OPTIMAL EXPERIMENTAL DESIGN FOR MODEL DISCRIMINATION OF SEMI-BATCH EMULSION POLYMERIZATION PROCESSES

Abstract

A number of methods for model-based experimental design for obtaining parameter precision are regularly used in industry and academia. 1-lowever, the application of Optimal Experimental Design (OED) methods for model discrimination is limited. Michalik (Michalik et al.,2010) have proposed a method for OED-model discrimination (OED-MD) method based on the Akaike Weights Design Criterion (AWDC). The implementation of this method for a simple literature example shows that the AWDC based methods result in significant reduction on experimental efforts compared to the classical design criteria. This work focuses on using the AWDC-based OED-MD method for a medium scale, but significantly complex process. The process considered for this study is a two monomer semi-batch emulsion polymerization process. This is a complex process as it involves multiple phases, mass transfer between multiple phases and complex reaction kinetics. It would, thus, be interesting to implement the performance of the AWDC-based OED-MD method. The model for semi-batch emulsion polymerization was developed by VSC LIT. Prague. This model contains 19 differential equations, 82 algebraic equations and 34 parameters. To apply AWDC criteria for model discrimination and optimal experimental design, various model alternatives were required. This "Main Model" was further modified and various models with different assumptions were proposed. Softwares used for the work includes. gPROMS 3.3.1 and MATLAB 2011 b. Optimization is done using IPOPT suite. The above mentioned models developed were further used in discrimination of model candidates. The dissertation concludes by specifying the best model of the process, which may be used for determining optimum operating conditions and (or) for designing suitable control system. Also, it improves the understanding of the complex process taken into study, such as, sensitivity with the assumptions considered while model building. This work has been done at RWTH Aachen. Germany and lIT Roorkee, India (German Academic Exchange Programme) under joint supervision of Prof.Alcxander Mitsos and Prof.Surendra Kumar.