Facilitating Screening of MOFs for Mixed Matrix Membranes Using Machine Learning and the Maxwell Model

Metal organic framework (MOF)-based mixed-matrix membranes
(MMMs),
which embed MOF particles in polymer matrices, combine the advantages
of polymeric and inorganic membranes. Multiple previous studies have
used the Maxwell model together with molecular simulations and machine
learning (ML) to predict the performance of MOF/polymer MMMs. However,
the assumption of rigid MOF frameworks in molecular simulations limited
the accuracy of the data used in the predictions, particularly in
predicting molecular diffusivities. We developed a novel workflow
integrating ML models with consideration of MOF flexibility to predict
the permeability and selectivity of 131,722 MMMs for CO2/CH4, O2/N2 and He/H2 separations. The full range of achievable MMM performance within
the Maxwell model was analyzed, and several promising MOFs were identified
using this workflow. This approach offers an efficient tool for screening
any polymer and MOF combination in gas separation applications.