Research On Intelligent Classification Of Scrap Based On Machine Vision And LIBS Technology

The scrap production in our country has exceeded 200 million tons per year,which provides a solid raw material guarantee for the development of EAF steelmaking.Using scrap for steelmaking can save 1.65 tons of iron ore,350 kg of standard coal,1.6 tons of CO2 and 4.3 tons of solid waste.The scrap is used as steelmaking raw material only after rough sorting and processing,without rapid quantitative detection and fine classification according to the surface corrosion,surface coating and alloy element content.As a result,the composition of molten steel can not be accurately controlled,and the alloy elements in some scrap can not be efficiently utilized.In this paper,combining machine vision and machine learning technology with LIBS technology,some key problems and industrial application of intelligent classification of scrap are studied.The main contents and results are as follows.(1)Based on the color histogram and K-means clustering analysis,the distribution characteristics of red brown rust color in RGB,HSV and YCbCr color space of rusted scrap are studied.The results show that RGB color space is not suitable for the extraction of rust color features.The H component value of red brown rust color in HSV space is mainly concentrated in the range of[0-45],while in YCbCr space,the Cb component value is mainly in the range of[70-120],and Cr value is mainly in the range of[130-170].Based on the gray level co-occurrence matrix and its characteristic parameters,the texture characteristics of rusty scrap are analyzed.The results show that the energy and correlation of rusted steel image are lower than that in the rust-free area,while the entropy of rusted area is higher than that of image background.The difference of contrast is not significant,but the boundary between rust and rust-free area can be distinguished.According to the analysis of color and texture characteristics,an intelligent algorithm is proposed to judge whether the scrap is rusty or not based on the fusion of color and texture characteristics.The accuracy of the algorithm is 98.14%in the training set and 96.88%in the prediction set.(2)The LIBS spectrum evolution characteristics,scrap coating identification and coating thickness detection methods of four kinds of coated scrap with zinc,tin,nickel and chromium are studied based on LIBS technology.The results show that for the scrap with thin coating,the spectral line strength of coating element decreases rapidly with the increase of laser pulse number.For scrap with thick coating,the spectral line strength of coating element first increases and the decreases with the increase of laser pulse number.For the uncoated scrap,the changing tendency of all spectral line strength is similar.The spectral line strength increases during the first few tens of pulses and then remains stable.The method of determining whether there is coating on the scrap surface based on the standard deviation threshold of normalized strength of Fe element spectral line.The standard deviation threshold is set to 0.02.The method of determining the coating element of scrap by the accumulated value of normalized intensity of element spectral lines is proposed.The evoluation law of three-dimensional morpholopy of ablation crater with the number of laser pulses is studied.The mathematical model between the depth of ablation crater and laser pulse number is established.The maximum between-cluster variance algorithm is used to calculate the critical pulse number.Finally,the thickness of scrap coating is calculated.(3)For a large number of complex social scrap,it is necessary to classify them according to the alloy composition.The key of this kind of classification method is to realize online,rapid and quantitative detection of alloy elements in scrap.In this paper,based on LIBS technology,the rapid quantitative detection of low alloy scrap and high alloy scrap is studied by using calibration method and free calibration method,respectively.A correction method based on GA-KELM algorithm is proposed to correct the matrix effect in LIBS calibration process.This method has the advantages of fast training speed,no need to adjust parameters artificially and good generalization performance.The results show that the root mean square errors of Si,Mn,Cr,Ni,V,Ti,Cu.and Mo in 47 low alloy steel samples are 0.2405%,0.1632%,0.0661%,0.0792%,0.229%,0.0411%,0.0759%and 0.0404%,respectively.A self absorption correction algorithm combining self absorption correction coefficient and genetic algorithm is proposed to improve the quantitative analysis performance of free calibration method.The results show that the root mean square errors of Cr and Ni in 7 high alloy steel samples are 2.80%and 2.19%,respectively.The detection accuracy of free calibration method is lower than that of calibration method,but it can be used for semi quantitative detection of high alloy scrap composition,based on which the high alloy scrap can be identified.It is beneficial for the rapid sorting of high alloy scrap.(4)In order to realize the automatic focusing of laser beam on the surface of scrap in the process of industrial application,a linear structured light measurement system is developed in this paper.The system parameter calibration and the algorithm of extracting the center of line structured light stripe are studied.Based on Zhang Zhengyou calibration method and 12 X 9 aluminum checkerboard calibration board,the camera internal parameters are calibrated,and the reprojection error is less than 0.1 pixel.Based on cross ratio invariable method and moving target method,the linear structured light plane and the moving direction of displacement platform are calibrated.The effect of scrap surface roughness,color and shape on stripe quality is studied.The results show that the rougher the scrap surface is,the closer the color is to silver white and the smoother the surface is,the higher the stripe quality is.The principal component analysis(PCA)method is proposed to extract the normal direction of the stripe,and then the Gaussian fitting method is used to further solve the stripe center in the normal direction of the stripe.The system is used to reconstruct the surface morphology of five kinds of scrap and the measurement accuracy of the system is verified by standard gauge block.The measurement error of the system is within 0.202 mm,which lays the foundation for the realization of LIBS automatic focusing function.(5)The prototype system of intelligent identification and classification of scrap is designed.The laboratory model is built and the control software is compiled.The collaborative control of laser,spectrometer,CCD camera and one-dimensional displacement platform is realized.The automatic control of the intelligent identification and classification model system of scrap in laboratory is realized,which lays a foundation for the industrial application in the future.