Polarization Vision-based Surface Defect Detection Of Wind Turbine Annular Components And Its System Research

Wind power as a new green energy has attracted widespread attention from society.The surface quality of the annular component,which is a key component of the wind turbine,directly affects the safe operation of the entire machine.Currently,surface defect detection of annular components mainly relies on manual inspection,with low accuracy and efficiency.Compared with traditional detection techniques,the advantage of polarized vision imaging technology is that it expands the information content from three degrees of freedom to seven,which helps improve the accuracy of target detection.This thesis takes a typical 8MW annular component(with a diameter above 5m)as the research object,first analyzes the types of surface defects,and then constructs a polarized vision detection system to mathematically model and analyze the optimization of surface defect detection of annular components.Secondly,the operational speed of 2D polarized image fusion is improved,the problem of the azimuth ambiguity caused by the 3D polarized reconstruction process is solved,and the detail expression ability of surface defects of annular components is effectively enhanced.Finally,through deep learning networks,defects are identified and classified,with a comprehensive accuracy rate of 93.4%,meeting the performance requirements of surface defect detection of annular components.The main research contents are as follows:(1)Polarized theory research of the annular component surface.First,polarized theory knowledge is studied,and the Equally Weighted Variance(EWV)is introduced as an indicator to measure the Full Stokes Vector(FSV)estimation performance.The Shuffled Complex Evolution(SCE)global optimization algorithm is used to optimize the model,realizing FSV detection and synchronous self-calibration of the hardware system.Secondly,an improved Wasserstein Generative Adversarial Network(WGAN)is constructed to generate partially polarized images,reducing the image acquisition frequency of FSV detection from two to one.(2)Polarized information preprocessing of the annular component surface.A polarized image enhancement method is proposed to improve the detail expression ability of the images.The traditional median filtering algorithm is improved,not only removing most of the noise but also preserving the edge information well.At the same time,the polarization image fusion method of lifting wavelet transform is used to analyze and optimize the surface information that cannot be locally detected by traditional imaging techniques,making the contrast between target surface defects and background significantly enhanced.(3)Semantic segmentation and classification of surface defects on the annular component.The causes of poor recognition performance of traditional defect recognition algorithms on the annular component surface are analyzed,and a Bilateral Segmentation Network(Bise Net)is constructed to achieve complex background defect segmentation.Finally,the proposed method achieves more accurate defect segmentation and good classification performance using deep learning defect classification networks.(4)Software design of the annular component surface defect detection system.Based on the detection requirements,the specific functions of the software system are determined,and integrating the studied algorithms into the software system,the surface defect detection system software is developed.The designed software system is tested,verifying the feasibility of the system operation.