# A DuMux Framework for Data-Driven Multi-Scale Parametrizations

## TL;DR

This paper introduces a framework that uses data-driven methods, including neural networks and pore metrics, to efficiently determine multi-scale parameters for porous media simulations, reducing the need for costly pore-scale calculations.

## Contribution

It presents a novel four-step framework combining pore data, neural networks, and metrics for improved multi-scale parametrizations in porous media modeling.

## Key findings

- Framework effectively estimates parameters from pore data
- Neural network generalizes pore-scale solutions
- Enhanced understanding of pore content through metrics

## Abstract

Presented in this work is a framework for the data-driven determination of multi-scale porous media parametrizations. Simulations of flow and transport in a porous medium at the REV scale, although efficient, require well defined parameters that represent pore-scale phenomena to maintain their accuracy. Determining the optimal parameters for this often require expensive pore-scale calculations. This work outlines a series of four steps where these parameters can be calculated from pore scale data, their solutions generalized with a convolutional neural network, and their content better understood with descriptive pore metrics.