Research On Combined Fault Diagnosis Of Misalignment And Broken Tooth Gear Fault Of Doubly-fed Wind Turbines

With the increasing demand for environmental protection,green and clean renewable energy is gradually replacing traditional fossil energy,and wind energy is one of the main forces of renewable energy.Wind turbines are located in remote areas with abundant wind resources,and their working conditions are complex and changeable.Compared with conventional machinery,they are prone to failure,and the probability of combined failure is higher.In order to reduce the operation and maintenance cost of wind turbines and ensure the safe and stable operation of them,combined fault diagnosis of wind turbines is studied.At present,most wind turbine combined fault diagnosis methods are based on manual time frequency domain analysis,and the fault diagnosis methods using machine learning algorithm usually only consider one single fault situation.In this dissertation,machine learning method is used to study the two common faults:misalignment and broken tooth gear.First,the vibration characteristics of misalignment fault and broken tooth fault are analyzed.According to the operation principle of doubly-fed wind turbine,the wind turbine drivetrain fault experimental test rig is built.Experiments using the test rig to simulate normal,misalignment fault,broken tooth gear fault and the combined fault are designed,the vibration data at various speeds are collected.Next,the support vector machine(SVM)model is studied,and grey wolf optimizer(GWO)is introduced into the process of supporting vector machine hyperparameter tuning.The experimental results on several data sets show that GWO is superior to particle swarm optimization on hyperparameter tuning.GWO hyperparameter tuned SVM has better stability,and has good performance on the vibration simulation data set of wind turbine misalignment faults.Therefore,GWO hyperparameter tuned SVM is selected as the diagnosis model.Then,the theory of higher-order cumulant images is introduced into wind turbine combined fault diagnosis,and the diagnosis effect of wind turbines using different features based on higher-order cumulant is studied.A fault diagnosis method using higher-order cumulant grayscale images texture features(contrast,inverse different moment and correlation of gray-level co-occurrence matrices generated from the cumulant grayscale images in 0,45,90 degrees directions)and mean in time domain is proposed.Compared with the direct use of high-order cumulants,the method improves the generalization ability of the diagnostic model,the accuracy of diagnosis,and also false alarm rate and miss rate.Last,based on the empirical wavelet transformation,low-pass filtering empirical wavelet transformation(LPFEWT)is proposed to extract features in time-frequency domain,which is proved better than the empirical wavelet transformation and empirical mode decomposition.LPFEWT can eliminate the confusion phenomenon of diagnosis.The influence of the number of LPFEWT Fourier spectrum segment on the diagnostic results is explored,and a specific selection method of it is given.After feature selection among common time domain features,frequency domain features and the time-frequency domain features extracted by LPFEWT,a combined fault diagnosis method of wind turbines using energies of LPFEWT components and margin factor is proposed,which has high accuracy,low false alarm rate and low miss rate.The SVM was compared with other machine learning algorithms,which proves the superiority of SVM in this diagnosis problem.In this dissertation,two effective fault diagnosis methods of wind turbine combined fault of misalignment and broken tooth gear fault based on machine learning are proposed.Compared with manual time-frequency domain analysis of signals to identify faults from the spectrum,it reduces the dependence on human experience,omits the process of calculating fault frequency,and improves the diagnosis speed.In these two methods,the fault diagnosis method based on LPFEWT is recommended,which has low computational cost and high accuracy.