Carbon capture and storage is a critical element of world efforts to mitigate the effects of climate change. The 2018 Intergovernmental Panel on Climate Change (IPCC) reported that limiting atmospheric warming to less than 2 degrees Celsius will require both capture of carbon emissions from point sources such as power and industrial plants as well as the removal of existing atmospheric carbon dioxide. Amine solvents have gained the most traction in large-scale CCUS implementation, but limitations include high energy penalties associated with regeneration, cost, toxic degradation products, and corrosiveness. Improvements in stability, binding capacity, kinetics, and vapor-liquid equilibria have been achieved by blending different primary, secondary, and tertiary amines, but accelerating their discovery with machine learning and AI has seen limited exploration. There is further need to both measure and predict the ongoing performance of novel amine solvents and mixtures during use and in situ at carbon capture facilities. We have built an apparatus and method for rapid testing and measurement of capacity, kinetics, and thermodynamics of CO2 binding to binary and tertiary blends of standard primary, secondary, and tertiary amines. Our approach replicates gas sparging of CO2 through liquid solvents and leverages NDIR and other sensor modalities in conjunction with temperature swing to rapidly measure CO2 absorption and desorption from complex mixtures of simulated flue gas. The resultant data is processed using physics constrained machine learning methodology to extract the kinetics equations for the reversable solvent reaction. Physical parameters including but not limited to capacity, enthalpy and entropy of activation are directly computed from the discovered reaction kinetic model. With these data sets and associated discovered model, we further leverage machine learning to compute theoretical optimal process parameters for novel solvent blends and rank performance of these mixtures. New blends are tested and validated, and these data are fed back to a chemical predictive model to benchmark and improve its capabilities with the end goal of predicting novel materials and solvent blends with improved performance for carbon capture. The chemical predictive model is based on computational chemistry and AI methods that include but are not limited to structural, topological and chemoinformatic analysis.

Automated discovery of chemical reaction kinetics for CO2 capture solutions

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.