Optimization-Based Framework for Robust Modeling and Design of Kinetic Systems

Abstract

To understand the mechanism of a catalytic system, computational analysis is essential. Once a potential reaction mechanism has been identified, it typically involves (a) evaluating the energetics for the species and transition states using a computational chemistry method such as density functional theory (DFT), (b) relating these microscopic properties to macroscopic quantities such as reaction time. Microkinetic models may not match experimental data due to (1) inaccuracies in the energetic estimation, (2) inadequate catalyst models that are not representative of the surface environment. As a result of using experimental data, a more accurate model may be developed. If the model-experiment mismatch is resolved, a microkinetic model solution compatible with DFT assumptions may be identified. This work presents a generic optimization framework for solving parameter estimation and catalyst design problems in catalysis. Using a stochastic optimization method, Differential Evolution with Tabu List in conjunction with Aspen Plus, and considering experimental data, various activation energy and kinetic constants values were predicted. A sequential approach is a traditional approach to solving parameter estimation problems. Issues with the stiffness of the microkinetic model and the optimizer’s capacity to tackle such highly nonlinear systems are common challenges. This proposal has the potential to use all the reactor models present in Aspen Plus, as well as be able to use all the kinetic models and solvers, avoiding numerical difficulty in optimization solutions. This method has several advantages, including ease of implementation, which leads to physically realizable steady-state solutions, and a reduced overall optimization problem in terms of the number of variables involved. To validate this method, a dimerization of isobutane to produce isooctane is used as base line. The data considered as base case were previously worked considering experimental and simulation work. Once the parameter estimation was performed, the error produced was almost zero, and it was possible to generate the same kinetic data, concentration profile and molar flow produced.