Quantitative analysis of stibnite content in raw ore by Raman spectroscopy and chemometric tools

This paper presents a non-destructive and fast methodology to quantify the stibnite content in raw ore samples using Raman spectroscopy. A total of 120 raw ore reference samples were obtained from a froth flotation process, and multipoint collection and averaging were used to obtain Raman spectra from the raw ore samples. Our aim was to create the best multivariate calibration model for the quantitative analysis of stibnite contents in the raw ore samples. Several strategies were evaluated to generate a robust model; these strategies included preprocessing methods (de-nosing, baseline correction, and vector normalization), selecting the key wavenumbers for the quantitative analysis of stibnite, and building a multivariate calibration model. Synergy interval partial least squares (PLS), backward interval PLS, stability competitive adaptive reweighted sampling, and PLS with a genetic algorithm (GA) were investigated to build the linear multivariable models, whereas artificial neural network (ANN) and support vector regression (SVR) preceded by GAs (GA-ANN and GA-SVR) were implemented to build the non-linear multivariable models. Experimental results showed that the best multivariable calibration model for stibnite was obtained by a combination of an ANN and GA (GA-ANN). In contrast to the PLS model based on the full spectrum, the root mean square error of calibration and root mean square error of prediction of the GA-ANN method for the calibration and validation sets were reduced to 0.2038 from 0.3552 and to 0.2196 from 0.3927, respectively, and the squares of the correlation coefficients of the calibration and validation sets were increased to 0.9369 from 0.8053 and 0.9219 from 0.7561, respectively. The above results indicate that the multivariable calibration model for stibnite is stable. Furthermore, this method could be used in the froth flotation process to precisely determine the stibnite content in raw ore samples.