Research On Defect Detection And Measurement Of Wind Turbine Blade Based On Computer Vision Technology

With the rapid development of society prompting the increasing demand for energy,the reform of energy forms is gradually on the agenda of countries around the world.Wind energy,as a more easily accessible clean energy source,has become the focus of energy transformation.Wind turbines,as a kind of equipment to collect wind energy,are respected worldwide because of their high energy conversion ra te and non-polluting production process.In order to obtain more wind energy resources,wind turbines are usually installed in harsh environments such as the Gobi,mountains and deserts.Long-term harsh environmental impact will make wind turbine blades produce a variety of defects,the accumulation of damage will reduce the bearing capacity of the wind turbine blade structure,so the wind turbine blades for d efect detection on the safe operation of wind turbines has important significance.Most of the existing wind turbine blade defect detection technologies are based on contact detection methods,which rely on the use of sen sors and have the disadvantages of expensive,unguaranteed survival rate and easy environmental pollution of the collected data.With the innovation of computer technology,the introduction of computer vision in the field of structural health inspection ca n circumvent these problems to a large extent.Therefore,this study is based on computer vision technology to start the research of wind turbine blade defect detection method.Although the existing computer vision inspection technology is booming,it is not mature enough and there are various problems,such as the traditional inspection algorithm is not high in accuracy,the field of view is limited to the overall structure of the wind turbine blade can not be fully inspected and the quantification of defects is difficult.In view of this,the main content of this study consists of the following four aspects:(1)The current wind turbine blade defect detection method only targets the local area and cannot reflect the damage condition of the overall blade structure.To a ddress this problem,this study proposes a high-definition lossless image stitching algorithm,which is based on Harris corner point detector,optimized by adaptive non-maximum suppression algorithm(ANMS),random sample consensus(RANSAC)outlier rejection algorithm,and finally adjusted by image weighted fusion algorithm to achieve high-quality two-dimensional reconstruction of the overall structure of wind turbine blades,and to pave the way for the subsequent study to propose The final image fusion algorithm is adjusted to achie ve a high-quality 2D reconstruction of the overall structure of the wind turbine blade,which paves the way for the subsequent study of defect detection to achieve a comprehensive inspection of the overall blade structure.(2)To achieve more refined detection,this study proposes a semantic segmentation network for multiclassification defect detection,VGG16 Unet.the network framework of VGG16 Unet is derived from the replacement of the encoder network of the Unet network by the VGG16 network with excellent classification capability,and also introduces batch normalization a nd Dice loss function for optimization.Considering the expensive cost required to collect the dataset,this study also employs migration learning techniques to pre-train the semantic segmentation network for data augmentation purposes and to address the p roblem of insufficient features in the context of small data training.(3)The accuracy of defect quantification is determined by the accuracy of the detection algorithm.In order to achieve accurate quantification of defects,this study proposes the Seg Former-OHEM detection model to achieve the closest detection of each type of defects in wind turbine blades to the real damage situation by pixel masking.The Seg Former-OHEM model uses the Seg Former model based on Transformer blocks as the network framework,and adds the online hard example mining operation(OHEM)to improve the network detection accuracy by strengthening the hard examples in the dataset during the training process,so that the detection results can be used for the quantitative analysis of various defects.(4)A pixel matching-based defect quantification technique is proposed for the coexistence of multiple types of defects in wind turbine blades,which can accomplish the measurement of various types of defects under the coexistence of multiple types of defects.The measurement algorithm consists of two parts,the first part is the extraction part,which aims to simplify the multi-classification problem into a binary problem,and the second part is the measurement part,which performs the corresponding post-processing for each type of defects based on the pixel values of the extraction results to obtain the damage information of each defect.The measurement algorithm is designed in terms of multi-scale and adaptivity,and can be used to quantify defects in the overall structure of wind turbine blades,and introduce damage ratios to achieve estimation of the actual size of defects with practical engineering significance.