Decoupling solvent features to address spurious correlations in ceramic Organic Solvent Nanofiltration membranes

Abstract

Organic Solvent Nanofiltration has emerged as an energy-efficient alternative to traditional thermal methods; yet its widespread implementation is hindered by its poorly understood transport mechanism. Improving the prediction of membrane flux and retention is therefore essential, which has recently been primarily advanced through data-driven modeling. Prediction and interpretation of organic solvent nanofiltration transport, however, are complicated by the collinearity between solvent size and Hansen solubility, as they are dependent on one another for common organic solvent nanofiltration solvents. Considering both solvent size and solubility are known to correlate strongly with flux, the collinearity of these properties obscures the impact of either of them. We break this collinearity by performing flux measurements on outlier solvents in unmodified and methyl-grafted ceramic membranes at room temperature. Breaking collinearity is achieved using propylene carbonate and glycerol-water mixtures, whose molecular size and solubility (at room temperature) significantly deviate from the trend established by common organic solvent nanofiltration solvents. To quantitatively identify the true driver of flux for each membrane separately, linear models using either solvent kinetic diameter, molar volume, or Hansen solubility as predictors are compared using statistical tests. Our analysis indicates that molecular size (specifically, the squared reciprocal kinetic diameter) is the true predictor of flux for unmodified and methyl-grafted titania membranes separately, while molar volume and Hansen solubility add no further predictive power. This method can be extended to, e.g., investigate membrane and temperature dependence. Decoupling solvent size and Hansen solubility can help improve the understanding of organic solvent nanofiltration transport and, via dimensionality reduction, aid the development of data-driven modeling.