Calibration of the Composition of Low-Alloy Steels by the Interval Partial Least Squares Using Low-Resolution Emission Spectra with Baseline Correction

Abstract

Express determination of the elemental composition of steels and iron-based alloys is an urgent problem. Laser induced breakdown spectroscopy can be applied for its decision. The disadvantage of single- and multivariate modeling the elemental composition of steels is the semi-quantitative accuracy of the models. The aim of the study was developing quantitative multivariate calibrations of the concentrations of a set of chemical elements sufficient to identify low-alloy steels using low-resolution emission spectra. The multivariate partial least squares method was used to create the calibrations. Reducing the effect of redundancy of wideband emission spectra on the results of quantitative analysis was achieved by searching combination moving window containing one spectral variable more than the optimal number of latent variables for the wideband multivariate model. Further improvement of calibration accuracy was achieved by using the adaptive iteratively reweighted penalized least squares algorithm for spectrum baseline correction. Based on the laser emission spectra of 65 reference samples of low-alloy steels registered in the wavelength range 172–507 nm with a spectral resolution of 0.5 nm and a step of 0.1 nm, the following calibration models were developed: for carbon concentration with a root mean square error 0.059 % in the range ≤ 0.8 %, for manganese – 0.02 % and 2.0 %, respectively, chromium – 0.009 % and 1.0 %, silicon – 0.021 % and 1.2 %, nickel – 0.04 % and 0.8 %, copper – 0.019 % and 0.5 %, vanadium and titanium – 0.005 % without range limitation. The obtained multivariate models are quantitative for eight elements. These models give the possibility to identify the grade of low-alloy steels in an express manner at the stages of production or recycling.