(141c) Experimental Design for Reaction Rate Model Discrimination

Reaction rate modeling is a crucial tool for understanding and optimizing chemical reactions. Multiple kinetic models can be developed to describe the same reaction, which often poses a challenge for practitioners in selecting the most appropriate model for their studies. In this case, one should perform experiments that will maximize the discrimination capacity. Some of the well-known model discrimination based experimental designs include: Hunter and Reiner’s criterion [1], Box and Hill’s criterion [2], T-optimality criterion [3], and Akaike weights design criterion AWDC [4]. While various solutions are available for addressing this challenge, emerging optimization tools encourage faster and more efficient methods for discriminating between increasingly more complex models. In this work, we aim to investigate model discrimination and the design-of-experiments capabilities using Bayesian Optimization (BO). BO utilizes a non-parametric Gaussian Process model, which does not have a defined structure and relies on prior knowledge about the behavior of reaction rates. Thus, we will study the usefulness of the proposed strategy in discriminating between nonlinear parametric models with as little experiments as possible. To this end, different experimental design procedures for model discrimination will be compared. The proposed methodology will be tested using the Methanol Synthesis Reaction Modeling Case Study [5]. In this case study, twenty kinetic models are proposed and only 27 data points are provided. Each model is a function of four experimental conditions: temperature, partial pressures of CO, H₂, and MeOH.

Keywords:

Experimental design; Bayesian Optimization; Reaction rate modeling; Model Discrimination

References:

[1] W. G. Hunter and A. M. Reiner, “Designs for Discriminating Between Two Rival Models,” Technometrics, vol. 7, no. 3, pp. 307–323, Aug. 1965, doi: 10.1080/00401706.1965.10490265.

[2] G. E. P. Box and W. J. Hill, “Discrimination among Mechanistic Models,” Technometrics, vol. 9, no. 1, pp. 57–71, 1967, doi: 10.2307/1266318.

[3] A. C. P. de Leon and A. C. Atkinson, “Optimum Experimental Design for Discriminating Between Two Rival Models in the Presence of Prior Information,” Biometrika, vol. 78, no. 3, pp. 601–608, 1991, doi: 10.2307/2337029.

[4] L. Cai, S. Kruse, D. Felsmann, C. Thies, K. K. Yalamanchi, and H. Pitsch, “Experimental Design for Discrimination of Chemical Kinetic Models for Oxy-Methane Combustion,” Energy & Fuels, vol. 31, no. 5, pp. 5533–5542, May 2017, doi: 10.1021/acs.energyfuels.6b03025.

[5] R. J. Berger, J. Hoorn, J. Verstraete, and J. W. Verwijs, “Software functionality assessment for kinetic parameter estimation, model discrimination and design of experiments : The four test cases,” Eurokin, Apr. 2001, [Online]. Available: http://eurokin.tudelft.nl/?cat=30.