Statistical determination of chemical composition and blending fraction of copolymers by multivariate analysis of ¹H NMR spectra of binary blends of the copolymers

Highlights

• Chemometric quantitative analysis was conducted for binary blends of copolymers.

• Blending parameters were predicted by multivariate analysis of ¹H NMR spectra.

• Predictions of composition and blending fractions of component copolymers were achieved.

• Assignments of ¹H NMR signals were unnecessary to conduct the predictions.

• Quantitative accuracy was improved through stepwise optimization of variables.

Abstract

A chemometric approach for the quantitative structural analysis of binary blends of copolymers was conducted. Three types of copolymers were synthesized by radical emulsion copolymerization of two out of three monomers—acrylonitrile, styrene, and α-methylstyrene—to prepare three series of binary blends of these copolymers. Partial least-squares (PLS) regression and least absolute shrinkage and selection operator (LASSO) regression were conducted with datasets in which the ¹H nuclear magnetic resonance (NMR) spectral matrix of the binary blends (explanatory variables) is combined with the blending parameter matrix (objective variables) of the binary blends. The blending parameters, such as chemical compositions and mole fractions of the component copolymers, were successfully predicted without any assignments of the ¹H NMR signals through stepwise optimization of the objective and explanatory variables. LASSO regression exhibited higher accuracy than PLS regression, suggesting that the variable selection in LASSO regression was responsible for the improvement in the quantitative prediction.

Introduction

A large number of industrial polymers are copolymers of several different monomers. The blend of copolymers is usually adjusted to achieve the properties that best suit the desired purposes. The properties of the copolymer blends depend greatly not only on the molecular structures, such as the chemical composition, of the component copolymers but also on the blending fraction. For example, acrylonitrile (AN) – butadiene – styrene (ST) (ABS) resin is a typical rubber-modified polymer and exhibits the so-called “sea-island” morphology, in which the rubbery polybutadiene phase is dispersed over a rigid continuous phase comprising the copolymer of AN and ST [1]. To enhance the property of heat resistance, a copolymer of AN and α-methylstyrene (αMS) with a higher glass transition temperature is often blended with the sea phase composed of the copolymer of AN and ST. Therefore, quantitative analysis of the blending parameters, such as the chemical composition and the blending fraction of the component copolymers in the sea phase, is very important to improving the properties of ABS resin.

Separation analysis, such as size-exclusion chromatography (SEC) and gradient polymer elution chromatography, is the first method of choice to analyze the features of copolymer blends [[2], [3], [4], [5]]. Chromatographic separation requires some differences in chemical properties, such as molecular weight or solubility, between the component copolymers. However, the only difference between the component copolymers, poly(AN-co-ST) and poly(AN-co-αMS), of the sea phase in ABS resin is the presence of a methyl group in the αMS units. Therefore, the chromatographic separation of these copolymers is difficult because of the similarity of their chemical properties; the development of another characterization method is required.

We have reported that multivariate analysis of nuclear magnetic resonance (NMR) spectra is useful for the structural analysis of synthetic (co)polymers [[6], [7], [8], [9], [10], [11]]. For example, in the principal component analysis of the ¹³C NMR spectra of copolymers of methyl methacrylate and tert-butyl methacrylate with various chemical compositions, the corresponding homopolymers and blends of the homopolymers with various blending fractions allowed successful extraction of information on not only the chemical composition but also the monomer sequence, without assigning the individual signals [6,7]. The chemical compositions of the copolymers were predicted rationally by partial least-squares (PLS) regression of ¹³C NMR spectra, in which the spectral data of the corresponding homopolymers and their blends were used as a training set. Recently, we also found that a similar analysis could be conducted by using the ¹H NMR spectra instead of the ¹³C NMR spectra [11].

In this study, we investigated the extent to which multivariate analysis of the ¹H NMR spectra of synthetic copolymers is applicable to extracting blending parameters in binary blends. Three kinds of copolymers were synthesized by radical emulsion copolymerization of combinations of two of the three monomers AN, ST and αMS to prepare three series of binary blends of the copolymers. PLS and least absolute shrinkage and selection operator (LASSO) regression, which are highly interpretable linear regression models, were used for multivariate analysis. The averaged chemical compositions in the binary blends were successfully predicted, whereas direct prediction of the blending parameters, such as the chemical compositions and the mole fractions of the component copolymers, failed. However, the blending parameters were successfully predicted by stepwise optimization of the objective and explanatory variables.