Quantitative and classification analysis of slag samples by laser induced breakdown spectroscopy (LIBS) coupled with support vector machine (SVM) and partial least square (PLS) methods

The laser induced breakdown spectroscopy (LIBS) technique coupled with a support vector machine (SVM) and partial least square (PLS) methods was proposed to perform quantitative and classification analysis of 20 slag samples. The characteristic lines (Ca, Si, Al, Mg and Ti) of LIBS spectra for slag samples can be identified based on the NIST database. At first, quantitative analysis of the major components (Fe2O3, CaO, SiO2, Al2O3, MgO and TiO2) in slag samples was completed by SVM with the full spectra as the input variable, and two parameters (kernel parameter of RBF-gamma and sigma(2)) of SVM were optimized by a grid search (GS) approach based on 5-fold cross-validation (CV). The performance of the SVM calibration model was investigated by 5-fold CV, and the prediction accuracy and root mean square error (RMSE) of SVM and PLS were employed to validate the predictive ability of the multivariate SVM calibration model in slag. The SVM model can eliminate the influence of nonlinear factors due to self-absorption in the plasma and provide a better predictive result. And then, two type of slag samples (open-hearth furnace slag and high titanium slag) were identified and classified by a partial least squares-discrimination analysis (PLS-DA) method with different input variables. Sensitivity, specificity and accuracy were calculated to evaluate the classification performance of the PLS-DA model for slag samples. It has been confirmed that the LIBS technique coupled with SVM and PLS methods is a promising approach to achieve the online analysis and process control of slag and even in the metallurgy field.