(394k) Predicting Vapor Liquid Equilibrium Curves of Acid Gas Absorption Using Machine Learning and Molecular Simulations

In the search for new solvents to remove acid gas (CO₂, H₂S, etc.) the use of molecular simulations and chemoinformatics is challenging because of the complexity of the absorption process, and, more specifically, due to its reactive nature. This talk will highlight several of our recent significant advances in computer-aided (mainly machine learning) developments in the field of absorption. The study case is absorption of CO2 and H2S in aqueous MDEA, a widely used industrial solvent.

The key thermodynamic property of acid gas removal is the acid gas-solvent Vapor-Liquid Equilibrium (VLE). These isotherms reflect the amount of acid gas absorbed for a given partial pressure of acid gas at equilibrium conditions. Measuring VLE curves is a time-intensive endeavor necessitating numerous repetitions for various temperatures and solvent compositions. Existing theoretical models for VLE prediction (e.g. e-NRTL) require the knowledge of a wide range of thermodynamic parameters of the studied pure, binary, ternary, etc. systems.

As a solution to this resource-intensive process, machine learning (ML) modeling has been employed to predict acid gas vapor pressure based on its concentration in the absorbent, temperature, and MDEA concentration in H2O. Data was collected from 15 articles, resulting in 512 data points (61 isotherms) for CO2 in MDEA+H2O and 239 datapoints (34 isotherms) for H2S in MDEA+H2O. Data analysis techniques were applied to the available data, and the resulting models were evaluated for their robustness and predictive capability. The model was used to predict new experimental data.The prediction error was deemed acceptable. A methodology is developed to eliminate suspicious experimental data, as well as unphysical trends. The accuracy of the model is also compared with a standard thermodynamic model fitted against experimental data and with the molecular simulation predictions presented in the next paragraph.

We will also briefly present the open-source chemical reaction equilibrium solver (CASpy, https://github.com/omoultosEthTuDelft/CASpy) developed to compute the concentration of species in any reactive liquid-phase absorption system. As a case study, we computed the CO₂ and H₂S absorption isotherm and speciation in an aqueous MDEA solution at 313.15 K and compared the results with available data from the literature. The results show that the computed CO₂ isotherms and speciations are in good agreement with experimental data, demonstrating the accuracy and the precision of our solver.

Selected publications:

1. Mert Polat, Frédérick de Meyer, Céline Houriez, Othonas A. Moultos, Thijs J.H. Vlugt, Solving Chemical Absorption Equilibria Using Free Energy and Quantum Chemistry Calculations: Methodology, Limitations, and New Open-Source Software, Journal of Chemical Theory and Computation, 19, 2616-2629, 2023.
2. Mert Polat, Hirad S. Salehi, Remco Hens, Dominika O. Wasik, Ahmadreza Rahbari, Frédérick de Meyer, Céline Houriez, Christophe Coquelet, Sofia Calero, David Dubbeldam, Othonas A. Moultos, Thijs J.H. Vlugt, New features of the open source Monte Carlo software brick-CFCMC: Thermodynamic integration and hybrid trial moves, Journal of Chemical Information and Modeling, 61 (8), 3752-3757, 2021.