Speaker
Description
The rapid growth of experimental data and imaging information and simulation results has led to increased adoption of machine learning (ML) techniques for studying porous media. The research evaluates current ML applications which analyze flow and transport and chemical reactions in porous and fractured media systems for CCUS and subsurface carbon mineralization and geothermal systems and hydrogen energy storage and remediation and unconventional resource recovery. The focus is exclusively on reviewing published approaches rather than proposing new algorithms or modeling strategies.
The review surveys at the pore scale demonstrate how ML techniques apply to image processing and digital rock physics and microstructure characterization. The research evaluates current deep learning and graph-based and unsupervised and self-supervised learning methods which analyze imaging data to detect pore-scale features and identify complex geometries and determine flow-related properties. The document presents a summary of ML applications which help speed up pore-scale simulations and help determine model parameters for upscaled models.
The review investigates ML-based surrogate models and proxy models and reduced-order representations which scientists use to create approximations of multiphase flow and reactive transport and flow-deformation coupling at bigger measurement sizes. The research investigates how these models have been integrated with modern physics-based simulation platforms which preserve their mass conservation features and their thermodynamic characteristics and physical significance. The research literature presents Physics-informed ML approaches which use governing equations and constraints to build learning formulations.
The review examines all published research which uses machine learning to merge data sets while it discusses the process of uniting experimental and field-based measurements and monitoring data analysis and simulation-data combination methods. The paper reviews applications which study mixing and dispersion and chemical reactions that occur in both homogeneous and broken rock formations while documenting the observed difficulties which stem from biased data and the need to move results between different rock types and fluid patterns and the process of quantifying uncertainties.
The review unites existing research data to show ML success while it shows its present difficulties and its position as an extra method which supports physics-based modeling in porous media research.
| Country | United States of America |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
