Computation of effective elastic moduli of rocks using hierarchical homogenization

TL;DR

This paper introduces a hierarchical homogenization approach combined with FFT-based elasticity solving to efficiently compute the effective elastic moduli of rocks from micro-CT images, demonstrating power law error scaling and improved accuracy.

Contribution

It presents a novel hierarchical homogenization method that reduces computational cost and improves accuracy in elastic property estimation of rocks from micro-CT data.

Findings

Error in homogenized elastic constants follows a power law with exponent -1.

Including surrounding materials reduces homogenization error.

Richardson extrapolation improves approximation accuracy.

Abstract

This work focuses on computing the homogenized elastic properties of rocks from 3D micro-computed-tomography (micro-CT) scanned images. The accurate computation of homogenized properties of rocks, archetypal random media, requires both resolution of intricate underlying microstructure and large field of view, resulting in huge micro-CT images. Homogenization entails solving the local elasticity problem computationally which can be prohibitively expensive for a huge image. To mitigate this problem, we use a renormalization method inspired scheme, the hierarchical homogenization method, where a large image is partitioned into smaller subimages. The individual subimages are separately homogenized using periodic boundary conditions, and then assembled into a much smaller intermediate image. The intermediate image is again homogenized, subject to the periodic boundary condition, to find the final homogenized elastic constant of the original image. An FFT-based elasticity solver is used to solve the associated periodic elasticity problem. The error in the homogenized elastic constant is empirically shown to follow a power law scaling with exponent -1 with respect to the subimage size across all five microstructures of rocks. We further show that the inclusion of surrounding materials during the homogenization of the small subimages reduces error in the final homogenized elastic moduli while still respecting the power law with the exponent of -1. This power law scaling is then exploited to determine a better approximation of the large heterogeneous microstructures based on Richardson extrapolation