Study On Rapid Separation And Quantitative Detection Method Of Bauxite In Laser-Induced Breakdown Spectroscopy

China is a big country in the aluminum industry,and bauxite is the most important raw material in the aluminum industry.The content of various chemical components in bauxite directly affects its smelting process and application.The traditional bauxite composition detection method is time-consuming,complex and cannot detect multiple elements at the same time.It is urgent to find a simple and rapid detection technology.Laser-induced breakdown spectroscopy(LIBS)detection technology has the advantages of fast,simple,real-time and simultaneous detection of multiple elements,which can provide technical means for rapid detection of bauxite.However,LIBS spectrum has the disadvantages of weak intensity,high dimension and large number,which seriously affects its detection accuracy.Machine learning methods can provide effective tools for high dimensional data analysis in LIBS.In this paper,a high-precision detection method based on cavity-constrained laser-induced breakdown spectroscopy was proposed for fast and efficient detection of bauxite components.By constructing LIBS detection system,the mechanism of LIBS plasma spectral enhancement and self-absorption correction was explored,qualitative and quantitative analysis method based on machine learning combined with LIBS was mainly studied,which overcomes the limitations of time-consuming,complex and unable to detect multi-elements simultaneously in traditional bauxite detection,and realizes the rapid and accurate measurement of main components of bauxite.The main research work is as follows:(1)Aiming at the problems of weak spectral intensity and strong background radiation of traditional LIBS,a method based on spatial constraint was proposed,and the experimental parameters of bauxite LIBS detection system were optimized,which lays the experimental foundation for subsequent qualitative and quantitative analysis.The LIBS spectra were constrained by cavities with different diameters(from 1mm to 6mm)and different heights(from 1mm to 6mm).Through the analysis of spectral intensity,signal-to-noise ratio(SNR),electron temperature and other spectral parameters,the optimal cavity was obtained.The size was 5mm in diameter and 4mm in height.The effects of pressure,laser energy and delay time on the characteristic spectral intensity and SNR of Si and Al elements in bauxite were analyzed under the optimal cavity constraint.The experimental results show that the spectral stability is the best when the pressure is 150MPa,when the laser energy is 80mJ and the delay time is 1?s,the SNR reaches the maximum.(2)Aiming at the influence of self-absorption phenomenon on the quantitative analysis accuracy of bauxite,a self-absorption correction method based on electron temperature and electron density was proposed,and a self-absorption correction model was established to realize the self-absorption correction of main elements of bauxite in LIBS.The main characteristic spectral lines of Al,Si,Fe and Ti elements in bauxite samples were screened by LIBS spectra and NIST library,and the internal reference lines were selected by using the energy level properties of elements.The self-absorption coefficient of the spectral line was determined by calculating the electron temperature,and the first correction of the spectral line was completed.The results showed that the accuracy of Al,Si,Fe and Ti concentrations was increased by 5.21%,5.94%,11.28%and 5.6%,respectively.Using the spectral line broadening theory and electron density,the second correction of the spectral line was completed.The results showed that the accuracy of the main element concentration was further improved by 3.19%,4.26%,4.49%and3.3%.After two corrections,the Boltzmann fitting coefficient of each element in the sample has been greatly improved,and the accuracy of quantitative analysis has been improved.The self-absorption correction method proposed in this paper has also been verified in 303 stainless steel samples,which proves that the method has certain universality.(3)Aiming at the problems of high spectral line dimension,low classification accuracy and time-consuming in the application of LIBS technology in bauxite classification,a bauxite classification method,SVM combined with LIBS based on TM and PCA dimensionality reduction,was proposed.The classification models of TM-SVM and PCA-SVM were established,and the accurate classification of bauxite samples was realized.The original data of bauxite in nine different mining areas were reduced by TM and PCA,and the 10-dimension data after dimension reduction were input into SVM algorithm for classification model training.The results show that the accuracy,precision,sensitivity and F1 values obtained by PCA-SVM model are 1,which are significantly higher than those obtained by TM-SVM discriminant model.Under the same dimension reduction condition,SVM classification performance is better than KNN,DNN and RF,indicating that PCA-SVM model can effectively realize the classification of bauxite in different mining areas.The TM-SVM and PCA-SVM models proposed in this paper were also verified in the classification of iron ore,which proves that the model has certain universality.(4)Aiming at the problems of high spectral line dimension,serious matrix effect and low detection accuracy in the application of LIBS technology in the quantitative analysis of bauxite components,a quantitative analysis method(RFE-RFR)based on recursive feature elimination(RFE)combined with LIBS bauxite multi-element quantitative analysis was proposed.The quantitative analysis model of RFE-RFR was established,and the high precision quantitative detection of Al,Si,Fe and Ti elements in bauxite was realized.The RFR regression training of LIBS spectral data in different bands shows that the regression performance of 250nm-500nm was best.The LIBS spectra of bauxite samples were eliminated by RFE algorithm.The quantitative analysis of bauxite samples was completed by training in the RFE-RFR model,and the parameters of RFR were optimized.Compared with PLSR and SVM regression algorithm,RFE-RFR regression model has the best performance,the determination coefficient R~2 was above 0.95,and the RMSE was 0.2874,0.2070,0.3802 and 0.0347,respectively,which are better than other regression algorithms.The results show that RFR combined with LIBS can effectively measure the main elements of bauxite,and RFE feature selection method can further improve the accuracy of RFR quantitative analysis model.