Speaker
Description
Understanding how pore-scale structure controls flow and transport in porous media remains a central challenge in pore-scale modeling and upscaling. While pore network models provide a physically grounded framework to simulate flow and transport, isolating the combined effects of geometric and topological heterogeneity, finite network connectivity, and structural disorder on velocity distributions and nonlinear transport behavior remains difficult. In this work, we use machine learning as a diagnostic and analysis tool, rather than a surrogate model, to systematically identify the structural characteristics of pore networks that govern flow and transport responses.
Large ensembles of synthetic pore networks are generated with controlled variations in coordination number, throat radius distributions, throat length distributions, and network connectivity. For each network, single-phase flow and advective-diffusive transport are simulated using pore network models, from which flow and transport metrics characterizing flow heterogeneity and transport nonlinearity, such as velocity and flow-rate distributions, dispersion coefficients, spatial moments, and breakthrough curve scaling, are extracted.
Interpretable machine learning models are then trained on statistical, geometric, and topological descriptors of the networks to analyze structure-property relationships. Feature importance and sensitivity analyses are used to identify dominant structural parameters and interactions controlling flow heterogeneity, preferential channeling, and the shape of transport distributions. By explicitly combining physics-based simulations with interpretable machine learning, this work provides new insight into the physical mechanisms by which pore-scale structure, connectivity, and finite-size effects influence nonlinear flow and transport, and demonstrates how machine learning can be used to support, rather than replace, traditional pore-scale modeling approaches.
| Country | Spain |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
