Research On Intelligent Fault Diagnosis Method For Gearbox Of Large Wind Turbine Based On Compressed Sensing

With the industrialization process in the 21 st century,wind power technology has gradually developed.The installed capacity and quantity of the wind turbine have increased,and the requirements for condition monitoring of the wind turbine have become more stringent.In the event of a failure of the wind turbine,the wind turbine is shut down or even the wind turbine is destroyed.In order to prevent dangerous events from occurring,wind turbine condition monitoring and fault diagnosis are required.This thesis summarized and studied the intelligent fault diagnosis methods for the vulnerable parts bearing and gearbox during the operation of wind turbines.The main contributions are as follows:(1)This thesis introduced the wind turbines' gearbox and bearing structure,analyzed the research status of wind turbine fault diagnosis at home and abroad,and summarized many common fault types and fault diagnosis models' advantages and disadvantages of large wind turbine gearboxes;(2)proposed an EMD-BPDN model to improve the EMD noise reduction method.It selects the appropriate IMF component and reconstructs the noise reduction signal by using the compressed sensing and its reconstruction algorithm,which proves that its noise reduction effect is better than other models.A new time-frequency compress sensing feature after compression is proposed.It combines time and frequency characteristics for fault diagnosis,laying the foundation for the diagnosis of intelligent models;(3)Aiming at the vulnerable parts bearing,this thesis proposed aPSO-AdaboostSVM bearing fault diagnosis model based on improved support vector machine with integrated algorithm.It used improved particle swarm optimization algorithm to optimize kernel function parameters and integration ideas,which improved classification performance;Aiming at the vulnerable parts gearbox gears,this thesis proposed aa GA Random Forest gear diagnosis model based on the improved decision tree.It optimized the size and characteristic properties of trees in random forests by genetic algorithm,PCA reduced feature quantities,and decision trees was integrated into random forests.Which improved the accuracy of the decision tree,The experimental data was applied to different models.The fault recognition rates of bearings and gears were 98% and 94% better than other models.Also validated the validity of the model in real wind turbine data(4)The thesis proposed the design scheme of the large-scale wind farm fault diagnosis system and designed a wind turbine gearbox experimental platform to verify the validity of the proposed model.