Automatic Mechanism and Kinetic Model Generation for Gas‐ and Solution‐Phase Processes: A Perspective on Best Practices, Recent Advances, and Future Challenges

Abstract

Completely automated mechanism generation of detailed kinetic models is within reach in the coming decade. The recent developments in this field of chemical reaction engineering are anticipated to lead to some groundbreaking discoveries in the future, extending our fundamental understanding and resolving many of today's society problems such as energy production and conversion, emission reduction, greener chemical production processes, etc. In the present review, the focus is on the core of these automated mechanism generation for gas‐phase and solution‐phase processes that is on how the reaction kinetics and thermodynamic and transport properties of species are estimated and calculated starting from the fundamental elements of the software. With tasks such as the definition of reaction rules and reaction families, the unambiguous representation of species, and the choice of different termination criteria, generating a good reaction mechanism is still not as simple as pressing a “run” button. One of the main challenges that still needs to be overcome is how to deal with data scarcity and the combination with affordable computational chemistry calculations seems the logical step forward. The best practices are illustrated in a butane pyrolysis case study, which also exposes the challenges in the field of automatic kinetic model generation.

Harnessing the Combined Power of Theoretical and Experimental Data through Multiscale Informatics

“…Higher levels of automation would improve the scalability of the approach to larger systems (where likely more parameters and more data are necessary), reduce uncertainties in the multiscale informed model (through techniques that generate more data used to further constrain model parameters), and create entirely new possibilities that progress toward completely autonomous scientific inquiry. Indeed, even creating additional scripts that link the multiscale informatics codes to other existing codes for automated model construction (e.g., RMG ) and automated theoretical calculations (e.g., Kinbot , Predictive Automated Phenomenological Elementary Rates (PAPER) , and Bhoorasingh and West ) would enable interesting possibilities. The author's expectation is that pushing the limits of automation will enable methodologies that rapidly accelerate the pace of scientific progress in the development of high‐accuracy models for complex chemical systems.…”

“…Though as evident by the list of model parameters (included in the uncertainty quantification) and their associated uncertainties in previous implementations , there appear to be opportunities for automation of this stage through rule‐based methods as well, which would further decrease the amount of time it takes to implement the approach. Additionally, while RMG was used to generate the starting kinetic model for the secondary reactions in previous multiscale informatics implementations , the resulting model was then only incorporated into the multiscale informatics approach manually. It may also be possible to link RMG software outputs directly through automated scripts.…”

An exponential expression for gas heat capacity, enthalpy, and entropy

“…Group additive correlations are often the most accurate [10,22] but also the most complex to handle [21,[23][24][25]. Thinh et al further developed additive models to calculate enthalpy, entropy and Gibbs free energy [26,27]. These correlations are accurate but cumbersome to implement because of the large number of parameters (not parsimonious).…”

Discerning complex reaction networks using automated generators

“…Figure displays the size of more than 50 kinetic mechanisms in terms of number of generated reactions R and species S for various combustion, oxidation, pyrolysis, and catalytic processes. The complexity of a detailed model grows linearly with the number of species included in the model as pointed out by Lu and Law, and Van de Vijver et al for combustion and pyrolysis. We augment this observation with additional types of chemistries and note that the linear trend, which can be approximated by the correlation R ≈ 5 S , is not dependent on the specific chemistry of the model.…”