Future Challenges of Modeling THMC Systems

Fracture-fluid flow coupling is critical for understanding society-relevant energy sources such as Enhanced Geothermal Systems (EGS), hydrocarbon extraction via hydraulic fracture, and carbon capture and storage (CCS). Although fluid-rock interactions are one of the most important processes in lithospheric geodynamics, they are one of the most difficult to properly simulate because of the orders-of-magnitude changes in hydraulic properties at the onset of fracture or slip. These large-scale changes in hydraulic properties result in the locally rapid flow that is likely to react with their new surroundings. Advances in modeling these processes must properly modeling the evolution of fracture networks coupled to the hydraulic (and thermodynamic) properties within the advecting, over-pressured, and reactive system.

Many THM simulators exist but with limited utility because their inherently low resolution do not allow for the development of evolving fracture networks. A promising approach to modeling is to develop algorithms written in the CUDA programming language and optimized for computations using the inherently parallel Graphics Processing Unit (GPU) computer architecture. We are developing such a model, with target resolutions on the order of 1000x1000x 500. This simulator models a pore-elastic-thermo-elastic-plastic rheology that includes hardening, softening, and damage, and coupled to a non-linear diffusion model of fluid pressure and a 2-phase fluid (water and gas). Both tensile and shear fractures evolve in response to far-field stresses, local stress perturbations associated with both shear and tensile fractures, and thermo-elastic and pore-elastic stresses. The advantage of this model over existing models is that the governing equations have been reformulated to run on a GPU cluster.

Results from numerical experiments show that this approach has great potential to study fluid-rock interactions at all scales. In this talk, I summarize the progress to date and future plans, and show that a GPU-based modeling approach is fast, high-resolution, and can reproduce experimental results of fluid injection experiments with numerical resolution at the grain-scale of the experiment.