Developing a Data-Driven Framework for Predicting Retention Profiles in Porous Membranes for Liquid Filtration

Predicting the filtration performance of porous membranes based on their structural characteristics remains difficult for researchers of membrane-based separation technologies. Although 3D porous structures can be acquired using state-of-the-art tomography techniques, volume sizes of the acquired structures are limited by experimental costs.  Acquiring 3D porous structures that span the entire membrane thickness and conducting filtration simulations to predict filtration performance remains prohibitively expensive, forcing researchers to consider alternative solutions.

In this work, we present a data-driven framework for predicting retention profiles in porous membranes for liquid filtration by combining a theoretical model of multi-layer filtration with a model trained using a structure-property database. A database of artificial porous structures was prepared using the GrainGeo module and the filtration performance of each 3D structure was determined using the FilterDict module. Structural parameter profiles obtained from 2D microscopy images were then used as predictors in the model to predict particle retention profiles spanning the entire membrane thickness. Qualitative prediction was demonstrated using this framework; however, further refinements to the methodology are required to enable quantitative reproduction of experimental results.