Searching a mid-infrared spectral library of solids and liquids with spectra of mixtures

Abstract

A novel method for searching spectral libraries with spectra of mixtures has been extensively tested and the results of these tests are presented herein. The algorithm, referred to as a Mix–Match search, uses principal component regression (PCR) to model a library of spectra and to predict pseudo concentrations of components in the spectrum of a mixture. The PCR method is combined with an adaptive filter which selectively removes target components from a subgroup of the library and from the spectrum of the mixture in order to identify successive components. In the present study, the Mix–Match search algorithm was successfully used to search a mid-infrared library of 1,021 solid and liquid organic compounds with two- and three-component mixtures. All components in these mixtures were identified correctly with the Mix–Match search, whereas traditional search methods such as the Euclidean distance search identified only 4 out of the possible 12 components in the five mixtures. In a statistical test of the library, 435 synthetic two-component mixture spectra were generated from 30 measured spectra of pure compounds. The Mix–Match algorithm correctly identified 832 out of a possible 870 components in these mixtures for 96% accuracy, whereas a Euclidean search correctly identified only 600 out of the possible 870 components for 69% accuracy. It was also shown the algorithm could correctly identify components at the 5 to 10% concentration levels. The major advantage of the Mix–Match algorithm is its ability to identify components in mixtures without any prior knowledge of compositions. In the presently investigation, it was also shown that processing times and computer memory requirements could be greatly reduced by taking the Fourier transforms of the spectra in order to compress the data.