A Machine Learning-Aided Equilibrium Model of VTSA Processes for Sorbents Screening Applied to CO2 Capture from Diluted Sources

Alexa Grimm; Matteo Gazzani

doi:10.1021/acs.iecr.2c01695

A Machine Learning-Aided Equilibrium Model of VTSA Processes for Sorbents Screening Applied to CO₂ Capture from Diluted Sources

Ind Eng Chem Res. 2022 Sep 21;61(37):14004-14019. doi: 10.1021/acs.iecr.2c01695. Epub 2022 Sep 6.

Authors

Alexa Grimm¹, Matteo Gazzani^{1

2}

Affiliations

¹ Utrecht University, Copernicus Institute of Sustainable Development, Princetonlaan 8a, 3584 CBUtrecht, The Netherlands.
² Sustainable Process Engineering, Chemical Engineering and Chemistry, Eindhoven University of Technology, 5612 APEindhoven, The Netherlands.

Abstract

The large design space of the sorbents' structure and the associated capability of tailoring properties to match process requirements make adsorption-based technologies suitable candidates for improved CO₂ capture processes. This is particularly of interest in novel, diluted, and ultradiluted separations as direct CO₂ removal from the atmosphere. Here, we present an equilibrium model of vacuum temperature swing adsorption cycles that is suitable for large throughput sorbent screening, e.g., for direct air capture applications. The accuracy and prediction capabilities of the equilibrium model are improved by incorporating feed-forward neural networks, which are trained with data from rate-based models. This allows one, for example, to include the process productivity, a key performance indicator typically obtained in rate-based models. We show that the equilibrium model reproduces well the results of a sophisticated rate-based model in terms of both temperature and composition profiles for a fixed cycle as well as in terms of process optimization and sorbent comparison. Moreover, we apply the proposed equilibrium model to screen and identify promising sorbents from the large NIST/ARPA-E database; we do this for three different (ultra)diluted separation processes: direct air capture, y _CO₂ = 0.1%, and y _CO₂ = 1.0%. In all cases, the tool allows for a quick identification of the most promising sorbents and the computation of the associated performance indicators. Also, in this case, outcomes are very well in line with the 1D model results. The equilibrium model is available in the GitHub repository https://github.com/UU-ER/SorbentsScreening0D.