Abstract:This paper presents a new deep learning data-driven model for predicting structure dependent pore-fluid velocity fields in rock. The model is based on a Convolutional Auto-Encoder (CAE) artificial neural network capable of learning from image data generated by direct numerical simulations of fluid flow through pore-structures, such as by Lattice Boltzmann or molecular dynamics methods. The main novelty of the model in comparison to previous CAE-based data-driven approaches consists of three parts. The first is a methodology for decomposing the full-domain of the porous media into sub-regions, or "sub-domains", in order to reduce the overall size of the CAE, batch process the sub-domains in parallel, and enable the CAE to learn local and generalizable nonlinear mappings of pore-fluid velocities. The second consists of embedding the finite difference solutions of the incompressible Navier-Stokes and continuity equations into convolutional layers prior to the CAE in order to provide the CAE with knowledge of fluid dynamics physics (PhyFlow). The third main novelty is that the training of the CAE is regularized with a hierarchical loss function that encourages the learning of fluid flow patterns (in a way similar to ranked modes in principal component analysis), ranking from most to least important. This is shown to increase the stability in learning, reduce over-fitting, and promote interpretability of the CAE neural network layers (HierCAE). The comprehensive new data-driven model, which we call the PhyFlow-HierCAE model, is shown to exhibit improved accuracy and generalizability of flow field predictions over conventional CAE models, attributable to the embedded physical knowledge and the hierarchical regularization, as well as realize orders of magnitude speed-ups in computation times as a surrogate for the direct numerical simulations. Examples of training and forward predictions on unseen pore-structures are provided and evaluated for data from Lattice Boltzmann and molecular dynamics simulations of pore-fluid flow. The model is shown to be a fast and accurate emulator (or "surrogate") for predicting effective permeability of unseen pore-structures based on learning from relatively small direct numerical simulation datasets.

A self-supervised learning framework based on physics-informed and convolutional neural networks to identify local anisotropic permeability tensor from textiles 2D images for filling pattern prediction

Permeability Prediction in Nonwoven Geotextiles Using Mathematical Modeling and Numerical Simulation

Numerical prediction strategy of fabric in‐plane permeability considering realistic inter‐ and intra‐tow geometry

Development of a Supervised Learning Model to Predict Length Scale Dependent Permeability of Porous Carbon Composites

Predicting permeability via statistical learning on higher-order microstructural information

Validation of Model Predictions of Pore-Scale Fluid Distributions During Two-Phase Flow

Learning a General Model of Single Phase Flow in Complex 3D Porous Media

Prediction of Water Transport Properties on an Anisotropic Wetting Surface Via Deep Learning.

Physics-based modeling of the effective gas transport properties of single jersey knitted fabrics based on images

Liquid wetting across porous anisotropic textiles

A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media

Physics-agnostic and physics-infused machine learning for thin films flows: modelling, and predictions from small data

GelFlow: Self-supervised Learning of Optical Flow for Vision-Based Tactile Sensor Displacement Measurement

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

From microstructural images to properties - an interpretable deep learning approach to predict elastic-plastic properties of fiber composites

Automated model discovery for textile structures: The unique mechanical signature of warp knitted fabrics

Physics Informed Deep Learning for Flow and Transport in Porous Media

Sensor Fusion for Simultaneous Estimation of In-Plane Permeability and Porosity of Fiber Reinforcement in Resin Transfer Molding

Descriptive modeling of textiles using FE simulations and deep learning

Physics-Informed and Hybrid Machine Learning in Additive Manufacturing: Application to Fused Filament Fabrication

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