Optimisation of fast quantification of fluorine content using handheld laser induced breakdown spectroscopy

Hydrofluoric acid represents the majority of the industrial applications of fluorine in the world. It is synthetized from fluorite, which is commonly purified by the froth flotation process to attain the high-grades required for hydrofluoric acid production. Besides, in metallic ores such as tungsten and phosphate ores, fluorite does not represent any added value compared to the extracted metals and is, therefore, considered as a gangue mineral that has to be rejected. In both cases, the fluorine content has to be known precisely in the flotation process as well as in all the industrial applications involving fluorine, to estimate the process efficiency and to optimise the operations. Nevertheless, fluorine quantification is difficult using conventional techniques since it usually includes heavy sample preparation such as dissolution. Though, Laser-Induced Breakdown Spectroscopy (LIBS) provides a multi-element detection that has been successfully used to quantify fluorine using either elementary or CaF molecular bands. Here, rock samples exhibiting a wide range of fluorine contents (from 1.48% to 40.73%) were analysed, the fluorine being mainly comprised in fluorite. These samples corresponded to the products of different flotation tests conducted on the same tungsten-skarn ore. The experimental conditions were optimised to study the two CaF molecular bands, located between 529 and 543 nm, and between 590 and 606 nm, respectively. Systematically, the LIBS emission intensities of the two studied bands were evaluated using the peak areas, which were normalised, averaged over several ablated zones, and correlated with the fluorine content determined by the fluoride-ion sensitive electrode method. The particle size played a key role as significant differences in the LIBS intensities were exhibited between the 10-150 mu m and the < 20 mu m powders, with no discernible correlation between the size ranges and signal magnitude. Furthermore, the matrix effects strongly impacted the LIBS intensities, which displayed a non-linear relationship with the fluorine contents: this induced the development of non-linear univariate models that were calculated on 27 training samples and validated on nine testing samples (3:1 ratio). Although non-linear models fitted adequately the experimental data, a multivariate approach considering the two studied CaF bands was adopted to overcome the matrix effects. A formula with linear, quadratic, and interaction terms was generated from the multivariate regression, predicting fluorine contents with R-2 = 0.94 and a mean average error of 2.18%F. The developed models demonstrated that a precise and accurate quantification of fluorine is possible using a calibrated handheld LIBS, providing an on-line estimation of the processes efficiency and a real-time adaptation.