Surrogate modeling of pressure loss & mass transfer in membrane channels via coupling of computational fluid dynamics and machine learning

• Published in: Desalination 2023-02, Vol.548, p.116241, Article 116241
• Main Authors: Binger, Zachary M., Achilli, Andrea
• Format: Article
• Language: English
• Subjects: Computational fluid dynamics; Machine learning; Mass transfer; Membrane spacers; Pressure loss; Surrogate model
• ISSN: 0011-9164; 1873-4464
• DOI: 10.1016/j.desal.2022.116241

Spacers are integral to the operation of membrane systems for both structural purposes and improving mass transfer dynamics that drive water permeation at the cost of increased pressure losses. 321 computational fluid dynamics (CFD) simulations were performed to provide high-fidelity data on hydrodynamic and mass transport behavior of spacer-filled membrane channels. CFD simulations were used to investigate the impact of the geometric parameters of spacers on pressure loss and concentration polarization in membrane channels. Spacer designs were characterized using six parameters that were varied in simulations to sample the domain of commercially available designs. Machine learning models were trained on CFD data to produce surrogate models for predicting pressure loss and mass transfer coefficients. These surrogate models consider more geometric parameters than existing empirical equations resulting in more representative and flexible models that can be integrated into existing module-scale or system-scale modeling software. Surrogate models were coupled with a particle swarm optimization algorithm and found that spacer designs with a diameter of 0.3 mm, length of 3.6 mm, angle between 42 and 46°, and moderate diameter necking (~60 %) best balances the trade-off between reduced concentration polarization and increased pressure losses in membrane channels with channel velocities between 0.05 and 0.35 m/s. [Display omitted] •Performed 321 computational fluid dynamics simulations of spacer-filled membranes•Employed realistic representations of non-woven spacers using six geometric parameters•Produced models for predicting pressure loss and mass transfer coefficient•Realized optimal spacer design to minimize pressure loss and maximize mass transfer•Demonstrated the use of machine learning to advance analyses of membrane systems