Wind Turbine Blade Surface Damage Detection Based On Convolutional Neural Network

Wind energy as a new energy,it has the characteristics of sustainable utilization,and meet the sustainable development strategy,so it has been vigorously developed.With the increase of wind farm installed capacity,the fan blade is also increased.As the core component of the wind turbine,the fan blade has high cost.Moreover,because of the low efficiency,strong subjectivity and easy visual fatigue of manual inspection of the fan blade,it is easy for the wind turbine blade to be inspected.These factors make part of the surface damage can not be effectively identified.Wind farm accidents occur every year.Therefore,an automatic and accurate detection method is urgently needed for blade surface damage.Therefore,this paper provides a scheme for the image acquisition of fan blades in actual wind farms,and by combining with the image processing technology,it can realize the precise classification and recognition of the surface damage of fan blades.Firstly,the surface damage image of the fan blade is obtained.Acquisition of wind turbine blade surface damage images using UAVs.According to the actual wind farm needs,select the UAV model and plan the UAV acquisition route.In addition,the image collected by UAV is used as the original data of this subject.The original data obtained were screened and size normalized.For fuzzy images,histogram equalization is carried out to make surface damage images better adapt to the neural network.Secondly,the surface damage images obtained after size normalization and histogram equalization are filtered,grayed out,binarized,morphologically processed and masked by combining digital image processing techniques to segment the wind turbine blade and background images.The segmented images are combined with data enhancement techniques to finally build a damage dataset containing three types of wind turbine blade surface damage,including 1016 crack damage images,496 coating peeling damage images and 800 no damage images,totaling 2312 images,and they are randomly assigned to the training set,test set and validation set in the ratio of 9:0.5:0.5,which are used as input of the neural network.Finally,the study of leaf surface damage classification recognition based on convolutional neural networks was conducted.By analyzing three convolutional neural network structures and conducting comparative analysis of different models through the prepared data set,and then select the most suitable for a network model of the blade surface damage classification and recognition,and the network model parameters optimization,evaluating the performance of the model under different parameters,the final test results show that the surface damage of the fan blade average recognition rate is 97.04%,therefore,meet the needs of wind farm actual checking for damage identification accuracy.