Based On Blind Source Separation

Nowadays Wind power industry is developing rapidly in our country, but high maintenance costs is always one of the important factors that restrict the wind farm's economy performance. As a vital components of wind turbine transmission system, the real-time condition monitoring and fault diagnosis for the bearings are beneficial to improve the operation stability of wind turbine and reduce the maintenance costs. In this paper, the blind source separation technology is used in the fault diagnosis of wind turbine bearings, and research on the fault feature extraction methods based on blind source separation in detail. Finally, the example analysis shows that this method can extract the fault feature of the bearings via analyzing the vibration signal of the wind turbine bearings. The details are as follows:(1).Study the three kinds of most commonly used methods of blind source separation, that is the blind resource separation method based on ICA, Fast ICA and the maximum signal-to-noise ratio. A few separation performance evaluation index is applied to evaluate and compare the separation results of three kinds of algorithms though a Matlab simulation. According to the comparison, It concluded that Fast ICA algorithm can achieve better separation precision in a shorter time. So we determine to use Fast ICA algorithm to analyse the vibration signal of wind turbine bearings.(2).Blind source separation methods tend to assume that the noises can be neglected, but practically, considering the complicated wind turbines, the fault signal of bearings is hard to be recognized for being vulnerable to affection by normal signal and the inference of the noise. In view of the above problems, a new method of the fault feature extraction is proposed. This method efficiently combines the wavelet- envelope demodulation analysis and the ideal blind source separation methods. Firstly, the method envelops and demodulates the original vibration signal in order to analyze the signal and implement the wavelet denoising. This step can restrict the inference of high frequency signal effectively. Then Fast ICA algorithm is applied to separate the signal that we get at the first step. Finally, the example analysis shows that this method can extract the fault feature of the bearings via analyzing the normal and fault vibration signal of wind turbine bearings.(3).Aiming at the case that it can only obtain single-channel signal in actual bearing fault diagnosis, a signal-channel BSS algorithm is presented. This method is based on Empirical Mode Decomposition and Fast ICA algorithm. Firstly, the observation signal is decomposed into a number of IMFs, so a virtual multichannel signal is composed by the observation signal and its IMFs. Then we estimate the source number of the virtual multichannel signal using the SVD method, and rebuilt the multichannel signal according to the number of source signal. Finally, Fast ICA algorithm is applied to separate the signal that we get at the second step. This method is used to analyse the vibration signal of gearbox bearings. Its effectiveness is proved by extracting the fault feature of the outer ring of bearings. Finally, the example analysis shows that this method can extract the fault feature of the bearings via analyzing the normal and fault vibration signal of wind turbine bearings. At the same time, the example analysis proves that this method used in the early fault diagnosis for bearings is effective.