Research On Feature Extraction Methods Of Vibration Signals For Diagnosis Of Wind Turbines Bearings

With the rapid development of wind power industry, the reliability andmaintainability of the wind turbine system are very urgent. Condition monitoring and faultdiagnosis technology of wind turbine system cause the extensive concern of the academiaand industry. Bearings are as the core components of wind turbine mechanicaltransmission system and generator system, there has a realistic significance to make acondition monitoring and fault diagnosis to them. In this paper, aimed at feature extractionmethods of wind turbine bearing diagnosis, applied Local Mean Decomposition (LMD),Shannon entropy and nonlinear dynamic parameters, in view of the transientcharacteristics description and nonlinear feature analysis research, the proposed methodsare verified by simulation experiment and experimental platform. The proposed methodsprovide a solution for wind turbine bearing condition monitoring and fault diagnosis. Thespecific research ways are as follows:Firstly, the wind turbine bearing operation characteristics and the failure mechanismare discussed, besides, aiming at nonstationary and nonlinear characteristics of bearings,the transient signal decomposition technique based on LMD are studied; the quantitativedescription method based on information entropy are analyzed. Both of that is in order toeffective extraction and accurate description of wind turbine bearing vibratory signals.Secondly, a transient characteristic extraction method based on LMD andWigner-Ville spectral entropy is proposed, in order to quantitatively describe thetime-frequency energy distribution of bearing vibratory signals under different condition.After that, a intelligent fault diagnosis model based on LS-SVM is used for automaticclassification and recognition of bearing faults. Simulation experiment and experimentalplatform verified the proposed method and diagnosis model.Finally, in view of nonlinear dynamics, a nonlinear feature extraction method nameda multi-scale permutation entropy based on LMD is proposed. The proposed method caneffectively represent nonlinear complexity characteristics of bearing vibratory signals andidentify different fault degree of bearing. Simulation experiment and experimental platform verified the proposed method.