Rapid quantitative analysis of the acidity of iron ore by the laser-induced breakdown spectroscopy (LIBS) technique coupled with variable importance measures-random forests (VIM-RF)

Abstract

Rapid and online analysis of the acidity of iron ore is extremely important for reasonable and efficient utilization of mineral resources. In this study, the laser-induced breakdown spectroscopy (LIBS) technique coupled with variable importance measures-random forests (VIM-RF) was proposed and applied for rapid and effective analysis of acidity in iron ore. LIBS spectra of 50 iron ore samples were collected, and the characteristic spectral lines of major elements (Ca, Mg, Si and Al) in iron ore samples were identified based on the National Institute of Standards and Technology (NIST) database. Different pre-processing methods, input variables and RF calibration model parameters were investigated and optimized by 5-fold cross validation (CV), and variable importance measurement (VIM) was used to optimize the input variables of the RF calibration model. In order to further verify the predictive ability and robustness of the VIM-RF calibration model, three calibration models of VIM-RF, partial least squares (PLS) and least squares support vector machine (LS-SVM) were applied for the quantitative analysis of acidity in iron ore, and the correlation coefficient (R²) and root mean squared error (RMSE) were evaluation indices. The results show that the VIM-RF model exhibits an excellent predictive performance compared with the other two calibration models both for the calibration set and prediction set. Therefore, the LIBS technique combined with VIM-RF can achieve a rapid acidity analysis of iron ores, and it will provide a new method and technology for selection and quality control of iron ore in the metallurgical industry.