Research On Fault Diagnosis Of Rolling Bearing Of Wind Turbine Based On HHT And XGBoost

With environmental pollution becoming increasingly serious and policy-driven,the wind power industry has received great attention of the world.Along with the rapid development of the wind power industry,the demand for its reliability is getting higher and higher.Rolling bearings are the key components of wind turbines and it is also the highest fault rate components.Therefore,real-time monitoring and fault diagnosis for the rolling bearings can not only make the entire wind turbine more reliable,but also reduce the costs of operation and maintenance.In this paper,based on the analysis of the characteristics of rolling bearing's faults and its development process,by comparing the commonly used signal analysis methods,the Hilbert-Huang transform is selected as the feature extraction method of the vibration signal of the rolling bearing.And the energy value histogram is used to dimension reduced for eigenvalues.The extracted eigenvalues are used as the input of classifier,and the XGBoost algorithm is used to identify various faults of rolling bearings.The main work of this paper is as follows:(1)A detailed analysis of the feature extraction method of the vibration signal of the rolling bearing was carried out.And the simulation method was used as an example to test the proposed method.So that viewed the feature extraction process directly.In order not to destroy the structure of the extracted eigenvalues,the energy histogram is used instead of the time-frequency spectrum to reduce the dimension.Taking the experimental data of rolling bearing as an example,an experimental comparison of this paper and the other feature extraction methods is made to prove the advantage of Hilbert's yellow transformation method in the feature extraction of vibration signals of rolling bearings.(2)Aiming at the problem of choosing intrinsic mode function which decomposed from the empirical mode in Hilbert-Huang transform,a method which combination the correlation coefficient and relative entropy is proposed.Because the correlation coefficient method is prone to misjudgment near the set threshold,and the relative entropy method is poorly adaptable.So no mater what method used it is easily lead the misjudgment of IMF component.This article combines the two methods to avoid their own shortcomings.Through simulation experiments,it is proved that this method is feasible.(3)By comparing the methods of common machine learning,XGBoost is used as a rolling bearing fault classification algorithm.In order to reduce the hyperparameter optimization time of the algorithm and avoid the local hyperbola of the selected hyperparameter,this paper proposes a hyperparameter optimization method which based on the message queue.Comparison the effect to commonly used classification algorithms and the parameters adjustment methods for the vibration data of rolling bearings,it is concluded that the classification algorithm used in this paper has the highest accuracy,and the tuning parameters method can not only find the global optimal solution,but also save the time spent by the optimization process remarkable.(4)Using the data of the wind turbine main bearing and the k-fold cross validation method made an experiment in this paper,and the corresponding time-frequency distribution of bearing at different locations is obtained.At the same time validates the feature extraction method and classification used in this paper is effective for fault diagnosis of rolling-element bearings in wind turbines,and the diagnosis results are ideal.