Machine learning-assisted selection of adsorption-based carbon dioxide capture materials

Abstract

Recently, carbon capture has gained increased attention as a sustainable way for mitigating global warming. One of the promising technologies for carbon capture is the adsorption-based process. Accordingly, many researchers are focusing on the development of novel adsorbents with desired characteristics, such as high capacities and selectivities, to accelerate the deployment/design of adsorption-based CO2 capture processes. This paper aims at identifying a group of optimum adsorbents with high capture capacity and selectivity based on the previously published data on different adsorbents. It provides different stakeholders (including experts in process design) with comprehensible and clear human interpretable patterns/rules. This facilitates the selection of optimum adsorbents for carbon capture at the simulation, lab-scale, and pilot-scale levels. Moreover, this work benefits other stakeholders such as researchers in the field of material design by providing the adsorbent characteristics needed to have high capacities and selectivities. Hence, data was collected from different sources including research papers, literature reviews and publicly available datasets. Additionally, an ensemble of diversified (explainable and interpretable) machine learning techniques was utilized to classify these collected data with a classification accuracy above 98%. The extracted rules for achieving the desired characteristics were revised by experts and were found to be appropriate. For instance, it is recommended to design a chemically modified adsorbent with high surface area, large pore volume and small pore diameter to be more selective. The preferred operating conditions are low temperature and low pressure, i.e., near atmospheric. Regarding capacity, high pressures are preferred. Explainable and interpretable machine learning were selected to overcome the black-box prediction problems of commonly used techniques.