An artificial intelligence framework for separation performance prediction of polymeric pervaporation membranes

Abstract

Pervaporation (PV) is an efficient technology for selective separation of challenging water-based mixtures, which is not feasible by conventional methods such as thermal distillation. Despite its promising application potential, the intricate interactions among various operational parameters and membrane structural characteristics complicate the understanding of the separation processes involved in water–organic mixtures. In this work, we introduce a machine learning (ML)-based predictive framework to forecast the separation performance of polymeric pervaporation membranes in water–ethanol systems through systematically optimizing membrane structures and operating conditions. Through utilizing Shapley additive explanations (SHAP) and partial dependence analysis, the crucial factors influencing separation performance were identified and demonstrated, including membrane effective area, swelling degree, and temperature. Notably, a significant enhancement in separation efficiency was observed when the membrane effective area exceeded 50 cm${}^2$ alongside operational temperature remained between 25 and 40 °C. Our findings demonstrate the dual utility of machine learning in ensuring predictive capability and uncovering the core parameters governing water–ethanol selectivity, establishing fundamental guidelines for rationally designing high-performance pervaporation membranes through parameter optimization.