Abstract:Determining effective elastic properties of rocks from their pore-scale digital images is a key goal of digital rock physics (DRP). Direct numerical simulation (DNS) of elastic behavior, however, incurs high computational cost; and surrogate machine learning (ML) model, particularly convolutional neural network (CNN), show promises to accelerate homogenization process. 3D CNN models, however, are unable to handle large images due to memory issues. To address this challenge, we propose a novel method that combines 3D CNN with hierarchical homogenization method (HHM). The surrogate 3D CNN model homogenizes only small subimages, and a DNS is used to homogenize the intermediate image obtained by assembling small subimages. The 3D CNN model is designed to output the homogenized elastic constants within the Hashin-Shtrikman (HS) bounds of the input images. The 3D CNN model is first trained on data comprising equal proportions of five sandstone (quartz mineralogy) images, and, subsequently, fine-tuned for specific rocks using transfer learning. The proposed method is applied to homogenize the rock images of size 300x300x300 and 600x600x600 voxels, and the predicted homogenized elastic moduli are shown to agree with that obtained from the brute-force DNS. The transferability of the trained 3D CNN model (using transfer learning) is further demonstrated by predicting the homogenized elastic moduli of a limestone rock with calcite mineralogy. The surrogate 3D CNN model in combination with the HHM is thus shown to be a promising tool for the homogenization of large 3D digital rock images and other random media

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to efficiently determine the effective elastic properties of large - scale digital core images by combining Convolutional Neural Network (CNN) with Hierarchical Homogenization Method (HHM). Specifically, the paper proposes solutions to the following challenges: 1. **High computational cost**: Direct Numerical Simulation (DNS) for calculating the elastic behavior of rocks requires a large amount of computational resources, which is especially obvious when dealing with large three - dimensional digital core images. The method proposed in the paper significantly reduces the computational cost by using CNN as a surrogate model to accelerate the homogenization process. 2. **Memory limitations in handling large images**: 3D CNN models encounter out - of - memory problems when handling large images. The CNN - HHM method proposed in the paper effectively solves this problem by dividing large images into multiple small sub - images, homogenizing these small images using 3D CNN models, and then assembling these homogenized sub - images and using DNS for the final homogenization. 3. **Improving prediction accuracy and generalization ability**: The 3D CNN model in the paper is designed to output two scalars \( f_K \) and \( f_\mu \) between 0 and 1, which correspond to the positions of the bulk modulus and shear modulus of the input image between the Hashin - Shtrikman (HS) upper and lower limits respectively. This design not only improves the prediction accuracy of the model, but also enhances the generalization ability of the model, making the model applicable to different types of rocks. ### Main contributions of the paper 1. **Proposing a new CNN - HHM method**: This method combines 3D CNN models and hierarchical homogenization methods, and can efficiently process large - scale digital core images and determine their effective elastic moduli. 2. **Solving the memory limitation problem**: By dividing large images into small sub - images, the 3D CNN model can run under limited memory, thus expanding the application range of the model. 3. **Improving prediction accuracy and generalization ability**: By using Hashin - Shtrikman upper and lower limits to constrain the output of the model, the prediction accuracy of the model is improved, and the generalization ability of the model for different types of rocks is further improved through transfer learning. 4. **Verifying the effectiveness of the method**: The paper verifies the effectiveness of the CNN - HHM method in processing digital core images of 300×300×300 and 600×600×600 pixels through experiments, and the prediction results are highly consistent with the results of direct numerical simulation (DNS). ### Formula summary - **Hashin - Shtrikman upper and lower limits**: \[ K_{\text{HS}}^{\text{upper}} = K_{\text{mineral}} + \frac{1 - \phi}{\left( K_{\text{pore}} - K_{\text{mineral}} \right)^{-1} + \phi \left( K_{\text{mineral}} + \frac{4}{3} \mu_{\text{mineral}} \right)} \] \[ K_{\text{HS}}^{\text{lower}} = K_{\text{pore}} + \frac{\phi}{\left( K_{\text{mineral}} - K_{\text{pore}} \right)^{-1} + (1 - \phi) \left( K_{\text{pore}} + \frac{4}{3} \mu_{\text{pore}} \right)} \] \[ \mu_{\text{HS}}^{\text{upper}} = \mu_{\text{mineral}} + \frac{1 - \phi}{\left( \mu_{\text{pore}} - \mu_{\text{mineral}} \right)^{-1} + 2\phi \left( K_{\text{mineral}} \right)}

Homogenizing elastic properties of large digital rock images by combining CNN with hierarchical homogenization method

Hierarchical homogenization with deep‐learning‐based surrogate model for rapid estimation of effective permeability from digital rocks

Computation of effective elastic moduli of rocks using hierarchical homogenization

Deep CNNs as universal predictors of elasticity tensors in homogenization

Deep Convolutional Neural Networks Predict Elasticity Tensors and their Bounds in Homogenization

Three-dimensional convolutional neural network (3D-CNN) for heterogeneous material homogenization

DH-Net: Deformed Microstructure Homogenization via 3D Convolutional Neural Networks

Predicting permeability from 3D rock images based on CNN with physical information

Efficient multiscale modeling of heterogeneous materials using deep neural networks

Estimating Elastic Parameters from Digital Rock Images Based on Multi-Task Learning with Multi-Gate Mixture-of-experts

Multiscale fusion of digital rock images based on deep generative adversarial networks

Estimating Two-Dimensional Physical Parameters of Digital Rocks Using Deep Learning

Estimating permeability of 3D micro-CT images by physics-informed CNNs based on DNS

Physical Accuracy of Deep Neural Networks for 2D and 3D Multi-Mineral Segmentation of Rock micro-CT Images

Machine Learning Assisted Prediction of Porosity and Related Properties Using Digital Rock Images

An End-to-end Approach to Predict Physical Properties of Heterogeneous Porous Media: Coupling Deep Learning and Physics-Based Features

A fully scalable homogenization method to upscale 3D elastic media

Deep Convolutional Neural Network-Based Method for Strength Parameter Prediction of Jointed Rock Mass Using Drilling Logging Data

Predictive Soft Computing Methods for Building Digital Rock Models Verified by Positron Emission Tomography Experiments

V P and V S prediction from digital rock images using a combination of U-Net and convolutional neural networks

PH-Net: Parallelepiped microstructure homogenization via 3D Convolutional Neural Networks