Fast determination of oxide content in cement raw meal using NIR spectroscopy with the SPXY algorithm

Abstract

NIR spectroscopy combined with the partial least squares (PLS) algorithm has been investigated to yield a new analytical method for the fast determination of CaO, SiO₂, Al₂O₃ and Fe₂O₃ in cement raw meal samples in the process of cement production. An improved algorithm based on Kennard–Stone (KS), named sample set partitioning based on joint x–y distance (SPXY), which takes into account both x (spectral variables) and y (chemical values) spaces, was applied to divide cement raw meal samples into calibration and prediction subsets, and the effect of SPXY was compared with that of random sampling (RS), KS and Duplex. Optimal predictions were obtained in most cases with the SPXY method, and the average prediction errors were 0.1430%, 0.1206%, 0.0667% and 0.0306% for CaO, SiO₂, Al₂O₃ and Fe₂O₃, respectively, showing that the calibration subset selected by the SPXY method is superior to that of the other three methods, and NIR spectroscopy coupled with SPXY and PLS algorithms can achieve rapid and accurate determination of oxide content in cement raw meal samples.