ИСТИНА

X-Ray fluorescence (XRF) is a useful tool for rapid multielemental analysis of geological samples that does not require acid digestion. One of the important tasks given to XRF is the simultaneous quantification of lanthanides in ores, which is a problem due to strong line overlapping. Thus, high resolution wavelength-dispersive XRF (WDXRF) is used. However, the question still stands: can cheaper low-resolution energy-dispersive spectrometers be used for this task? To answer this question we provided an assessment of total reflection XRF (TXRF) as a tool for direct analysis of REE-rich ores. In this work, we used partial least squares (PLS) and principal component regression (PCR) to circumvent spectral interferences in both TXRF (S2 Picofox, Bruker GmbH) and WDXRF (S8 Tiger, Bruker AXS) spectra. To provide a uniform distribution of concentrations of lanthanides in a test set we implemented specialized design of experiment (DoE) based on Latin hypercube sampling (LHS) [1]. We obtained the matrix of 5 factors (elements) and 20 levels (concentrations as well as samples). The maximum correlation is 0.03, which is several times better than in [2]. In addition, this design has an advantage over DoE in [3] where the number of samples equals a square of the levels. Thereafter, our model was applied for the determination of Ce (460–39500 ppm), La (260–24100 ppm), Nd (150–11800 ppm) by both TXRF and WDXRF in standard reference materials of niobium and uranium ores used for validation. In addition, WDXRF provided quantitative determination of Pr (47–3800 ppm), Sm (24–1500 ppm) in these samples. We also provide a comparison of PLS, PCR and integrated software results. Chemometrics provided twice the best average root mean square error of prediction (RMSEp) for Ce, La, and Nd for TXRF and thrice for WDXRF (average RMSEp for all elements) compared to the integrated software.