The subsoil is a porous medium capable of trapping fluids within its volume. This characteristic can be exploited, for example, in $\text{CO}_2$ capture and sequestration processes. The subsoil is also characterized by the presence of fractures and faults, which represent discontinuities in the rock. Accurate measurements of subsurface properties are inherently complex. Therefore, our knowledge of the subsurface and the associated phenomena is subject to uncertainties. Numerical simulations are essential tools for studying and predicting site phenomena, particularly for assessing the safe exploitation of the subsoil. A primary challenge arises in the simulation of fluid flow. Numerical solvers typically encounter difficulties in simulating multiphase flow in porous media, primarily due to non-linear solver convergence issues. It is also important to accurately model the interaction between fluid flow and the fractures or faults in the medium since the latter highly influence the flow paths. A second challenge is constituted by the uncertainties related to subsoil, as they lead to a multitude of possible scenarios that must be taken into account. Standard numerical solvers are generally too computationally expensive to simulate all possible scenarios. In this work, we address the simulation of two-phase flows in domains containing fractures and faults, with a particular focus on the countercurrent flow regime which is one of the most challenging regimes for this type of flow. We extend a finite volume discretization method to mixed-dimensional domains. To handle the inherent uncertainties, we employ data-driven reduced order model techniques based on the use of neural networks. We study single-phase flow in fractured porous media, where both geometric and physical properties are uncertain. We compare the performance of this approach with the well-established proper orthogonal decomposition method for single-phase flow problems. Subsequently, we extend the focus to the time-dependent two-phase flow problem and investigate appropriate neural network architectures for these scenarios. Several case studies are presented to thoroughly evaluate the strengths and limitations of the proposed methods. We show results for single-phase flow, two-phase flow, multi-query applications, and a realistic industrial scenario of interest. This research is conducted in collaboration with Eni S.p.A.
Il sottosuolo è un mezzo poroso capace di intrappolare fluidi al suo interno. Questa caratteristica può essere sfruttata, ad esempio, nei processi di cattura e sequestro di $\text{CO}_2$. Il sottosuolo è anche caratterizzato dalla presenza di fratture e faglie, che rappresentano discontinuità nella roccia. Le misurazioni accurate delle proprietà del sottosuolo sono intrinsecamente complesse. Pertanto, la nostra conoscenza del sottosuolo e dei fenomeni associati è soggetta a incertezze. Le simulazioni numeriche sono strumenti essenziali per studiare e prevedere i fenomeni del sito, in particolare per valutare lo sfruttamento sicuro del sottosuolo. Una delle principali sfide riguarda la simulazione del flusso di fluidi. I solutori numerici incontrano tipicamente difficoltà nella simulazione di flussi multifase in mezzi porosi, principalmente a causa di problemi di convergenza del solutore non lineare. È anche importante modellare accuratamente l'interazione tra il flusso di fluidi e le fratture o faglie nel mezzo, poiché queste influenzano fortemente le traiettorie del flusso. Una seconda sfida è costituita dalle incertezze legate al sottosuolo, che implicano una moltitudine di scenari possibili da considerare. I solutori numerici standard sono generalmente troppo costosi dal punto di vista computazionale per simulare tutti gli scenari possibili. In questo lavoro, affrontiamo la simulazione di flussi bifase in domini contenenti fratture e faglie, con particolare attenzione al regime di flusso controcorrente, uno dei regimi più complessi per questo tipo di flusso. Estendiamo un metodo di discretizzazione a volumi finiti a domini a dimensione mista. Per gestire le incertezze intrinseche, impieghiamo tecniche di modelli a riduzione d'ordine basate sui dati, utilizzando reti neurali. Studiamo il flusso monofase in mezzi porosi fratturati, dove sia le proprietà geometriche che fisiche sono incerte. Confrontiamo le prestazioni di questo approccio con il metodo ben consolidato della proper orthogonal decomposition (POD) per problemi di flusso monofase. Successivamente, estendiamo il focus al problema del flusso bifase dipendente dal tempo e analizziamo le architetture di reti neurali più appropriate per questi scenari. Vengono presentati diversi studi di caso per valutare a fondo i punti di forza e le limitazioni dei metodi proposti. Mostriamo risultati per flussi monofase, flussi bifase, applicazioni multi-query e uno scenario industriale realistico di interesse. Questa ricerca è condotta in collaborazione con Eni S.p.A.
Flow and mechanics in fractured porous media: from high fidelity models to efficient reduced order solutions
Ballini, Enrico
2024/2025
Abstract
The subsoil is a porous medium capable of trapping fluids within its volume. This characteristic can be exploited, for example, in $\text{CO}_2$ capture and sequestration processes. The subsoil is also characterized by the presence of fractures and faults, which represent discontinuities in the rock. Accurate measurements of subsurface properties are inherently complex. Therefore, our knowledge of the subsurface and the associated phenomena is subject to uncertainties. Numerical simulations are essential tools for studying and predicting site phenomena, particularly for assessing the safe exploitation of the subsoil. A primary challenge arises in the simulation of fluid flow. Numerical solvers typically encounter difficulties in simulating multiphase flow in porous media, primarily due to non-linear solver convergence issues. It is also important to accurately model the interaction between fluid flow and the fractures or faults in the medium since the latter highly influence the flow paths. A second challenge is constituted by the uncertainties related to subsoil, as they lead to a multitude of possible scenarios that must be taken into account. Standard numerical solvers are generally too computationally expensive to simulate all possible scenarios. In this work, we address the simulation of two-phase flows in domains containing fractures and faults, with a particular focus on the countercurrent flow regime which is one of the most challenging regimes for this type of flow. We extend a finite volume discretization method to mixed-dimensional domains. To handle the inherent uncertainties, we employ data-driven reduced order model techniques based on the use of neural networks. We study single-phase flow in fractured porous media, where both geometric and physical properties are uncertain. We compare the performance of this approach with the well-established proper orthogonal decomposition method for single-phase flow problems. Subsequently, we extend the focus to the time-dependent two-phase flow problem and investigate appropriate neural network architectures for these scenarios. Several case studies are presented to thoroughly evaluate the strengths and limitations of the proposed methods. We show results for single-phase flow, two-phase flow, multi-query applications, and a realistic industrial scenario of interest. This research is conducted in collaboration with Eni S.p.A.File | Dimensione | Formato | |
---|---|---|---|
thesis_enrico_ballini.pdf
solo utenti autorizzati a partire dal 14/01/2026
Descrizione: pdf tesi
Dimensione
17.89 MB
Formato
Adobe PDF
|
17.89 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/232512