Research On Tracing Method Of Wind Turbine Inspection Based On UAV Vision

The wind turbine may suffer surface or internal damage during operation,which will affect the maximum operation efficiency.The wind turbine blades shall be inspected regularly.In this thesis,the image processing method according to the characteristics of different wind turbine generator components and the visual navigation algorithm of UAV patrol wind turbine generator based on visual servo are proposed to realize the autonomous target detection and tracking in UAV patrol wind turbine generator and the tracking control of patrol UAV.Under the limitation of highaltitude image captured by UAV,the feature recognition and tracking of wind turbine generator components are studied.The tracking control scheme is designed according to the visual information of UAV,and the tracking flight of UAV is controlled by visual servo.The realization and verification of the flight control system are completed based on the simulated physical environment and ROS platform.The main research work of this thesis is as follows:1.Research on the detection and tracking of the characteristics of the main components of the wind turbine generator.Firstly,the wind turbine generator hub is selected as the central feature component,and the image features of the wind turbine generator hub are extracted independently based on the yolov5 deep convolution neural network algorithm.The CBAM attention mechanism integrating channel and spatial information is designed to improve the attention of target features,improve the backbone network of yolov5,and improve the detection efficiency and accuracy.By establishing the hub data set and training the regression network,the wind turbine generator hub detection model is obtained,and the algorithm accuracy is 99.5%,Reasoning speed up to 212 FPS.For the wind turbine tower and blade,the Hough line detection method is used to extract the blade line,and the identification position of the hub is used as the constraint condition to assist the blade line detection.For the blade tip,the corner extraction method based on fast feature is adopted.After completing the detection of the central components,a Kalman filter is designed for the moving target UAV to realize the continuous and stable tracking of the positioning results.The mainstream convolutional neural network is combined with the classical filter algorithm to obtain stable and efficient feature tracking results.2.The tracking method of UAV Based on vision is studied.After identifying the components of the wind turbine generator and positioning them in the image,establish the camera imaging model,analyze the mapping of the three-dimensional coordinates of the physical world in the two-dimensional pixel plane coordinate system,and connect the changes of characteristic parameters in the image with the changes of UAV pose.A new idea of speed and position adjustment of flight control system is provided.Based on the principle of visual servo,according to the requirements of patrol inspection task,the tracking control scheme of phased UAV patrol inspection under different scenes is designed,including the tracking control of initial operation point and blade inspection.Combined with the flight constraints of UAV,the speed control quantity of UAV is calculated.3.Build a three-dimensional simulation physical simulation environment,establish a three-dimensional model of the wind turbine generator,import the UAV and camera sensor model,and develop the flight control system on the ROS platform in combination with the flight controller firmware of the UAV open source platform.Deploy the deep learning algorithm to the ROS environment,design the communication node,take the sensor data of the virtual physical world as the input,output the control message through each module processing algorithm,control the autonomous flight of the UAV,and complete the UAV positioning to the inspection starting point directly in front of the hub and the inspection process of single blade.