MODEL DISCRIMINATION VIA DESIGNED EXPERIMENTS - DISCRIMINATING BETWEEN THE TERMINAL AND PENULTIMATE MODELS BASED ON TRIAD FRACTION DATA

Computer simulations have been used to study the application of statistical model discrimination techniques to the modelling of copolymerization reactions based on triad fraction data. Since there are six triad fraction measurements, the problems of parameter estimation and model discrimination become multivariate problems. The multivariate forms of three model discrimination methods (exact entropy, Hsiang and Reilly, and Buzzi-Ferraris et al.) are presented. Programs have been developed to simulate the application of these techniques to the systems styrene-acrylonitrile, styrene-methyl methacrylate and styrene-butyl acrylate. The simulation programs are explained, including the method used to simulate triad fraction data. Evidence is presented that the simulation programs are capable of duplicating the type of measurements that would be obtained in the laboratory. The simulation results show that model discrimination methods are able to accurately and reliably discriminate between the terminal and penultimate models. The use of simulated model discrimination methods leads to reliable discrimination in fewer experiments than have been used in past work. Also, model discrimination methods are able to detect smaller penultimate effects than those found by Hill et al. for styrene-acrylonitrile. In addition, the results show that the use of four of the six triad fractions, versus one copolymer composition measurement, should lead to more precise reactivity ratio estimates and an increased ability to discriminate between the terminal and penultimate models. Our work suggests that use of model discrimination methods will indeed lead to improvements in the modelling of copolymerization reactions.