Research On Defect Detection Method Of Wind Turbine Blade Based On Ground Image

As a green and pollution-free clean energy source,wind energy is favored by the energy supply structural reform strategies of countries around the world.As an important part of wind turbine,damage to the surface of wind turbine blades will not only reduce the power generation efficiency,but also may cause damage to other facilities of the wind turbine and even threaten the safety of the staff.Therefore,it is essential to health monitoring of wind turbine blade surface.At present,researchers have proposed many methods for detecting damage to the surface of wind turbine blades,but there is still a shortage of methods that combine economy,safety and efficiency and can accurately locate the location and area of damage.To address the surface damage detection of wind turbine blades,this paper uses computer vision and image processing techniques to study the classification,localization and damage area measurement of wind turbine blade surface damage detection met hods.Firstly,it is proposed a fast detection method of wind turbine blade surface damage based on improved Cascade RCNN convolutional neural network.Specifically,it includes: using the superior one-stage detector Center Net to replace the original area proposal network in the two-stage detector;adjusting the higher IOU threshold;and using the training method of joint training of the first and second stages.The models validate the effectiveness of the method on the COCO dataset and the wind turbine blade damage dataset,respectively.Secondly,it is proposed for a method to improve the backbone network of the Cascade RCNN model to address the problem that the improved Cascade RCNN model has a large error in detecting small target damage of wind turbine blades.The method uses Res Net50,which fuses multi-scale information,as the backbone network of the model,and adds a CBAM attention mechanism to correct the problem of channel information loss due to the addition of Res Net modules with multi-scale information.The improved CR-Center Net network improves the recognition accuracy of small target damage of wind turbine blades significantly,and the average accuracy of the model is also slightly improved.The experimental results verify the effectiveness of the wind turbine blade damage detection model based on the improved CR-Center Net network.Finally,for the wind turbine blade damage localization and image stitching problem,a ground head remote acquisition wind turbine blade image model is established.The distortion coefficient between the sampled images is obtained through the geometric relationship and the images are corrected,a curve multi-feature point matching stitching algorithm based on edge extraction is proposed,and the corrected images are stitched to determine the specific location of the blade damage.Meanwhile,the corresponding coefficients between the pixel points of the corrected image and the real wind turbine area are obtained by geometric relations,and the binarized blade damage area is calculated.The experiments show that the method effectively improves the accuracy of locating the wind turbine blade damage and the accuracy of measuring the damage area.