Speaker
Description
Scaling heterogeneous aperture distributions into equivalent permeability tensors allows for using coarser grids to simulate flow in fractured porous media, significantly reducing computational costs while maintaining accuracy. This work introduces a framework that leverages Conditional Generative Adversarial Networks (CGANs) to upscale the permeability of single fractures efficiently. The framework is tested on three types of aperture distributions: layered media, Zinn & Harvey transformations, and self-affine fractals. The CGAN model predicts pressure distributions within fractures, which are used to compute equivalent permeability tensors. Results demonstrate that this approach accurately captures both the anisotropy and orientation of fracture permeability while achieving substantial computational speed-ups compared to traditional numerical methods. This framework highlights the potential of machine learning to revolutionize modeling practices for fractured porous media and supports the development of efficient multi-scale simulations, bridging the gap between fine-scale physics and large-scale field applications.
| References | Ferreira, C. A. S., Kadeethum, T., Bouklas, N., Nick, H. M. 2022. A framework for upscaling and modelling fluid flow for discrete fractures using conditional generative adversarial networks. Advances in Water Resources, 166, 104264. https://doi.org/10.1016/j.advwatres.2022.104264 |
|---|---|
| Country | Denmark |
| Acceptance of the Terms & Conditions | Click here to agree |
