Exploratory data analysis reveals the impact of solvent and solute on the performance of native and grafted ceramic membranes in organic solvent nanofiltration

Abstract

Organic Solvent Nanofiltration (OSN) offers an energy-efficient and environmentally friendly alternative to traditional thermal separation methods. Industrial acceptance, however, is hindered by the complexity of membrane–solvent–solute interactions, slowing down the development and optimization of membranes. This study investigates the impact of solvent and solute on permeate flux and retention in native and methyl-grafted ceramic 0.9nm TiO2 membranes. Using over 250 measurements, encompassing 12 organic solvents, binary solvent mixtures, and 24 solutes at varying concentrations, statistical methods such as correlation coefficients were employed to identify trends. The analysis revealed strong correlations between flux and the solvent, with viscosity, polarity (dielectric constant, Hansen solubility parameters), and size (molar volume, kinetic diameter, molecular weight) as key properties. Correlations strengthened in the methyl-grafted membrane, where solute effects on flux were significantly reduced. Retention was more strongly impacted by the solute, especially in the native membrane, with molecular weight and polarity as key properties, while it was less sensitive to the solute after methyl grafting. Additionally, a consistent relationship between retention and flux was revealed in both membranes for fixed solute-solvent mixtures, effectively described by the Spiegler–Kedem model. Moreover, methyl-grafted membranes show optimal size exclusion for various solvent–solute mixtures. These findings demonstrate how the surface modification of methyl grafting leads to unexpectedly large changes in membrane performance: it significantly reduces solute–membrane interactions, simplifies flux behaviour, and leads to more predictable, size-selective behaviour. These insights may help enrich the understanding of the OSN separation process in ceramic membranes and help guide the development of data-driven models.