Research On Wind Generation Recurring Technologies Dedicated To Condition Monitoring And Fault Diagnosis

Condition monitoring and fault diagnosis is a crucial and essential technical method in the early fault detection for the wind turbine systems.In the guidance of the condition monitoring and fault diagnosis,the preventive maintenance is scheduled to ensure the safety and health of the wind turbine systems,decreasing the operational and maintenance cost and promoting the economic benefits.Due to the volatility of the natural wind,the wind turbine gearbox is running under the varying rotational speed conditions to track the maximum power point of the wind energy.Hence the vibration signals acquired from wind turbine gearbox show the typical non-stationary characteristic on the influence of the varying condition,which leads to the difficulty of the condition monitoring and fault diagnosis.Funded by the National Natural Science Foundation of China(Grant No.51275453),a scaled-down wind turbine simulator,which is oriented for the researches of the condition monitoring and fault diagnosis,is developed based on the similarity criterion.The dynamic modeling of the wind turbine system,optimal electrical control-strategies,the processing method of vibration signals in the variational mode decomposition and the multi-information fusion diagnosis in support vector machine are studied deeply in this dissertation.The main contents of this dissertation are as follows:In the Chapter 1,the scientific backgrounds and significance of this thesis are presented.Then,state-of-art for several key technologies concerning this field are elaborated.The structure and main contents of this dissertation are depicted.In the Chapter 2,a scaled-down wind turbine simulator is developed on an electricity circulation test-rig.The corresponding software and hardware are summarized in brief.The similarity principle of the scaled-down simulator is analyzed based on the three laws of similitude theory.In addition,dynamic model of the simulator system is dicussed.In the Chapter 3,the recursive least square algorithm is used to identify the motor's parameters on line in eliminating the flux coupling terms.Moreover,a control strategy for the three-phase rectifier module is imported,which contains a current decoupling loop without the inductance parameter.Additionally,feed-forward compensation control strategy based on the load-current observer is adopted to improve the anti-interference of the DC bus.The experiments verifies the validity of the control strategies.In the Chapter 4,a wind model based on the wind turbulence combining the probability density distribution theory is built up.The wind-shear and tower-shadow effects is considered in the equivalent wind model for the aerodynamic characteristics of the wind turbine.Then control strategies of the double fed induction wind turbines is studied.Experimental results demonstrate that the proposed recurring replication method is effective and accurate in terms of both static and dynamic performances.In the Chapter 5,a novel signal processing method by using the variational mode decomposition(VMD)and Renyi entropy is proposed to evaluate the bearing health of the wind turbine gearbox.The health evaluation method,which is called as "VMD-Renyi",is based on a performance review over a range of operating conditions,rather than at a certain single operating condition.Experimental investigations were performed which verified the efficiency of the evaluation method,as well as a comparison with the previous method.In the Chapter 6,an intelligent diagnosis method based on the multi-information fusion with support vector machine(SVM)is proposed.The multi-classification by SVM is constructed through the signal preprocessing,characteristics selection and uniformization.Four SVM classifications are constructed by selecting the non-dimensional time-domain feature,the wavelet packets relative-energy feature,the VMD relative-energy feature and the multi-information fusion feature.The performances of the four different SVM classification are studied and compared by using the confusion matrix,ROC curve and the space mapping set.Moreover,the penalty factor and the RBF kernel value of the SVM is optimized by the twice mesh search method.In Chapter 7,the conclusions and prospects are briefly depicted at the end of this thesis.