Research On Surface Damage Analysis Of Wind Turbine Blade Based On Image Recognition

Wind turbines are generally installed in the higher altitude of the mountain area,the working environment is harsh,the blade is exposed to the outside for a long time is easy to produce cracks,coating damage,oil stains,pitting and icing and other surface defects,if the blade can not be repaired in time,over time,the blade may break and cause a large accident,so it is necessary to detect defects on the surface of the fan blade to prevent accidents.At present,the defect detection of wind turbine blades basically relies on human observation,and human inspection not only has the problem of low efficiency and high cost,but this method is often prone to misjudgment.In this paper,the UAV is used to collect pictures and videos of wind turbine blade surface damage,and the machine vision field combined with the algorithm in the field of deep learning is used to process the wind turbine blade surface damage pictures to achieve the purpose of automatic and accurate classification of pictures.Firstly,the software and hardware of the fan blade image acquisition device and the image processing device were designed and installed.The M210 RTK UAV is equipped with a Zenmuse Z30 gimbal camera to collect pictures and videos,and after screening,a four-damage type dataset is constructed,and the images are enhanced to expand the data set to ensure the number of pictures and the robustness of the network model after training.Secondly,with the help of the traditional image processing method in the field of machine vision,the image is selected from the series of algorithmic processes of grayscale processing,filtering processing,binary segmentation processing,morphological processing and connected domain marker processing.It provides a guarantee for the accurate classification of leaf damage pictures using deep learning in the future.Finally,using the deep learning object detection algorithm to accurately classify the processed images,the three convolutional neural networks with high classification accuracy,VGG16,Google Net and Res Net50,are selected to classify the above-mentioned processed images.By comparing the accuracy rate and loss value curve of each network model after multiple iterations,the accuracy rates of VGG16,Google Net and Res Net50 verification sets were 85.98%,89.02%,and 92.07%,respectively,and the network model of the convolutional neural network with the highest accuracy and the smallest loss value was initially selected as the network model for image classification.In order to improve the performance of the network model,the Res Net50 network optimized by Adam and RAdam optimizer is further compared,and the classification results of the two optimization models are compared,and the RAdam optimization network model is determined to have higher accuracy and faster convergence speed.In this paper,some leaf damage pictures are collected from the website and the wind field,and the accuracy of each class is verified by the confusion matrix,and the accuracy of the final classification result is 95.83%,which meets the requirements of image classification.The paper finally selects the RAdam optimized Res Net50 network to classify and test the surface damage of the wind turbine blades,which can achieve rapid and accurate classification of the pictures.