Prediction of mass transport properties in 3D microstructures using 2D CNNs

Abstract

Porous materials and the relationship between their 3D microstructure and their mass transport properties is of interest in multiple fields. To analyse this relationship and build an understanding of it requires a great quantity of data, but obtaining experimental 3D data is difficult and expensive. An alternative is to generate virtual microstructures and simulate their mass transports, which can then be used to estimate the relationship. 2D experimental data is easier to obtain and work with than 3D experimental data, e.g. it requires less storage space and memory. It is of interest to investigate models that can estimate mass transport properties of 3D microstructures from 2D data. In this work, 2D data is extracted from a pre-existing 3D virtual microstructure dataset and the viability of using 2D convolutional neural networks (CNNs) to predict the mass transport properties is explored. Keywords: