Data-driven investigation of process solvent and membrane material on organic solvent nanofiltration

Abstract

Organic solvent nanofiltration (OSN) studies are largely limited to small and specialized datasets, hindering the investigation of broader relationships and contexts. Larger datasets have recently emerged but they are limited to a single membrane and few solvents. To improve the understanding of solute rejection in OSN, we introduced a large dataset containing 1938 rejection values derived from three membranes and ten industrially relevant green solvents. We examined two polydimethylsiloxane membranes, namely, GMT-oNF-2 and Solsep 030306, and a custom polybenzimidazole membrane. Structure–property relationship methods were used to identify the connections between the performance of membranes, solvents, and solutes. We observed polarity selectivity, which was explained using the classical solution diffusion model, and demonstrated the translation of the rejection database into the corresponding rejection selectivity dataset to characterize separation performance. The obtained rejection selectivity data enabled the process-oriented analysis of solvent and membrane characteristics. Our selectivity-based investigation highlighted the inadequacy of the solute molecular weight to properly characterize membrane material and separation performance. Consequently, our findings support the need for more comprehensive modeling approaches for rejection and process performance prediction, while providing process-oriented insights into the performance of OSN membranes.