Research On Fault Early Warning Method For Wind Turbine Based On Condition Identification

In recent years,the rapid growth of wind power industry has led to a series of issues including frequent failures which results in the consequent expensive operation and maintenance costs.Strengthening the effective monitoring of the operating state of wind turbine is so important to ensure the safe and stable operation of the unit and reduce the maintenance expenditure.At the same time,the changeability and complexity of large wind turbine operating conditions make it difficult to accurately evaluate the operating state.Therefore,on-line monitoring of wind turbine based on condition identification has become an important research d irection in wind power development.This paper first carries on the analysis and identification of wind turbine operating conditions,then starting from two kinds of data sources of SCADA parameters and vibration parameters and taking wind turbine gearbox as the research object,carries out the research work about fault warning and fault diagnosis respectively based on the multivariate state estimation model and the variational mode decomposition method.The main contents are as follows:Firstly,after analyzing the operating characteristics of wind turbine and selecting appropriate parameters from the SCADA data system,p artition the operational space based on fuzzy C-means clustering algorithm.Secondly,the early fault warning of wind turbine gearbox based on multivariate state estimation is studied,the typical faults of the gearbox are analyzed,the principle of MSET model is introduced in detail,selection of the model variables is researched.The proposed model is verified by actual wind turbine dataset and concludes that condition identification can reduce the false alarm rate.Finally,the fault diagnosis of rolling bearing based on the variational mode decomposition is studied.Aiming at the problem that the early faul t signal of rolling bearing is too weak to extract effective information,a feature extraction method based on VMD,AR model and singular value decomposition is proposed.Simulating actual data to validate the superiority of VMD in suppressing the mode mixing compared with EMD.Aiming at high dimensions of feature vectors in classification of vibration signal,a classification method based on FCM and KPCA is put out to reduce the dimensions of data processing to improve the validity of classification method.Through simulation test analysis with laboratory data,the result points that the different working conditions obviously affect the accuracy of fault diagnosis.