Machine Learning‐Guided Design and Synthesis of Eco‐Friendly Poly(ethylene oxide) Membranes for High‐Efficacy CO₂/N₂ Separation

Abstract Machine learning (ML)‐guided polymer design and synthesis will enable the next‐generation membrane material discovery for CO 2 capture. Herein, ML is leveraged to establish a structure‐performance relationship for the eco‐friendly poly(ethylene oxide) (PEO) membrane and guide its design for high‐efficacy CO 2 /N 2 separation. Through a rational fragment representation method and knowledge sharing across membranes fabricated by different methods, the precise prediction of CO 2 /N 2 separation performance for PEO membranes with high Pearson correlation coefficients (0.973 for permeability and 0.875 for selectivity) despite data scarcity is demonstrated. Expertise knowledge and external monomer databases are then utilized in a human‐in‐the‐loop workflow to effectively explore high‐performance PEO membranes in the design space. Several discovered thermally crosslinked PEO membranes achieve CO 2 /N 2 separation performances close to the 2019 Robeson upper bound, which are promising for practical large‐scale carbon capture applications. Model interpretation techniques are employed to provide data‐driven insights into the design of PEO membranes for high‐efficacy CO 2 /N 2 separation. Further life cycle assessment results reveal the outstanding advantage of discovered PEO membranes in terms of environmental friendliness. The work highlights the enormous potential of ML in expediting the discovery of high‐performance carbon capture membrane materials.