TL;DR
This paper introduces a novel non-local upscaling framework for simulating fluid flow in complex fractured media. By leveraging recent multiscale methods, it constructs an accurate and efficient coarse-grid model that captures the essential physics of flow in both the rock matrix and fracture networks, leading to average pressure values that are physically meaningful.
What Problem Does This Paper Address?
Simulating fluid flow in geological formations, especially those with intricate fracture networks, is computationally demanding. The sheer detail of these formations requires extremely fine computational grids, making large-scale simulations impractical. Upscaling techniques aim to bridge this gap by creating coarser models that retain accuracy while significantly reducing computational cost. Traditional upscaling methods often struggle with the complex interactions between different continua (like rock matrix and fractures) and the non-local effects that fractures introduce. This paper tackles the challenge of developing a rigorous and accurate non-local upscaling framework specifically for multi-continua systems, like fractured media, that can capture these complex interactions efficiently.
Methods
The core of this work lies in a modified Generalized Multiscale Finite Element Method (GMsFEM). The authors propose a novel approach that involves:
- Identifying Continua: The method assumes prior knowledge of individual fracture networks within each coarse grid block. This allows for the definition of distinct "continua" representing the matrix and each fracture.
- Simplified Basis Functions: Instead of complex spectral basis functions, they propose using simple piecewise constant functions for each identified continuum (matrix and fractures). This significantly reduces the degrees of freedom and leads to an upscaled equation with minimal size.
- Local Problems in Oversampled Regions: Local problems are solved within "oversampled regions" (larger than the target coarse block) to compute basis functions. These problems are formulated to minimize local energy while ensuring orthogonality to other continua, effectively localizing the effects of each continuum.
- Non-local Transfer and Effective Properties: The computed local basis functions are used to derive non-local transfer coefficients (transmissibilities) between coarse blocks. These coefficients are non-local because they extend into the oversampled regions, capturing the influence of neighboring areas.
- Upscaled Equation: The resulting upscaled equation uses these non-local transmissibilities and represents average pressures in the matrix and fractures, making the variables physically interpretable.
Datasets
Not discussed in the paper. The authors present numerical results using a synthetic fractured media setup.
Key Insights and Ideas
- Piecewise Constant Basis Functions for Continua: The use of simple piecewise constant basis functions for each continuum (matrix and fractures) is a key simplification that drastically reduces the complexity of the upscaled system while retaining physical meaning in the variables (average pressures).
- Localization of Fracture Effects: The method demonstrates that even though fractures can introduce non-local effects, their influence can be effectively localized within carefully chosen oversampled regions, allowing for efficient computation.
- Non-local Transmissibilities are Crucial: The paper highlights the importance of non-local transmissibilities, which extend beyond the immediate coarse block, to accurately capture the flow behavior in fractured media.
- Physical Interpretability of Variables: A significant advantage is that the upscaled variables directly represent average pressures in the matrix and fractures, which is highly beneficial for practical reservoir simulation.
Key Results
The numerical results demonstrate that the proposed non-local upscaling framework achieves good accuracy.
- Accuracy: The upscaled solutions show good agreement with the averaged fine-scale solutions, with errors decreasing significantly as the oversampling region size increases. For instance, with 4 layers of oversampling, the error can be as low as 0.008% for a coarse mesh size of 1/20.
- Efficiency: The method allows for accurate results with relatively small local regions (oversampling), suggesting computational efficiency.
- Decay Property: The local solutions exhibit a decay property outside the oversampling region, confirming the effectiveness of localization.
- Transmissibility Accuracy: The transmissibility matrices computed with sufficient oversampling (e.g., 4 layers) are shown to be accurate compared to a reference computed using the global domain.
Limitations and Future Work
- Prior Knowledge of Fractures: A significant limitation is the assumption that the fracture networks are known a priori. This is a drawback for cases where fracture characterization is incomplete.
- Generalization to More Complex Systems: The current work focuses on relatively simple fractured media. The authors suggest that future work will involve considering more general and complex fracture systems.
- Analytical Non-local Upscaled Models: The authors plan to explore the analytical derivation of non-local upscaled models, which could further enhance understanding and application.
- Multiphase Flow: The current framework is for single-phase flow. Extending it to multiphase flow would be a valuable future direction.
Summary updated