Research On Intelligent Inspection Method Of Wind Turbine Blade Based On UAV Image Acquisition Technology

Wind turbine blade is the key component of wind turbine to capture wind energy.Due to the long-term harsh outdoor working environment,its surface will inevitably produce all kinds of damage.If it is not found and repaired in time,it will lead to serious failure consequences.At present,the research of wind turbine blade damage detection in China mainly focuses on the analysis of sensor signals to detect the health status of wind turbine blades.This analysis method is limited by the problems of sensor installation,data storage and transmission,and the signal obtained by the sensor is easily affected by the external environment,Moreover,the installation of a large number of sensors on the blade will also lead to the complex structure of the blade and affect the performance of the blade to capture wind energy.In this context,aiming at the task of wind turbine blade damage detection,this paper uses the technology of unmanned aerial vehicle(UAV)image acquisition to capture the image data of wind turbine blade,and then uses the method based on deep learning to detect the damage of wind turbine blade.The main work of this paper is as follows;(1)In this paper,UAV is used to inspect the fan blades,and the images captured by UAV are used to train the neural network model.The images captured by UAV are preprocessed,and the images with damage are selected.According to the number of damage type images and damage intensity,three types of damage are determined: crack,surface wear or fall off and lightning strike.The data are expanded by using conventional and mosaic image enhancement technology respectively.Labelimg labeling software was used to label the data,and the training set,verification set and test set needed for neural network model training were made.(2)In this paper,yolov5(you only look once)algorithm is applied to the task of wind turbine blade damage detection for the first time.The limitation of yolov5 algorithm in blade damage detection is improved.In view of the poor detection effect of adjacent damage occlusion in the experimental results,the non maximum suppression(NMS)of the output prediction part of the model is improved to Diou＿nms(Distance-Io U)。 In view of the slow convergence speed and low accuracy of the regression loss function,four regression loss functions are analyzed and compared,and the regression loss function giou(generalized IOU)is improved to CIO(complete IOU).After the improvement,the experimental results show that the fan blade damage detection based on yolov5 model achieves 96.1% map(mean average precision)on three types of damage,The average detection time of one image is 7 ms,which meets the real-time requirements,and realizes the identification and location of fan blade damage type.(3)According to the trained neural network model,the wind turbine blade health detection system is developed based on LabVIEW platform.The automatic detection module of the system uses the trained yolov5 model to realize the visual operation of the damage detection process of the fan blade.