Fault Diagnosis Approaches To The Wind Turbine Planetary Gearbox Based On Support Vector Machine

Wind power has developed rapidly in recent years.It is not only highly valued by all countries in the world,but also is the fastest growing new energy in China.Because of the harsh working environment,wind turbines are prone to failures,and planetary gearbox is the component with the highest failure frequency.Because of its special physical structure,the general fault diagnosis methods are difficult to apply to practical engineering,and its fault diagnosis has become an urgent technical problem to be solved.Relying on the wind turbine drivetrain diagnosis simulator,this thesis designs the fault diagnosis experiments of planetary gearbox,collects experimental data,analyzes the characteristics of vibration signal and extracts fault feature.These works establish foundation for the following fault diagnosis researches.According to the characteristics and difficulties of planetary gearbox,its fault diagnosis is studied based on support vector machine(SVM)in this thesis.The main contents include:Firstly,to selecting the Gauss Kernel parameter of SVM reasonably,an improved particle swarm optimization(PSO)algorithm is proposed to optimize the parameters,and compares it with other methods.Then,planetary gearbox fault diagnosis is studied based on SVM with reasonable parameters.Secondly,there are various fault types,but SVM can only complete binary classification problem.An improved directed acyclic graph support vector machine(DAG-SVM)algorithm is proposed to solve multi-classification problem.This algorithm can effectively overcame the shortcomings of traditional DAG-SVM,and a decision structure with high classification accuracy is selected.The experimental results of planetary gearbox show that the proposed algorithm is effective.Finally,this improved DAG-SVM algorithm can not complete the fault diagnosis task when the unknown fault types occur.Therefore,an improved support vector data description(SVDD)is proposed.This method separately establish Hyper-Spheres for each class sample,new sample type is consistent with Hyper-Sphere type.When the new sample does not belong to any known Hyper-Sphere,new fault type is identified and new Hyper-Sphere of new type is constructed.By adding classified Hyper-Spheres iteratively,the diagnosis performance of classification model is constantly improved.The experimental results show that the method is effective.