Intelligent Acoustic Monitoring System For Wind Turbine Blade Health

Wind turbine blade is a crucial part of wind turbine for driving motors to generate electricity.Due to the long-term operation in harsh environment,the blade is prone to fracture,lightning,ice and other faults,resulting in significant economic losses and even casualties.In view of shortcomings of existing detection methods and monitoring systems,this paper comprehensively uses acoustic fault diagnosis,machine learning technologies to research and design intelligent acoustic monitoring system of wind turbine blade health.The specific research contents and achievements include:(1)According to intelligent monitoring requirements of large-scale wind turbine blades,the overall framework and implementation scheme are presented.The intelligent system consists acoustic diagnosis node,network message middleware and intelligent analysis platform.Moreover,this system has rich intelligent functions such as acoustic fault diagnosis,online optimization of recognition model,fault trend prediction,remote monitoring,controlling node and so on.(2)Aiming at improving recognition accuracy of the initial SVDD diagnosis model for new samples,an on-line incremental learning method based on density reduction is proposed.Firstly,the sample subset that violates the KKT condition is selected from incremental sample set,and the overall aggregation degree of sample subset is subsequently evaluated.Secondly,density value of the remaining sample subset is calculated according to information entropy,and the remaining sample set is filtered again with the proportion control factor.Finally,the reduced incremental sample subset,support vectors of the initial model and samples near the boundary are together used to train incremental recognition model.Simulation results show that compared with the CISVDD and kmeans-SVDD,the DRISVDD greatly reduces the training time.(3)Combined with the SCADA data of wind turbine,blade fault trend prediction model based deep learning is proposed.On a basis of the correlation analysis between multi-dimensional features,new features are constructed to describe the state of wind turbine blade.By optimizing parameters such as learning rate,sliding step and time window length,fault trend prediction model based on Long Short-Term Memory(LSTM)network is established.Finally,the validity of proposed method is verified by the open icing prediction data set of wind turbine blade.Simulations have been carried out to show that purposed LSTM model is superior to the RNN,SAE and XGBOOST.Finally,integrating the incremental SVDD,LSTM trend prediction and shallow data analysis algorithms,the intelligent acoustic diagnosis system for blade health is designed and then verified by measured data from wind farms.The research results have significant value for engineering applications.