Identification of the effective distribution function for determination of the distributed activation energy models using Bayesian statistics: Application of isothermal thermogravimetric data

Abstract

The new procedure for identification of the effective distribution function for determination of the distributed activation energy models, which is based on the Bayesian statistics, has been established. The five different continuous probability functions (exponential, logistic, normal, gamma and Weibull probability functions (the extended set of distributions)) were used for searching the most appropriate reactivity model for two heterogeneous processes: (a) the isothermal reduction process of nickel oxide under hydrogen atmosphere and (b) the isothermal degradation process of bisphenol-A polycarbonate (Lexan) under nitrogen atmosphere. Using the Bayes weights, it was shown that for both processes, the most suitable distributed reactivity model is the Weibull distribution model. The kinetic parameters (ln A, Ea) attached with the Weibull distribution model were calculated for both investigated processes, using three different computational methods (the maximum likelihood method (MLM), the nonlinear regression analysis (NRA), and the posterior mean (the expected value of scale parameter η, E(η)). It was shown that there is an excellent agreement between the values of kinetic parameters calculated by the MLM, NRA, and E(η) approaches. Using Bayes weights, it is possible to discriminate between different probability models and to quantify how well a distribution fits the experimental data. For the formal reactivity model comparison, the use of the (nonnormalized) Jeffreys prior is recommended. © 2010 Wiley Periodicals, Inc. Int J Chem Kinet 42: 641–658, 2010