Место издания:Tomsk Polytechnic University Tomsk, Russia
Первая страница:40
Последняя страница:41
Аннотация:Laser-induced breakdown spectrometry (LIBS) is an emerging technique for materials analysis, based on laser sampling and simultaneous detection of the light emission from a laser-induced plasma. The possibility of a non-contact sampling by focused laser radiation makes possible real-time analysis of samples of different origin (construction and composite materials, coating layers, alloys, environmental and geological objects, pieces of the art, etc.) with minimal sample preparation and extremely short analysis time. These advantages of LIBS give an opportunity for rapid direct analysis of materials during operation process. Quantitative analysis of steels is focused both on impurities and doping components determination, but there are numerous spectral interferences due to extremely complex emission spectra of iron. In the field of control of both metallurgy process and assessment of constructions (e.g. railway rails) it is important to solve this problem for reliable and accurate analysis result.
Multivariate calibration techniques are commonly applied for analysis in the case of overlapping signals. In the present work we focused on searching an appropriate calibration strategy for LIBS determination either metal or non-metal components in various steels. We tried two multivariate regression methods (PCR and PLS) for solving the problem of spectral interferences for determination of silicon, chromium, nickel, manganese and carbon in steels. Data pre-processing includes the averaging of the spectra from several laser pulses and removing of the background as the minimal intensity in the spectrum. The stability of the model was verified by the one-leave-out cross-validation procedure. A special criterion for the determination an optimal number of principal components was used. We compared the results of multivariate regression with ordinary univariate linear regression. In cases of Mn and Cr we found experimental conditions to isolate an analytical lines and obtained a good univariate calibration with the use of baseline correction and an appropriate internal standard (R2~0.996). For Si multivariate calibration provides worse results than univariate calibration despite the existing of spectral interferences of Si with the lines of Ni and Cr (R2~0.86, R2~0.94, respectively). Nevertheless, multivariate linear regression methods gives moderate prediction capability for all elements of interest in the case of overlapping signals (R2~0.86 – 0.99).