Research On Method For Extraction Of Fault Signal And Remaining Life Prediction Of Gearbox In Offshore Wind Turbine

Environmental degradation and energy shortages are getting worse,we need to develop some clean energy forms.Wind energy,a clean renewable energy,is easier to form an industrial scale than other conventional energy forms.Now it plays a central role in more and more national long-term energy plans.Wind turbine is a complex electromechanical system composed of electrical system,hydraulic system,control system,transmission system,etc.They often work in extremely harsh environments,and the rapid changes in ambient temperature,air pressure,and alternating load conditions cause different types of failures in wind turbines.If the gearbox fails,it will cause huge production costs.We can use scientific and reliable technical means to predict the remaining life of the wind turbine gearbox,and we can take maintenance and remedial measures before the gearbox fails.The support vector machine is used to construct a gearbox remaining life prediction model.First,the situation of my country's wind industry and wind turbine installed capacity is introduced.Wind turbine status assessment and modeling is introduced.The research on the status assessment and status modeling of wind turbines is divided into two categories.Several research methods for life prediction of mechanical equipment are introduced,calculated and analyzed.They are divided into prediction methods based on physical models,prediction methods based on knowledge,prediction methods based on statistical experience,and prediction methods driven by data tables.Three processing methods of vibration signals are introduced in detail.Then,the SVM is used to train the sample data and use the PSO to globally optimize the model parameters.First,feature extraction is performed on the collected vibration signals.Due to the data redundancy problem between each feature parameter,the modeling process is complicated and the model error is large.The principal component analysis method is used to reduce the dimensionality of the characteristic parameters of the vibration signal in this article,the correlation between the various parameters is reduced to avoid the occurrence of data redundancy.The data after dimensionality reduction is input into the established SVM model to construct a predictive model.The selection of penalty parameters and kernel parameters affects the accuracy and generalization ability of SVM,a method using PSO to optimize the parameters of SVM globally is proposed.Finally,the experiment of the influence of principal component analysis on the prediction accuracy is completed.At the same time,the grid method and manual tuning method are compared with the method developed to verify the superiority of this research.The model constructed by the particle swarm optimization support vector machine for data dimensionality reduction using principal component analysis shows strong advantages in prediction accuracy and generalization ability.