Research On Visual Inspection System Of Photovoltaic Cell Defects

With the progress of the photovoltaic power generation technology,solar energy is widely used.During the manufacturing of solar cells,it is inevitable to produce miscellaneous defects which will reduce photoelectric conversion efficiency and life span.Because traditional manual inspection and various common detection means suffer from slow detection speed,untimely response and high labor cost,they do not satisfy the fast,accurate and economic detection requirements.Therefore,in this paper,solar cells in the coating process are taken as the research object and a machine vision-based online defect detection system for solar cells is studied and developed which involves image acquisition,preprocessing,combination of traditional and deep learning algorithms and software functions.It is of engineering significance and practical value for this system to realize the automatic defect detection of solar cells.The research contents of this paper are summarized as below.(1)Inspection system design.The photoluminescence(PL)system for image acquisition and detection of defects inside and outside solar cells is built by analyzing production technologies of solar cells and detection requirements,selecting hardware and developing software functions.(2)Traditional detection algorithm design.First,a correction algorithm based on the correction coefficient matrix,grayscale projection and sub-pixel ROI extraction on Sigmoid function is proposed for nonuniform imaging of raw solar cell image.In addition,after analyzing the characteristics of different types of defects,an improved adaptive local threshold segmentation is presented to extract the block-type defects by algorithms such as morphology,region transform and Gaussian line detection.(3)Deep learning method design.The YOLO-v3 Model is applied to solve the problem of low inspection accuracy of some defects detected by conventional detection algorithms,and the YOLO-v3 is improved for black spot defects.Compared with the original model,the recall rate and accuracy rate of the improved one are increased by 6.33%,1.95%,respectively.(4)Experiment and result analysis: An online testing environment is realized by building a visual inspection system for cell defects,and the effectiveness of the proposed method is verified in terms of inspection accuracy,consumed time,and system performances.The experimental results show that the missing detection rate is less than 1.5%,the average accuracy rate is about 95.38%,and the consumed time is less than 1.5 second per cell,which meets the various technical indicators for the on-line defect detection of solar cells.