Prediction Method For Icing Mass Of Wind Turbine Blade Based On Modal Frequency

The wind turbine blades installed in cold regions are subjected to icing problems inevitably,which will affect the aerodynamic performance,structural dynamics and other characteristics of wind turbine blades.It will increase the operation risk of wind turbine and reduce the utilization rate of wind energy.The different icing location and icing mass will lead to different natural frequency changes.Therefore,the thesis uses the artificial neural network technology to achieve the purpose of predicting the icing location and icing mass of the blade based on the natural frequency of the iced blade.The ice accumulated on the tip,middle,root or the whole blade of a 1k W wind turbine respectively.The natural environment icing experiment of the blade and iced blade modal experiment are carried out respectively.The relationships between the natural frequency change rate of the iced blade in the four cases and the icing mass are constructed,which uses two hypothesis methods.The BP and RBF neural networks are trained with the data set and their prediction ability is analyzed.Finally,the 2MW large iced wind turbine blades are analyzed using the prediction method based on the finite element modal analysis technology.The results show that:(1)The model of 1k W wind turbine blade is established by Faro 3D laser scanner and Solid Works software,and the modal of blade is carried out by the finite element modal analysis.The results are compared with the modal parameters measured by B&K modal test system.The errors of the first five natural frequencies are within 10% and the modal shapes are consistent.(2)Considering the influence of icing at different location on the natural frequencies of all iced blades,the relationships between the change rate of natural frequencies of all iced blades and the icing mass of tip,middle and root of 1k W wind turbine blades are constructed,The relative error of the relationships is smaller.The relationship is used to train BP and RBF neural networks,and the relative error rate of predicting the total icing mass of blades is within 10.5%.(3)When the actual icing mass of the blade is smaller,the relative error rate of the predicted icing mass is larger,and the relative error rate decreases with the increase of the actual icing mass.The relative error rate of BP neural network is lower than that of RBF neural network,which is more suitable for this case.The average error of BP neural network is 5.53%.(4)The 2MW large iced wind turbine blades are analyzed by using the finite element modal analysis.The relative error rate between the predicted icing mass and the actual icing mass is 13.21% by using BP neural network.Compared with Gantasala’s works that used the same neural network to predict the icing quality of 5MW blade,the prediction accuracy is improved by 1.56%.Therefore,the artificial neural network technology can be used to predict the icing location and icing mass of the blade by using the natural frequency change rate of the icing blade,so that the deicing system can work more accurately,quickly and efficiently.