Nondestructive Speciation of Solid Mixtures by Multivariate Calibration of X-Ray Absorption Near-Edge Structure Using Artificial Neural Networks and Partial Least-Squares

Abstract

Two multivariate calibration methods, artificial neural networks (ANN) and partial least-squares (PLS), have been applied to the quantitative determination of iron species in solid mixtures by X-ray absorption near-edge structure (XANES). XANES spectra were successfully resolved by both methods, and the iron compounds in solid mixtures were quantified, even though the spectra of different compounds showed serious overlap. When iron compounds that were not contained in the model mixtures were subjected to the calibration model, ANN recognized the patterns of their XANES spectra as the nearest spectra of model compounds in shape, and gave more robust results than PLS. The self-absorption effect on the calculated values from XANES measured in the fluorescence mode was examined by comparing with transmission mode; it turned out that a spectral distortion by a self-absorption effect is irrelevant to the prediction performance of these multivariate calibration methods. The present study demonstrated that ANN and PLS are applicable to the chemical speciation of elements by XANES measured in the fluorescence mode.