Automated Artificial-intelligence Inverse Design of Polymer Membranes for CO₂ Capture and Separation

Abstract

Climate change is mainly due to $ CO_2 $ emissions occurring in energy production and transportation. A set of technologies are being developed to separate and sequester $ CO_2 $ emitted by point sources. Polymer membranes show certain advantages with regards to their storage and disposal properties, they allow for passive operation, have high tolerance to SOx and NOx content and can be integrated within an existing power plant steam cycle, i.e., post combustion application. Candidate materials for application in polymer separation membranes must fulfill two key requirements: high $ CO_2 $ permeability and high $ CO_2/N_2 $ selectivity. However, there is a tradeoff between these two requirements: increasing a material’s permeability decreases its selectivity, and vice-versa. Screening the large number of candidate materials for molecular properties requires an automated design and validation process which does not yet exist. Here, we report our progress in automated discovery of membrane polymers through inverse molecular design. We have computationally generated and validated hundreds of polymer candidates designed for application in post-combustion carbon dioxide filtration. Specifically, we have validated each discovery step, from training dataset creation, via graph-based generative design of optimized monomer units, to molecular dynamics simulation of gas permeation through the polymer membranes. We compare the permeability predictions obtained by AI models and molecular dynamics simulations with experimental results and discuss future extensions of the discovery approach.