The Research Of Quantitative Analysis Methods For Steel Based On Laser-induced Breakdown Spectroscopy Technology And Chemometrics

Iron and steel industry play an important role in the metallurgical industry market, its production increases year by year, and has arisen rigorous energy consumption at the same time. However, the traditional technologies for steel analysis need the complicated sample preparation, which is difficult to ensure each link (smelting, processing and inputting market) of the iron and steel smoothly, thus put forward a higher requirements to the traditional analysis technologies. Laser-induced breakdown spectroscopy (LIBS) is a material element analysis technology based on laser induced plasma emission spectrum, and has many advantages, such as no complicated sample preparation, accurate, rapid, multielement simultaneously analysis, remote detection and apply to any types of sample analysis, which has been widely used to many key areas especially in metallurgical industry. However, the collected LIBS spectra information usually exist lots of interference information because of experimental instrument and samples, which is not favor for accurate quantitative analysis of LIBS technology. Chemometrics methods can extract effective information from complex spectrum, which is one of the effective ways to improve accuracy of LIBS analysis. In this work, on the basis of practical problems of rapid and accurate steel analysis, chemometrics methods on the spectral preprocessing, quantitative analysis and feature extraction were studied and used to improve the accuracy of steel analysis. It enriches the research contents of chemometrics and metallurgical analysis, and provides new idea, method and technical support for the on-line analysis of metallurgical industry and quality supervision. There are four chapters in this study, and the main research content is as follows:1. With LIBS spectrum of steel as the research object, the wavelet analysis was applied to spectral baseline drift and noise problem. Firstly, LIBS spectra were corrected through the optimized wavelet basis function and the decomposition layers. Secondly, LIBS spectra were de-noised using the optimal wavelet basis function and decomposition layers. Compared with the Fourier Transform denoising, the wavelet analysis has a high speed, good reproducibility, and can solve the problem of baseline correction and denoising at the same time.2. Random Forest Regression (RFR) combined LIBS technology has utilized to quantitative analysis nonmetal elements of S and P in steel simultaneously. Based on optimized input variables and two important parameters (ntree and muy), RFR calibration model was built to predict content of S and P. Compared with classical Partial Least-Squares Regression (PLSR), RFR has a better accuracy of quantitative analysis and good stability.3. Based on the last research work, RFR method based on Sequential Backward Selection (SBS) was uesd to improve the accuracy of quantitative analysis of S and P in steel. Firstly, feature subsets were optimized by using SBS method for feature extraction according to root mean square error and the correlation coefficient as evaluation indicators; RFR model was built by using optimized feature subset as the input variables and then optimized the model parameter; Finally, on condition that the optimized model parameters and feature subset, RFR model was constructed to predict S and P of tested samples, the result shows the correlation coefficient is all over 0.9990. Compared with traditional RFR method, this method has the advantages of higher analysis precision, faster speed, and better reproducibility.