Deep Learning for Accurate Prediction of Physical Properties of Heterogeneous Digital Rocks

Abstract

Summary Accurate prediction of the physical properties of heterogeneous porous media based on digital models requires 3D high-resolution (HR) and large-scale images. It is, however, extremely challenging to acquire such images since the current imaging technologies cannot resolve the dilemma between the high resolution and large field of view and we often end up with low-resolution images but with a large field of view or HR images with a small field of view. Moreover, available HR images are limited and always unpaired with accessible low-resolution images. Therefore, we proposed a hybrid unsupervised end-to-end deep learning method to fuse the fine-scale structures from 2D HR images into 3D low-resolution CT images for reconstructing 3D HR and large-scale digital rocks based on limited unpaired training images. The presented method is accurate since the porosity, pore size distribution, multiple-point correlation, and permeability of the reconstructed digital rocks are in good agreement with laboratory measurements.