Data-Driven Optimization of Regenerant Formulations for Efficient Regeneration of PFAS-Laden Ion-Exchange Resins

Rational design of regenerants for perfluoroalkyl and polyfluoroalkyl substances (PFAS)-laden ion-exchange resins is hindered by the complex interplay among solution components, resin properties, and PFAS structures. This study applied machine learning (ML) with 916 experimental data points to predict the liquid-phase desorption efficiency (DE) of PFAS-laden resins and identify critical influencing factors. We observed that DE positively correlated with the application of polyacrylic resins and organic solvent concentration, but was negatively influenced by PFAS chain length. Subsequently, the interpretable ML model was employed to design targeted regeneration strategies, guiding a systematic assessment from single-component to multicomponent regenerants. The model revealed that polyacrylic resins achieve high DE (>95%) using single-chloride salts (e.g., 5-10% NaCl) through ion exchange. In contrast, most polystyrene resins require combined alkali-salt solutions to surpass 80% DE by disrupting hydrophobic interactions. Model-proposed optimal formulations for nine resins across four PFAS were experimentally validated, demonstrating a close match between predictions and outcomes with a mean absolute deviation of 10.5-16.7 percentage points. This study offers a transferable and data-driven strategy for optimizing resin regenerant formulations, highlighting interpretable ML's role in advancing efficient recovery practices.