Topological Data Analysis combined with Machine Learning for Predicting Permeability of Porous Media

Ebru Dagdelen 1, Catherin Neena Lalu 2, Aakash Karlekar 1, Manav Arora 1, Matthew Illingworth 1, Jonathan Jaquette 1, Linda Cummings 1, Lou Kondic 1 · 2026

Read Topological Data Analysis combined with Machine Learning for Predicting Permeability of Porous Media by Ebru Dagdelen 1, Catherin Neena Lalu 2, Aakash Karlekar 1, Manav Arora 1, Matthew Illingworth 1, Jonathan Jaquette 1, Linda Cummings 1, Lou Kondic 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-17581.txt.

Introduction

Ebru Dagdelen and colleagues combine topological data analysis with machine learning to predict permeability of porous media. Inputs span geometry traits, connectivity traits and pore-network traits, fed into a standard learning algorithm with separately computed exact values as ground truth. Comparing input sets clarifies which traits help most. Those traits in particular combine easily with learning methods to give meaningful flow predictions from porous structure.

Catalog record

CREATOREbru Dagdelen 1, Catherin Neena Lalu 2, Aakash Karlekar 1, Manav Arora 1, Matthew Illingworth 1, Jonathan Jaquette 1, Linda Cummings 1, Lou Kondic 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44320
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.17581

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Topological Data Analysis combined with Machine Learning for Predicting Permeability of Porous Media Ebru Dagdelen 1, Catherin Neena Lalu 2, Aakash Karlekar 1, Manav Arora 1, Matthew Illingworth 1, Jonathan Jaquette 1, Linda Cummings 1, Lou Kondic 1 Abstract Flow in porous media is difficult to address using standard analytical or numerical methods due to its complexity. However, since synthetic representations of porous media are easy to produce and data from physical experiments are becoming more widely available, the problem is well-suited to studies that include machine learning (ML) techniques. We discuss a number of features that can be extracted from such data, and their utility as input variables into a standard ML algorithm. These features include structural measures describing the geometry of the porous media, topological measures describing the connectivity, and network measures obtained by modeling the porous media as simplified pore networks. These features enable the prediction of the permeability of the considered (synthetic) porous materials using ML techniques that also leverage the separately computed exact permeability (ground truth). Comparing results obtained using different input variables helps develop a better understanding of the utility of various measures for predicting permeability based on the porous media structure. We show, in particular, that topological data analysis (TDA) provides a useful set of features that can be easily combined with ML to yield meaningful results. I Introduction I.1 Motivation and Overview Machine Learning (ML) and Artificial Intelligence (AI) approaches already play an important role in materials science research, and their use is expected to expand significantly in the years to come. The goal of the present work is to provide a simple description of approaches for analyzing porous media, specifically problems involving flow through porous media, which are important in a variety of applied fields. This system is a good candidate for ML-based approaches, since the exact (numerical) solutions are hard to compute due to the computational complexity; and the underlying data (digitized representations of porous media) can be easily generated, as in the present work, or obtained from experimental images. Extensive studies of such porous media systems, both with and without consideration of ML, have been carried out 6 ; 34 ; 35 ; 39 ; the distinguishing feature of the present work is that it (i) presents the techniques used in an accessible manner, avoiding complexities and technical details, which, while possibly important for any particular application, may obscure the big picture; (ii) uses well-documented software packages and libraries that can be implemented with minimal preparation; and (iii) provides detailed, clear, and precise measures that quantify the utility of ML in the present context. In addition, the paper presents new results illustrating the value of topological data analysis (TDA) measures, particularly those emerging from persistence homology (PH) 24 , when applying ML to porous media. To illustrate one possible application of the present work, let us point out that a significant part of the techniques and some preliminary results emerged from a semester-long undergraduate (Capstone 25 ) project in the Department of Mathematical Sciences at New Jersey Institute of Technology; the senior authors hope that the presented paper may be useful to other educators and researchers looking for a material that can be used to introduce students to the ML and TDA techniques, which they may find useful in their future professional lives, either in an academic or industrial environment. I.2 Flow through porous media Permeability is a fundamental property of a porous material that quantifies the ease with which fluid can pass through it under pressure. It plays a crucial role in Darcy’s Law, which states that the volumetric flow rate of a fluid through a porous medium is directly proportional to the pressure gradient and the medium’s permeability, and inversely proportional to the fluid’s viscosity; see 32 for a concise review. Accurate estimation of permeability is essential in a wide range of real-world applications, including groundwater filtration, oil recovery, CO 2 sequestration, and industrial filtration processes 26 . In these examples, the porous material is soil, rock, or some manufactured porous membrane. Porous media have been studied using a variety of modeling approaches to understand how fluids flow through the complex internal pore structures. We note that porous media flow is closely related to percolation theory, see 15 ; 38 for standard physics texts, or 7 for more mathematical perspective. For modeling porous media flow, standard methods include continuum and pore-scale models, which describe fluid transport using principles such as Darcy’s law and the Hagen–Poiseuille equation, and can be extended to capture additional phenomena that may be important in applications, such as pore blockage, adsorption, and the accumulation of deposits over time 5 ; 37 . Pore-network models provide a simplified representation of a porous medium as a network of interconnected pores of simple shape. Being more analytically tractable, such models can yield additional insight, allowing systematic exploration of how pore size, connectivity, and heterogeneity affect overall fluid transport 10 ; 14 ; 19 ; 20 . Traditionally, permeability is measured through laboratory experiments or by simulating fluid flow through digitized images of a porous medium using partial differential equations (PDEs). Although accurate, these PDE-based simulations are computationally intensive, often requiring fine-grained meshes, high-resolution image data, and substantial processing time to converge, especially for large or structurally complex samples. These challenges have motivated growing interest in data-driven alternatives and advanced tools that can efficiently and accurately account for realistic pore structures 3 ; 18 ; 34 . One approach proposed recently applies persistent homology (PH) (see 31 for a recent, accessible review, and 9 ; 12 for more in-depth studies), a tool from topological data analysis (TDA), to study the structure of pore networks 23 ; 39 . Such methods provide a potential foundation for data-driven approaches, including machine learning (ML), to efficiently predict permeability in complex porous materials. In this work, we investigate whether an ML model trained on geometric, topological, and network-based descriptors of a digitized representation of a porous medium can accurately predict its permeability. We begin our workflow by generating 1000 synthetic porous structures using the Porous Microstructure Analysis (PuMA) software package 16 . For each of these structures, we compute permeability directly using PuMA’s built-in capabilities. We then extract pore network representations from the structures using the SNOW2 algorithm implemented in PoreSpy 17 . Moreover, for each structure and its associated pore network, we compute a comprehensive set of descriptors that capture geometric complexity, topological features, and network connectivity. The combined structural, topological, and network descriptors are then used to train the ML model to predict permeability. Permeability prediction using ML can significantly reduce the computational cost and time of traditional methods while retaining physical interpretability. The objectives of the study are (i) to assess the predictive power of these structural, topological, and network descriptors, and (ii) to evaluate the computational efficiency of ML predictions relative to direct numerical simulations. Here, accuracy is measured by comparing with permeabilities computed with PuMA, while efficiency is evaluated by total computation time. Whenever possible, we attempt to describe the techniques used and the results in a simple yet detailed manner, allowing for a reasonably straightforward understanding and reproducibility. II Data Generation To facilitate ML-based permeability

Read the complete text