Design and prediction of metal organic framework-based mixed matrix membranes for CO2 capture via machine learning

Mixed matrix membranes (MMMs) based on metal organic frameworks (MOFs) have been extensively studied for carbon capture to combat global warming. Here we report the introduction of machine learning to get more insights. Random forest models are first trained by literature data on CO2/CH4 separation, which reveal the optimum MOF structure with pore size > 1 nm and surface area of 800 m(2) g(-1). Then, representative MOFs are blended into Pebax-2533 and polymer of intrinsic microporosity-1 to fabricate MMMs. The membranes demonstrate CO2 separation performances that not only agree well with model prediction, but also exceed the 2008 Robeson upper bound. In addition, knowledge transfer from CO2/CH4 to CO2/N-2 separations shows better agreement with literature data compared to direct modeling, so it enables fast and resource-saving machine learning. This work applies machine learning to solve domain-specific problems and may provide implications for other membrane development.

Automated summary

This study applies machine learning to the development of mixed matrix membranes based on metal organic frameworks, achieving improved CO2 separation performance.