In-line monitoring of polymer processing: A multidisciplinary approach

grantor: University of TorontoAn in-line monitoring system is composed of a process/monitor interface, instrumentation and software. By drawing upon areas ranging from processing through spectroscopy and optics to multivariate data analysis, this thesis focuses upon the interface and software components of the system. Experimental data obtained by the in-line monitoring of polypropylene/polyethylene blends during extrusion were used as a basis for this work. The principal objectives of the thesis were: (i) to design and assess process-monitor interfaces which permit probe replacement during processing; (ii) to develop methods for comparing the quality of interpretation approaches used with spectral data acquired in-line, and (iii) to explore the properties of factor-based multivariate calibration models. Regarding the first objective, three designs of process-monitor interfaces were introduced which allow different commercially available fiber optic probes to be inserted. at different times without process interruption. These interfaces provide a means for measuring a variety of properties (only polymer blend composition was measured in this study). A statistical methodology based on the analysis of variance was implemented to show that melt blending, dry blending and temperature effects did not adversely affect the accuracy and precision of the data. Regarding the second objective, data preprocessing by second derivatives (d2), multiplicative scatter correction (MSC), and a combination of MSC and d2 was applied. The latter was the best preprocessing method for removing both additive and multiplicative effects from the in-line spectra. A unifying "continuum regression" approach was used to successfully obtain the most robust and parsimonious calibration model for predicting the blend composition from the preprocessed in-line spectra. The third objective encompassed three separate topics. First, nonlinearity tracking plots were developed for factor-based multivariate calibration models. The effect of data preprocessing on the nonlinear regions was demonstrated. The nonlinearity tracking plots revealed where, and to what extent, each factor contributed to regenerate the original spectrum. Second, a calibration transfer model based on direct on was used to transfer the calibration model from a laboratory instrument to a process instrument. This was accomplished by placing emphasis on the least number of factors (model parsimony) and the smallest number of "transfer samples". Finally, mid-infrared spectra were regenerated from in-line near-infrared spectra. The attributes and limitations of such a transformation were elucidated.Ph.D