Modeling flow and transport in fracture networks using graphs

Area of Science:

• Geosciences
• Computational Science
• Fluid Dynamics

Background:

• Subsurface flow in low-permeability rock predominantly occurs through fractures, necessitating accurate modeling for applications like resource extraction and carbon sequestration.
• Discrete fracture network (DFN) models offer high fidelity but are computationally expensive, especially for large-scale systems and uncertainty quantification.
• Existing continuum-based methods lack the detail to capture complex fracture network behavior.

Purpose of the Study:

• To develop and validate a computationally efficient graph-based approach for simulating fluid flow and tracer transport in fractured subsurface systems.
• To assess the accuracy of the graph approach by comparing it against high-fidelity DFN simulations.
• To introduce a bias correction methodology to improve the graph-based approach's predictive capabilities.

Main Methods:

• Flow and transport simulations were performed on a graph representation of discrete fracture networks (DFNs).
• The graph approach was compared to high-fidelity DFN simulations using metrics like breakthrough times and tracer particle statistics.
• A novel bias correction methodology was developed and applied to the graph algorithm.

Main Results:

• The uncorrected graph approach exhibited a consistent bias, predicting breakthrough times up to an order of magnitude slower than DFN models.
• This bias was attributed to the underprediction of pressure gradients at fracture intersections by the graph algorithm.
• The bias-corrected graph approach demonstrated significantly improved accuracy, closely matching DFN predictions.
• The corrected graph method achieved computational speeds approximately 10^4 times faster than DFN simulations.

Conclusions:

• Graph-based modeling offers a computationally tractable alternative to DFNs for subsurface flow and transport simulations.
• The proposed bias correction methodology enhances the accuracy of graph-based predictions, making them suitable for complex fractured systems.
• The efficiency and accuracy of the corrected graph approach make it ideal for uncertainty quantification in subsurface applications.