Parametric study of spot size and multi-elemental quantification of geomaterials under complex matrix conditions using fiber-optic laser-induced breakdown spectroscopy

Abstract

Laser-induced breakdown spectroscopy (LIBS) has provided an optional access method to analyze multiple components of geomaterials with the advantages of rapid detection and minimal damage and preprocessing of samples compared with other conventional analytical methods. A fiber-optic LIBS (FO-LIBS) system with a compact laser probe was established to analyze ten different sediment samples, with Al, Fe, Ca, Mg, Ti, Mn, V, and Cr successfully detected. The spectra from the different samples are quite different due to matrix effects, and the spot size has a great influence on both plasma evolution and emission spectra for the same sample. The laser irradiance decreased from 1.382 to 0.855 GW/cm2 with increasing the spot diameter from 480 to 610 μm as the spectral fluctuation was gradually aggravated. Internal standardization (IS) was utilized to decline the spectral fluctuations caused by the spot size and matrix effects for the quantification of multiple elements. By adopting the most suitable combinations of emission lines, the fluctuation of emission lines caused by the spot size was also limited with a relative standard deviation lower than 30%. The R2 of all the calibration curves was higher than 0.9 with the major element Si out of the detection range and spectra mixed at different spot sizes, which outperformed normalization by the whole spectrum. Two multi-variate chemometric methods, including partial least squares regression and support vector regression, were also applied to establish quantitative models for the sediments. It was found that three methods provided similar prediction results, while IS could establish a more robust calibration model that was less sensitive to spot size. This study also demonstrated that FO-LIBS has the potential to quantitatively analyze the chemical compositions of geomaterials combined with IS and machine learning.