Neural Network Builds 3D Model of Porous Material From a Single Image
Researchers at the Moscow Institute of Physics and Technology (MIPT) and Sichuan University have trained a neural network to reconstruct a three-dimensional model of a porous material from a single two-dimensional image of a cross-section.
The SWNN method combines neural networks with sliced Wasserstein distance and can handle volumes containing hundreds of millions of voxels. The results were published on ScienceDirect.
Porosity can be specified as an input parameter, while heterogeneous structures can be generated using a mask. In terms of thermal conductivity and fluid permeability, the models are virtually indistinguishable from real samples. They also outperform diffusion and generative adversarial networks in reconstruction quality while requiring less computing power.
The method could be useful for developing porous anodes, catalysts and filters, as well as for modeling oil and gas reservoir rocks.
