Research On Wind Turbine Blade Damage Detection Based On Computer Vision

As one of the most widely used new energy sources in today’s society,the utilization of wind energy resources is mainly through wind turbines to achieve wind energy capture and conversion to electricity,and ultimately wind energy is used in the form of electricity for human life and social production.As the only component to capture wind energy,the health status of blades in wind turbines is the basis of wind energy utilization efficiency.However,most wind turbines work in natural environments such as plateaus and mountains.The surface of the blades is in direct contact with the complex and harsh environment and is affected by the alternating load for a long time.Wind turbine blades often have cracks or even fractures due to the above factors.The health status of blades is closely related to power generation efficiency,production safety,and economic benefits.Therefore,detecting damage to wind turbine blades is of great significance.Aiming at the problems of low efficiency and numerous limitations in the detection of wind turbine blade damage,this thesis deeply analyzes the characteristics of non-destructive detection methods such as manual inspection methods and infrared detection.Combined with image processing technology and the widespread use of drones,a new idea of wind turbine blade damage image detection based on deep learning is proposed.Through the in-depth study of blade damage image features,image edge segmentation technology,transfer learning,and convolutional neural network,an image edge detection method based on the Canny algorithm is designed and improved.A residual network model based on transfer learning and random forest algorithm is established to train the blade damage image to complete the detection of leaf health status and damage identification.The performance of the model proposed in this thesis is verified by simulation experiments.The model has good performance in the rapidity of the detection process and the accuracy of the results.It effectively solves the problems of safety,efficiency,and accuracy of wind turbine blade surface damage detection.The main contents of this thesis are as follows:1.The background and significance of wind turbine blade detection research are introduced.The common methods and research status of blade damage detection are summarized.The characteristics of various detection methods are analyzed.Combined with the increasingly mature computer image processing technology and the wide application of UAVs,the advantages of deep learning image recognition technology in blade damage detection are analyzed.2.According to the characteristics of the blade damage image.The image processing techniques such as image filtering,image enhancement,and image segmentation are analyzed,and the edge detection algorithm of wind turbine blade damage image based on the Sobel operator and Otsu threshold segmentation is designed.This algorithm is used to preprocess the blade damage image to reduce the influence of the image background on the detection results and highlight the blade-damaged area.3.Based on the successful application of deep learning in image classification,the Res Net model in the convolutional neural network is applied to the detection of blade surface damage.The problems of insufficient training data and low sensitivity in image detection and classification are deeply studied.An improved ResNet blade damage image detection model is proposed,which adds transfer learning and random forest to the Res Net model.Through transfer learning,the network is divided into a replacement layer and a retention layer,which improves the ability of the model to extract features and reduces the model operation data.A classifier based on the random forest is constructed to improve the model’s ability to classify blade image features.The superiority of the improved blade damage image detection and recognition model is verified by experimental comparison.4.The framework of blade damage image detection based on computer vision is proposed,including multi-scale image edge detection and an improved Res Net model.Compared with the support vector machine,the experimental results show that the proposed damage detection model framework has higher accuracy and faster detection time.