Membrane & Separation Processes

Research Contents

• CFD → PINN workflow: generate channel-level datasets and train PINNs that respect continuity/momentum constraints.

• Geometry & operation optimization: co-design spacer patterns and operating windows for flux, pressure drop, and fouling risk.

• Risk & sustainability: predict fouling/cleaning trade-offs and connect to TEA/LCA for process-level decisions.

• Digital twin & scale-up: translate channel models to module/system design and real-time monitoring.

Membrane processes are widely deployed for water treatment and CCUS, yet their design is hampered by complex hydrodynamics inside spacer-filled channels where turbulence, pressure drop, and near-wall transport interact to drive fouling and energy consumption. Empirical correlations often fail to generalize across geometries and operating windows, which complicates module scale-up and control. To address these limitations, we couple high-fidelity CFD with Physics-Informed Neural Networks that learn flow fields and mass-transfer while enforcing physical constraints, thereby capturing near-wall dynamics more accurately than purely data-driven surrogates. Sensitivity analysis links spacer geometry and operating conditions to shear, concentration polarization, and pumping penalties, enabling multiobjective optimization that balances flux, energy use, and fouling risk. The resulting models form the core of digital-twin workflows that translate channel-scale insight into module and plant decisions, providing actionable guidelines for high-flux, energy-efficient, and fouling-resilient separations.

Associated members: Sungjin Bae, Jongwoo Kim