Partial least squares assisted influence coefficients concept improves accuracy in X-ray fluorescence analysis

Routine application of X-ray fluorescence (XRF) spectrometry in laboratory practice requires special procedures to correct for matrix effects. One of the most popular procedures is the influence coefficients method. In this method the regression equation relating analytical signal to the target element concentration is extended with empirical terms correcting the influence from particular matrix elements. While the influence coefficients method is quite accurate, it is rather laborious as it requires individual selection of matrix terms for each element under study. The influence coefficients method is based on the least squares regression technique, thus the number of matrix correction terms is limited by the number of available standard calibration samples. Here we propose a very simple technique that can take into account an unlimited number of terms in the influence coefficients method through the employment of partial least squares regression (PLS) where spectral intensities, their ratios and squared intensities are employed as variables. Unlike traditional application of PLS in XRF studies where the regression model is built using spectral intensities only, the proposed approach inspired by classic influence coefficients allows elimination of complex specific XRF matrix effects. The paper describes the suggested method and demonstrates its' performance in two EDXRF data sets with significant matrix effects (ore and steel samples).

Automated summary

This paper introduces a new XRF spectrometry method that combines the influence coefficients method with partial least squares regression to correct for matrix effects and eliminate the influence of sample matrices. The proposed method is simple and effectively improves the accuracy and reliability of analysis.