Speaker
Description
Accurate simulation of two-phase flow in porous media is crucial for understanding fluid dynamics in reservoir engineering, particularly in SCAL experiments such as counter-current spontaneous imbibition (COUCSI). However, traditional numerical methods are computationally expensive and often struggle to efficiently handle flow uncertainties.
In this work, we explore the application of machine learning-based neural operator techniques, specifically Fourier Neural Operators (FNOs), to model two-phase flow dynamics in COUCSI experiments by predicting spatiotemporal saturation and pressure fields. The FNO framework, which operates in the frequency domain, is well-suited for capturing the nonlinearities of the governing partial differential equations (PDEs) while offering strong generalization across a wide range of system properties.
By training the FNO on a dataset of high-fidelity numerical simulations of COUCSI experiments, we demonstrate its ability to accurately predict fluid saturation profiles and imbibition rates across various rock/fluid properties and boundary conditions. The proposed FNO-based surrogate model achieves a significant computational speed-up over traditional numerical solvers, enabling rapid evaluation of COUCSI processes in many-query scenarios such as uncertainty quantification and parameter estimation from experimental data.
Our results indicate that FNOs generalize well to unseen geological configurations and fluid properties, making them a powerful tool for accelerating SCAL analyses and enhancing our understanding of two-phase flow in porous media. This approach holds promise for optimizing experimental design in core-scale studies and could be readily extended to more complex flow scenarios.
| Country | Norway |
|---|---|
| Student Awards | I would like to submit this presentation into both awards |
| Acceptance of the Terms & Conditions | Click here to agree |
