Machine learning-driven multi-objective optimization of electrospun nanofibrous membranes design for membrane...

Electrospinning is a versatile route to fabricate membranes for membrane distillation (MD) with tunable structures and properties that govern flux and wetting resistance. Achieving target MD-relevant properties requires coordinated control of interconnected solution and processing conditions. Conventional trial-and-error methods are fundamentally limited in their ability to optimize such a vast and non-linear parameter space. Here, we develop a holistic machine learning workflow that encompasses dataset construction, model training and tuning, and inverse design to predict and optimize key attributes of electrospun membranes, including thickness, contact angle, fiber diameter, and fiber morphology. Feature-contribution analysis reveals how polymer-solution descriptors and operating parameters quantitatively influence membrane properties. By coupling the model with Bayesian optimization, we enable data-guided selection of polymer systems and process windows from a candidate space. Membranes prepared under the suggested conditions matched the predicted properties. Overall, this provides a general and extensible framework for forward prediction and inverse design of electrospun membranes, supporting rapid screening and optimization for MD and related separations.