Feasibility Study on Multivariate Calibration Transfer in X-Ray Fluorescence Spectrometry

Abstract

Despite the widespread use of calibration model transfer in spectroscopy, its application to X-ray fluorescence (XRF) instruments has been largely unexplored. This study investigates the feasibility of transferring regression models between different XRF instruments: energy dispersive XRF (EDX) under varying conditions and between wavelength dispersive (WDX) and EDX. Metal ore samples were employed as a case study. We compared three model transfer algorithms (direct standardization, piecewise direct standardization, and slope and bias correction) for quantification of Al2O3, SiO2, K2O, CaO, TiO2, MnO, Fe2O3. It was found that piecewise direct standardization generally performed best in terms of prediction accuracy on a test set with transferred models. For example, in case of Fe2O3 the direct EDX data plugging into WDX model leads to tenfold increase in quantification error (from 5.34% for WDX to 52.75% for raw EDX in the range from 0.03 – 97.52%) which is definitely not acceptable even for semi-quantitative analysis. Model transfer allows decreasing the error down to original WDX model. The study demonstrates the potential and the practicability of calibration transfer protocol in XRF studies.

Keywords: X-ray fluorescence, element ore analysis, chemometrics, calibration model transfer