Spatially and angularly resolved diffuse reflectance spectroscopy for in-situ monitoring of suspension polymerisation reactions

  • Gledson Jose

Student thesis: Doctoral Thesis

Abstract

Despite the widespread use of optical spectroscopy for monitoring polymerisation processes, significant challenges remain for its successful application on suspension polymerisation reactions. The high heterogeneity and viscosity of such reaction media make sampling a challenging task and deteriorates the accuracy of reference and spectroscopic measurements. In this thesis, this problem is tackled by taking advantage of the stronger scattering susceptibility of the visible spectral range and the deeper penetration depth offered by the near infrared region. An investigation is carried out to evaluate whether the predictive capability of multivariate calibration models could be improved by introducing new variations in measurement configuration, in particular, spatially and angularly resolved illumination. An empirical approach based on multivariate calibration methods and chemometrics is also proposed for the extraction, pre-processing, fusing and modelling of such multidimensional information. How well these different measurements are integrated and how accurate the multivariate models can be, are one of the main questions of this thesis. They are first studied through a two-component system composed of polystyrene and water, followed by full suspension polymerisation reactions.The results suggest that the accuracy of multivariate calibration models can be improved by (i) including angularly orientated fibres, (ii) fusing information from two or more source-detector separations, (iii) by how data is fused or manipulated,and (iv) by the quality of the measurements. To the best of my knowledge, this work is the first attempt to employ spatially and angularly resolved diffuse reflectance spectroscopy as a Process Analytical Technology (PAT) tool for monitoring suspension polymerisation reactions, and also the first in which a data fusion approach based on Multiblock Partial Least Squares regression (MB-PLS) is evaluated for this purpose.
Date of Award1 Oct 2014
LanguageEnglish
Awarding Institution
  • University Of Strathclyde
SupervisorPaul Mulheran (Supervisor) & Anthony Morris (Supervisor)

Cite this