Tuning Pressure Drop in Isoporous Membranes: Design with Fabrication Variability

Abstract

AbstractIsoporous membranes consist of well‐defined micro and nanoscale pore architecture comprising uniform pore sizes with straight pore channels. In contrast to traditional random porous membranes with tortuous flow paths, isoporous membranes offer the opportunity to achieve a high degree of membrane customization and low pressure drop. Here, a physics‐based machine learning methodology that enables the predictive design of a single‐layer isoporous membrane in terms of the pressure drop is reported. In short, the methodology consists of a hybrid approach that includes experimental data on the variability of the pore architecture and the resulting pressure drop, training of a neural network with data from validated physics‐based simulations of laminar flow through the membrane, and Monte Carlo simulation (MCS) to stochastically account for the inherent variabilities of the pore architecture of fabricated isoporous membranes. Overall, the neural network and MCS predict the range of Δp for a given single‐layer membrane well. Experimental values fall within 90% of the minimum and maximum predicted Δp values. In addition, a sensitivity analysis with MCS is carried out to quantify how design and operating parameters affect the overall pressure drop. The methodology can be extended to membranes comprising multiple layers and to account for filtration efficiency.