Eric Cawi

Fracture networks are sets of cracks/pathways through non-porous material such as rock, through which fluid can flow. High fidelity simulations have been developed to compute fluid flow through a fracture network, but they are computationally intensive.

​

The breakthrough curve is a common metric used to analyze fluid flow in fracture networks, and can be computed using a subnetwork called the "Backbone". In this project we use machine learning to identify backbones in networks, and improve on previous work by enforcing physical constraints.

​

The main idea is to represent the fracture network as a graph, and treat that graph as a network of connected paths. We compute path features and then use those to identify Backbone paths. The backbones found by our method are able to capture the breakthrough curve of the networks while significantly reducing the network size.

Full description can be found here.

​

​