Research On Image Recognition Method Of Wind Turbine Blade Surface Damage

The research on the method of wind turbine blade surface damage is of great significance to improve the efficiency of wind turbine fault diagnosis and reduce the risk of wind turbine operation.When inspecting the surface of the wind turbine blade,visual inspection method,depending on manual inspection,may cause visual fatigue.But its results and efficiency can be affected by subjective consciousness of inspectors.Therefore,in order to improve limitations of manual inspection,the wind power plant urgently needs to apply the damage image recognition technology to actual sites.For wind power plants,however,long-distance inspection and dynamic environment will cause difficulties in obtaining damage images,which makes the identification of the damage image of the wind turbine blade surface more difficult.Taking the surface damage of wind turbine blades as the object of inspection and focusing on the limitations of the surface damage inspection of wind turbine blades,this paper aims to visually identify and inspect the surface damage of the wind turbine blades by the research content of the advanced machine learning visual inspection method.The main research of this paper is as follows.1.Three kinds of image acquisition schemes for obtaining the surface of wind turbine blades with manual inspection are proposed.Taking a feasible ground inspection robot as the hypothetical image acquisition scheme,the three schemes are demonstrated to determine technical specifications that are applied to the image acquisition and damage classification.2.An image data set,which is suitable for image identification of wind turbine blade surface damage,is established.It includes zero damage images,crack damage images and other damage images on the wind turbine blade surface.3.Based on the TensorFlow learning framework,a network model of image identification classification experiment task was built on Python 3.5.(1)Based on back propagation(BP)neural network,the experiment of the accuracy of classification ofwind turbine blade surface classification is realized.(2)Based on convolutional neural network(CNN),the accuracy of the damage inspection of the surface damage of the wind turbine blade is tested.(3)Based on full convolutional neural network(FCN),the accuracy of crack damage and other types of damage is inspected respectively among the damaged surface of wind turbine blades.The experimental results of test images with wind turbine blade surface damage were analyzed and evaluated.4.The visual inspection method of wind turbine blade surface damage adopted in this paper has achieved good inspection results.The average classification accuracy and precision of BP neural network on the surface crack damage of wind turbine blade and other types of damage is 76.7%.The average classification accuracy and precision of CNN on damage and zero damage of the wind turbine blade surface is 92.41%.The average classification accuracy and precision of FCN on the surface of the wind turbine blade without damage,crack damage and other types of damage is 92.63%.This paper established the image identification data set of wind turbine blade surface damage,specified the image acquisition scheme and specification of technical classification,and realized the image identification inspection of wind turbine blade surface damage.This paper will provide a method for the application of image identification algorithms to the field of wind turbine blade surface inspection.