Calibration methods to circumvent unknown component spectra for quantitative in situ Raman monitoring of co-polymerization reactions

Abstract

The use of Raman spectroscopy for reaction monitoring has been successfully applied over the past few decades. One complication in such usage is the applicability for quantitative reaction studies. This analytical problem is intrinsic to any reaction system, and is part of the larger transinformation necessary to bridge qualitative Raman spectroscopic information through multivariate calibration approaches. Another compounding issue is the presence of unknown component spectra that are encountered when investigating novel reactions, which is often fraught with either a lack of understanding of reaction stoichiometries/mechanisms, or inability to isolate reaction intermediates or products for obtaining their Raman spectra. To overcome these analytical challenges, three numerical approaches are tested using the model styrene–butyl acrylate co-polymerization reaction. They are partial least squares regression (PLSR), a novel minimisation of mixture spectrum residuals (MMSR) algorithm, and the first ever attempt at combining band-target entropy minimisation curve resolution with multilinear regression (BTEM-MLR). Multivariate calibrations are performed using inline Raman spectra from co-polymerization monitored with offline NMR to estimate the concentration of monomers without the need for additional information on reaction intermediates or products. All three multivariate calibration approaches produce comparable success. The specific calibration dataset utilized and relative Raman molar intensities of chemical species directly impact the quality of calibration. Furthermore, both MMSR and BTEM yield additional spectral reconstruction of the unknown co-polymer Raman spectrum.