Detection And Location Of Fan Blade Defects Based On UAV System

Wind energy is a kind of renewable clean energy,which is widely used all over the world for its environmental protection and pollution-free characteristics.The wind power industry is developing rapidly in China.By the end of 2020,China had built more than4,000 wind farms with a cumulative installed capacity of 220 million k W,ranking first in the world.Blade is the core component of wind turbine to convert wind energy into mechanical energy,which accounts for 15% ～ 25% of the total cost of wind turbine.If the fault of fan blade can not be found and dealt with in time,it will bring huge economic loss to wind farm and seriously affect the normal operation of wind turbine.Therefore,it is very important to detect and locate the fan blade timely and effectively.In this study,a complete wind turbine blade damage detection system was established based on computer vision processing technology and deep learning algorithm,which can realize the identification,classification and location of blade damage.It is of great significance to the healthy operation of wind turbine.The main research work of this paper is as follows:(1)Image processing: In view of the problems of uneven illumination,low color saturation and noise interference in the fan blade images collected by UAV,the collected original images are preprocessed to improve the data quality.In view of the characteristics of complex local details and large background interference in the image,in order to ensure the accuracy of fan blade surface defect detection,r GB-Trackbar was used to extract the color features and set the manual threshold to black the background and separate the blade from the background.Through morphological processing and multi-scale Retinex algorithm processing,the interference of uneven light and fog on image can be well eliminated,so that the image defect information can be retained,and the data set can be made based on this.(2)Defect detection: Findcontour operator was used to extract the contour of the blade defect area,and the defects were classified according to the comparison of defect area and length diameter.The method is applied to the defect detection experiment on the self-made fan blade data set,and the accuracy is 94%.Due to the complex environment of fan blades,it is necessary to constantly adjust and set relevant parameters to extract blade defect features using machine learning method.In order to achieve better detection effect,an improved residual network is adopted for defect detection.The residual network model combining multi-scale feature fusion and RBN was used to test the self-made data set,and the detection accuracy reached 99.55%.(3)Fan blade defect positioning: ASIFT operator is adopted to improve the spatial shape control,this paper will collect leaves all parts of the image matching with the original fan blade as a whole,according to the longitude and latitude Angle first model,the need to match the two images of affine transformation interpolation resampling,to transform the image feature matching,the defect area is located in the blade.The accuracy of the method is 93.8% in the self-made fan blade defect data set,and the experimental results show that the method can complete the task of blade defect location well.