(498a) Data-Driven Predictions of Complex Mixture Permeation in Polymer Membranes

Separations of nonaqueous or organic-water liquid mixtures by polymer membranes have been identified as a key enabler for reduced carbon emissions in chemical, biochemical, and petrochemical manufacturing processes. The fractionation by membranes is attractive when hybridized with existing separation modalities (e.g., distillations) for hydrocarbon separations, and has the potential to be applied to biobased complex mixtures (e.g., biocrude oils, crude tall oils) that are vulnerable to high temperature operations. A critical gap for this class of membrane separations is the difficulty associated with estimating membrane performance when challenged with a complex mixture. Current approaches rely on laborious and specialized experiment that requires sophisticated and time-intensive analysis to understand the effectiveness of the membrane. To enable rapid and quantitative predictions of the separation of any arbitrary complex mixture using any arbitrary linear polymer membrane, we developed machine learning algorithms capable of predicting guest diffusivities and solubilities based on the chemical structures of the polymer and the solvents in a mixture. These transport properties are parameterized in a Maxwell-Stefan approach to calculate the flux of each molecule in a complex mixture. The entire prediction framework was validated by several separation measurements from binary liquid mixtures to significantly complex mixtures such as real crude oils consisting of 60,000 molecules. The model accurately predicted the fluxes and compositions of permeates from the crude oil fractionation measurements within 6-7 %. This work will open new opportunities to efficiently explore new membrane materials and separation processes not with any experimental pre-inspection but only with easily accessible information of the polymer and mixture. Details of the future direction and potential applications of this work will be discussed.