Research On Highly Efficient System For Surface Defects Detection Of Solar Cells

Solar cell module is the core component for photovoltaic in the new energy system,and the solar cells is the key part of the whole set.While due to the limit of craft and production environment,defects such as broken gate,missing angle,scratch and smudge are common problems on the surface of the solar cells in the process of production.These defects not only reduce the yield of products,but also effect the transfer efficiency which is not very high originally.It is therefore essential to detect and get rid of the defective products before they are combined into a whole set.The common manual detection method is largely affected by personal experience and subjective factors while machine vision detection method is limited by the defect detection algorithm.According to the need of instantaneity,accuracy and high speed in the solar cells defect detection during production,this article conducts research on efficient detection of solar cells defects.Based on underdetermined equation model,this research puts forward the solar solar cells defect detection algorithm,which detects minor defects like lines and spots quickly and accurately.Then,to solve the problem of low identification rate of underdetermined equation on rim defect and annular defect,this research proposes the Dp-Unet model based on deep learning to reach high precision rate and high identification accuracy rate.Finally,they are united to form a systematic detection platform.Details and innovative points are as follows:(1)A defect detection algorithm based on underdetermined equation is proposed to improve the widely used frequency domain analysis method.This algorithm inspects digital images on the surface of solar cells by adopting compressed sensing theory;it builds underdetermined equation about projection coefficient that defect row makes on the wavelet domain;then through comparison of inner product,it finally defines locations of maximum values among high frequency coefficients and detects defective images.Tests show that this algorithm,by integrating compressed sensing theory,can detect minor line defects,spot defects and mixed defects quickly,easily and accurately.(2)To improve the unsatisfactory result of underdetermined equation's detection on rim defects,deep learning defect detection model Dp-Unet,which is based on dynamic pooling,is carried out by comparing projection coefficient of image lines on one dimensional wavelet domain and determining great value of high frequency component.This algorithm treats solar cells digital images as discrete line signals.Through locating positions of jump point in wavelet domain,dynamic pooling is reached and characteristic information is extracted.Result of model training shows that Dp-Unet is real-time,fast and accurate in the process of production in terms of accuracy,recall rate and precision.This model is a strong complementation to unsatisfactory results of underdetermined equation's detection on rim defects.(3)Finally,this dissertation adopts Visual Basic 6.0 and carries out the solar cells detection system based on underdetermined equation and Dp-Unet.This system can meet the need of detecting products defects on 16 production lines at the same time and can improve productivity effectively.It can also achieve functions of detecting solar cells online,getting rid of defective products and inquiring real-time and historical data.