InterPore2025

Determining Tracer Dispersion Properties in Porous Media Using the Galerkin Physics-Informed Neural Network Approach

19 May 2025, 15:05

1h 30m

Poster Presentation (MS15) Machine Learning and Big Data in Porous Media Poster

Speaker

Mr Luis Constante (Universidade Estadual do Norte Fluminense Darcy Ribeiro)

Description

This study addresses the challenge of modeling tracer dispersion through porous media, applying the Galerkin Physics-Informed Neural Network (Galerkin PINN) approach. The Galerkin PINN method has been systematically evaluated by comparing its results against both analytical solutions and numerical simulations using a Finite Element Method. In this work, experiments were conducted using Berea sandstone, a representative natural porous medium, to validate the model's accuracy in real-world scenarios. A significant aspect of this research involved solving the inverse problem to accurately determine the dispersion coefficient, which is crucial for predicting the behavior of tracers in geophysical and environmental engineering applications. The numerical results demonstrate that the Galerkin PINN approach not only reproduces precisely the empirical data but also offers an efficient alternative since shows an improvement in computational performance and adaptability, making it a promising tool for complex flow and transport problems in heterogeneous porous media.