Physics-informed convolutional neural networks for fluid flow through porous media
Rafał Topolnicki, Paweł Dłotko, Maciej Matyka · 2026
Read Physics-informed convolutional neural networks for fluid flow through porous media by Rafał Topolnicki, Paweł Dłotko, Maciej Matyka (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/PMC13547363.txt.
Description
Rafał Topolnicki, Paweł Dłotko, and Maciej Matyka trained in 2026 a neural network to predict how fluid moves through the tiny pore spaces of materials like rock and filters. That calculation normally needs slow, mesh-heavy simulation. Their encoder-decoder network reads the solid geometry and outputs the velocity field directly. A custom penalty function nudges predictions to obey physics: no compression of the fluid, no flow inside solid, and consistent wrap-around at the edges.
Tuning the penalty weights one at a time showed what each rule contributes, and the best variant stayed accurate on unfamiliar geometries, porosities, and boundary setups. Used as a starting guess for a standard solver, the predictions sped up convergence in more than ninety percent of cases. The authors offer the network as a fast front end for porous flow work.
Catalog record
| CREATOR | Rafał Topolnicki, Paweł Dłotko, Maciej Matyka |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 105430 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |