Monitoring emulsion polymerisation using UV-Vis-NIR spectroscopy – Impact of measurement configuration on calibration models

Near infrared (NIR) spectroscopy has been investigated as a tool for monitoring emulsion polymerisation reactions using multivariate calibration models for estimating mean particle size and monomer concentration. While the models were promising, the studies used small datasets which do not sufficiently span the reaction profile space with batches having very similar trajectories for all the reactions used to build and test the models. This leads to potential pitfalls in the building of robust calibration models and also in their rigorous assessment.

In general, the extraction of chemical information from dense particulate suspensions, such as that arising in emulsion polymerisation, using NIR spectroscopy is complicated by multiple scattering effects which lead to sample-to-sample pathlength variations. Empirical pre-processing techniques have shown promise in some cases. However, no one pre-processing technique has been shown to convincingly be superior for a majority of cases. It can be expected that the performance of empirical pre-processing and calibration models will be dependent on the measurement configuration in addition to the particular system under consideration.

In this work, calibration models for estimating monomer conversion and mean particle size over the course of emulsion polymerisation reactions are developed using 3 different measurement configurations namely, total transmittance, total reflectance and collimated transmittance modes. The region of the electromagnetic spectrum considered ranged from ultraviolet to the NIR (300-1900nm). Seventeen reactions were carried out. The experiments were designed such that the reactions varied significantly in terms of the reaction profile and duration, final conversion and particle size range in order to reduce batch-to-batch correlations in the time profile of conversion and particle size.

Partial least squares (PLS) regression was used to build models for estimating mean particle size and monomer conversion. In each case, the models were built for different wavelength ranges in order to select the best wavelength range for model performance. Different pre-processing methods were investigated and the results analysed to study the impact of measurement mode on the performance of calibration models and pre-processing methods. It was found that the best wavelength range for predicting monomer conversion was 1548-1876nm which is in the first overtone region. For mean particle size the best wavelength region consisted of a combination of the visible and near infrared region. It was found that the diffuse reflectance and diffuse transmittance spectra provided similar performing models which were better than those built using collimated transmittance measurements.