Comparison of univariate and multivariate data analysis models for uranium quantification in Trombay historical nuclear waste glass

Abstract

Abstract Laser induced breakdown spectroscopy recently has been investigated for analysis of nuclear waste glass for uranium quantification. The initial obtained accuracy and precision was ~15%. In this paper, we have compared the analytical merit of the univariative and multivariative PLSR regression models for the determination of U in barium borosilicate simulated waste glass containing significant amount of U. The analytical merit of a Czerny-Turner spectrograph with high spectral resolution and Echelle spectrograph with broadband spectrum recording capacity were compared using spectra simultaneously record from the same plasma. For univariative calibration the superiority of Czerny-Turner spectrograph over the Echelle has been demonstrated here. Multivariative chemometric PLSR model was found to drastically improve the results. It was also observed that selection of spectral window for analysis significantly affects the analytical merit of multivariative analysis. Echelle though shows relatively inferior analytical merit, but by applying Analytical spectral dependant PLSR in Echelle spectra, a much higher degree of improvement was observed. Using ASD-PLSR and Czerny-Turner spectrograph generated spectra an accuracy and precision of 2–2.5% was achieved in this study.