Hybrid CFD and Deep Learning Framework for Fouling and Oil-Rejection Prediction in Membrane Oil–Water Separation

Membrane separation is effective for treating oily wastewater, yet predicting fouling and oil rejection remains challenging due to variability in membranes and operating conditions. This work presents an integrated framework combining computational fluid dynamics (CFD), reduced-order modelling (ROM), and machine learning to analyse fouling dynamics and separation performance. CFD simulations evaluated hydrodynamics and particle deposition in tubular and flat membranes, while a Python-based ROM reproduced experimental flux decline with high accuracy (R² ≈ 0.95). Machine learning models predicted oil rejection from system descriptors (R² = 0.84). The proposed approach provides an interpretable and scalable tool for data-driven membrane design.