Research On Feature Extraction And Diagnosis Methods Base On Acoustic Emission Signal Of The Wind Turbine Gearbox Bearing Faults

Due to the serious situation of energy crises and environmental pollution, wind energy has been attached great importance home and abroad as a pollution-free and renewable energy. As the installed capacity of wind power becoming bigger and bigger, the fault occurrence rate of wind turbine gearbox becomes higher and higher, seriously influencing the utilization rate of wind power. Rolling bearing is the part in the wind turbine gearbox with higher fault rate, it may result in serious accidents if the bearing’s fault is serious. Due to the rolling bearing’s producing of acoustic emission signal in fault forming period of the beginning and development, adopting acoustic emission technology in the early fault diagnosis of rolling bearing has great significant value on avoiding serious accidents, reducing operating and maintenance costs.The research purpose is to discover a new method of feature extraction and fault diagnosis of bearing’s acoustic emission signal in wind turbine gearbox, in order to solve the problems existing in traditional method such as weak anti-noise interference capability, complex parameter selection, and hard-recognizing fuzzy sample etc., so as to improve the accuracy of feature extraction and the correct rate of fault diagnosis.The main research contents are summarized as follows:Firstly, when the acoustic emission signals are collected from the bearing’s acoustic emission signal in gearbox of wind turbines, there are multiple fault source signals’ compound problem in single channel. In view of this problem, a method of single channel blind source separation which based on the Ensemble Empirical Mode Decomposition(EEMD) and improved Fast ICA algorithm is proposed. This method decomposes one-dimensional single channel of compound fault of acoustic emission signal into mulit-dimensional Intrinsic Mode Function(IMF), then the same number of input signals are built according to the estimated number of source signal, finally separating it by improved Fast ICA algorithm. This method has resolved the under-determined problem of blind source separation in one-dimensional single channel and overcome the disadvantage of sensitive initial value, effectively separating the acoustic emission signals of complex faults(damage and fracture).Secondly, in view of the non-stationary and uncertainty of the bearing’s acoustic emission signals in wind turbine gearbox, a method of feature extraction based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN) and cloud model characteristic entropy is proposed. The method uses CEEMDAN in decomposing signal into IMFs. This method firstly selects sensitive IMF components by correlation coefficient method, then reconstructs the signal by sensitive IMF, finally reconstructed signals’ cloud model characteristic entropy as the characteristic parameters of signal is calculated by backward cloud generator. Through the experimental testing and analyzing, this method can not only effectively extract the feature of acoustic emission signal, but also can overcome the disadvantages of complex parameter selection and sensitive threshold valuing etc.in tradition entropy.Thirdly, in order to solve the problem of strong noise interference existing in the feature extraction of bearing’s acoustic emission signal in wind turbine gearbox, a method of feature extraction based on improved EEMD algorithm and Partial Mean of Multi-scale Permutation Entropy(PMMPE) is proposed. This method firstly selects sensitive IMF components by Cloud similarity measure method, then reconstructs the signal by sensitive IMF, finally calculates reconstructed signals’ PMMPE as the characteristic parameters of signal. This method has overcome the disadvantage of the misjudgment in selecting sensitive IMFs existing in traditional method, reduced the noise interference, so as to improve the accuracy of feature extraction.Finally, in order to solve the problem of the samples with uncertain factors which affect the correct rate of bearing’s fault diagnosis in wind turbine gearbox, a fault diagnosis method of Fuzzy Support Vector Machine(FSVM) based on Certainty Degree of Multidimensional Cloud Model is proposed. This method has utilized the Certainty Degree of Multidimensional Cloud Model as the membership degree of FSVM algorithm and overcome the disadvantage that it is impossible to identify the noise and wild value from the effective samples by traditional FSVM algorithm. Verifying it by the bearing’s fault acoustic emission data of the wind turbines gearbox, the results show that this method can inhibit the interference of uncertain information(noise or wild value) with higher performance in fault diagnosis.