Imperial College London
Portrait Picture

ProfessorMartinBlunt

Faculty of Engineering, Department of Earth Science & Engineering

Chair in Flow in Porous Media
 
 
 
//

Contact

 

+44 (0)20 7594 6500m.blunt Website

 
 
//

Location

 

2.38ARoyal School of MinesSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@article{Mosser:2017:10.1103/PhysRevE.96.043309,
author = {Mosser, L and Dubrule, O and Blunt, MJ},
doi = {10.1103/PhysRevE.96.043309},
journal = {Physical Review E},
pages = {1--17},
title = {Reconstruction of three-dimensional porous media using generative adversarial neural networks},
url = {http://dx.doi.org/10.1103/PhysRevE.96.043309},
volume = {96},
year = {2017}
}
Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - To evaluate the variability of multiphase flow properties of porous media at the pore scale, it is necessary toacquire a number of representative samples of the void-solid structure. While modern x-ray computer tomographyhas made it possible to extract three-dimensional images of the pore space, assessment of the variability in theinherent material properties is often experimentally not feasible. We present a method to reconstruct thesolid-void structure of porous media by applying a generative neural network that allows an implicit descriptionof the probability distribution represented by three-dimensional image data sets. We show, by using an adversariallearning approach for neural networks, that this method of unsupervised learning is able to generate representativesamples of porous media that honor their statistics. We successfully compare measures of pore morphology, suchas the Euler characteristic, two-point statistics, and directional single-phase permeability of synthetic realizationswith the calculated properties of a bead pack, Berea sandstone, and Ketton limestone. Results show that generativeadversarial networks can be used to reconstruct high-resolution three-dimensional images of porous media atdifferent scales that are representative of the morphology of the images used to train the neural network.The fully convolutional nature of the trained neural network allows the generation of large samples whilemaintaining computational efficiency. Compared to classical stochastic methods of image reconstruction, theimplicit representation of the learned data distribution can be stored and reused to generate multiple realizationsof the pore structure very rapidly
AU  - Mosser,L
AU  - Dubrule,O
AU  - Blunt,MJ
DO  - 10.1103/PhysRevE.96.043309
EP  - 17
PY  - 2017///
SN  - 1539-3755
SP  - 1
TI  - Reconstruction of three-dimensional porous media using generative adversarial neural networks
T2  - Physical Review E
UR  - http://dx.doi.org/10.1103/PhysRevE.96.043309
UR  - https://journals.aps.org/pre/abstract/10.1103/PhysRevE.96.043309
UR  - http://hdl.handle.net/10044/1/54509
VL  - 96
ER  -
Download