Study On The Incipient Fault Diagnosis Of Direct Drive Wind Turbine With Operating State

Xinjiang area is rich in wind resources,and it is an important wind power base in china.In this area,the wind turbine should be subjected to extreme environment such as high temperature,freezing,sandstorm,and so on.The wind turbine generator room is located at 50 ~ 80 m above the altitude,and naturally difficult to maintain.The fault diagnosis results are affected by wind speed,load and grid operation state,and the diagnostic accuracy is not high.Therefore,it is of great significance to study the monitoring and fault diagnosis technology of wind turbine generator in grid connected operation.Aiming at the problem of permanent magnet direct drive wind generator fault monitoring and diagnosis,a new method based on mechanical and electrical comprehensive information is proposed in this paper.This method can diagnose faults in wind speed steady,wind speed step change,three-phase grid voltage unbalanced working conditions,and normal operation,spindle eccentricity fault,spindle bearing wear fault can be detected.The details are as follows:(1)This paper makes a preliminary investigation on the fault condition of wind turbines in Xinjiang,and compares the fault characteristics of wind turbines in Xinjiang and Sweden.The fault types and fault characteristics of spindle eccentricity and bearing wear are analyzed and the influence of the fault on the vibration signal and current signal is analyzed qualitatively.The fault diagnosis methods are summarized,and the advantages and disadvantages of the diagnosis model are compared.Finally,SVM model is selected as the primary diagnostic model.(2)The multi-source fault information acquisition system of wind turbine is developed,and the fault detection test bench of wind turbine under variable working condition is built.Radial vibration and generator stator current signals in wind speed steady,wind speed step change,three-phase grid voltage unbalance operations under normal,spindle eccentricity of spindle,bearing wear cases were obtained.The fault feature extraction method of vibration and current signal is studied,and the fault feature sample library is established.The Mahalanobis distance is introduced to evaluate the separability of fault features.(3)The logistic regression function and the coupling method are used to improve the traditional SVM diagnostic model,so that SVM can output the fault type and fault probability.Fault diagnosis models were established respectively under wind speed steady,wind speed step change,three-phase grid voltage unbalance condition.The extracted vibration and current characteristic parameters were input respectively to train fault models which were built based on vibration characteristics and current characteristics and implement preliminary fault diagnosis by using single information source under various generator working conditions.(4)The traditional D-S evidence theory is improved,and the relationship between the reliability coefficient and the fusion weight is bridged by using the reliability matrix.The generalization ability as well as the diagnosis effect is taken into account,and different weights are given to the preliminary diagnosis results and the fusion diagnosis is carried out.Compared with the preliminary diagnosis method based on a single fault information source,the error diagnosis results are corrected and the diagnosis accuracy is improved obviously after weighted fusion.