CC BY
14LD-O-8
LASER DIAGNOSTICS AND SPECTROSCOPY
Spectral variables selection in multivariate calibration of concentrations of C, Mn, Si, Cr, Ni, AND Cu in low-alloy steels by LIBS method
M. Belkov, D. Borisevich, K. Catsalap, M. Khodasevich
B.I.Stepanov Institute of Physics, National Academy of Sciences of Belarus, Minsk, Nezavisimosti Ave., 68-2
email address: m.khodasevich@ifanbel.bas-net.by
The advantages of the laser induced breakdown spectroscopy (LIBS) for determining the composition of steels are express multi-element analysis in the open air and the relatively low cost of instrument implementation. The main drawback of LIBS is the lack of accuracy in quantitative measurements [1]. Spectrometers with low spectral resolution are used usually in portable and mobile LIBS devices. Due to the strong overlap of the analytical lines in this case, the classical univariate calibration is inapplicable and multivariate models considering a large number of spectral variables are widely used [2]. Earlier the wideband low-resolution emission spectra (190-440 nm, resolution 0,4 nm, step 0,1 nm) of low-alloy standard steels were recorded in experimental conditions described in [3]. For training dataset selection the Kennard-Stone algorithm is applied to 31 to 39 standard steels. Multivariate models for calibration of C, Mn, Si, Cr, Ni and Cu concentrations without spectral variables selection were developed with root-mean-square error (RMSE) of prediction 0,06 % for C, 0,12 % for Mn, 0,09 % for Si, 0,13 % for Cr, 0,07 % for Ni and 0,08 % for Cu. Here three methods of spectral variables selection are used for improving the calibration accuracy: method of ranking spectral variables by their correlation coefficient with the value of the calibrated parameter [4], a successive projection algorithm [5] and an original modification of the method of searching combination moving windows for partial least squares model (scmwiPLS) [6]. The last model is the best and quantitative for C (RMSE = 0,004 %, the residual prediction deviation (RPD) in the test dataset is 23,4 in the concentration range from 0,13 to 0,43 %), for Mn (0,04 % and 5,2 in the range of 0,47-1,15 %), for Si (0,003 % and 20,7 in the range of 0,15-0,33%), for Cr (0,04 % and 3,1 in the range of 0,09-0,43 %) and for Ni (0,01 % and 4,8 in the range of 0,05-0,25 %). Calibration model is considered quantitative if RPD exceeds 3. For Cu in the concentration range of 0,06-0,26 %, scmwiPLS model is qualitative only (RMSE = 0,04 % and RPD = 1,4). Figure 1 shows the predicted vs measured values of C concentration for wideband PLS model and for scmwiPLS model. One can see a significant improvement in calibration accuracy.
Fig. 1. Predicted vs measured values of C concentration for wideband PLS (left) and for scmwiPLS (right) models.
ALT'22
[1] D. Syvilay [et al.], Guideline for increasing the analysis quality in laser-induced breakdown spectroscopy, Spectrochimica Acta Part B: Atomic Spectroscopy, vol.. 161, pp. 1-34 (2019).
[2] Z. Wang [et al.], Recent advances in laser-induced breakdown spectroscopy quantification: From fundamental understanding to data processing, TrAC Trends in Analytical Chemistry. Vol. 143, pp.116385-1-21 (2021).
[3] M. Belkov [et al], Multivariate Calibration of Concentrations of C, Mn, Si, Cr, Ni, and Cu in Low-Alloy Steels from Raw Low-Resolution Spectra Obtained By Laser-Induced Breakdown Spectroscopy, Journal of Applied Spectroscopy, vol. 88, pp.970974 (2021).
[4] Z. Xiaobo [et al.], Variables selection methods in near-infrared spectroscopy, Analytica Chimica Acta, vol. 667, pp. 14-32 (2010).
[5] S.F.C. Soares [et al.], The successive projections algorithm, Trends in Analytical Chemistry, vol. 42, pp. 84-98 (2013).
[6] M. Khodasevich, V. Aseev, Selection of Spectral Variables and Improvement of the Accuracy of Calibration of Temperature by Projection onto Latent Structures Using the Fluorescence Spectra of Yb3+:CaF2, Optics and Spectroscopy, vol. 124, pp. 748-752 (2018).
