Mary Wheeler fracturing image
Joining and branching of hydraulic fractures (blue) in porous media using a phase field approach where the initial fractures are formulated by a given probability map.

Professor Mary Wheeler, director of the ICES Center for Subsurface Modeling and holder of the Ernest and Virginia Cockrell Chair in Engineering, with her collaborators has received a $1.5 million grant from the National Science Foundation to develop computational techniques that more effectively use big data to predict and model the pathways of naturally-occurring ground fractures and how induced fractures interact with them.

Wheeler is the lead principal investigator. Her co-investigators are Mrinal Sen, an ICES affiliate faculty member and holder of the John A. and Katherine G. Jackson Chair in Applied Seismology at UT's Institute for Geophysics; Sanjay Srinivasan, holder of the Leone Family Chair in Energy and Mineral Engineering at Pennsylvania State University; and Manish Parashar, Distinguished Professor of Computer Science at Rutgers University.

The injection of large volumes of fluids into the ground, such as during geologic sequestration of CO2 or hydraulic fracturing operations, creates fractures that can deform rock in the vicinity of the injection site. These deformations can trigger seismic effects near the site or alter the elevation of the landscape, a phenomena known as surface deflection. The network of natural fractures has an important effect on these types of responses, said Wheeler, either as locations along which micro-seismic events occur or by acting as conduits for fluid flow that influence the direction and extent of surface deflection.

The NSF grant will help fund methods that will predict potential side-effects by simulating the injection site, and how planned fractures will likely interact with natural topography at the time of injection as well as over long time periods. Wheeler says it’s a challenge that requires integrating information on the fracture environment with the physical processes of fracturing and fluid flow over various scales of time and space.

To do that requires computing immense amounts of data.

“These fields could be many kilometers in size, the timescale is hundreds of years, the physics can occur at the nanoscale…so you have many equations, very complicated flow, multi-phase, multi-component phase behavior, geomechanics and plasticity, wells, chemical reactions, thermal,” Wheeler said. “How do you incorporate all of this in a model? It sounds near impossible.”

Because of the computing challenges inherent in creating such detailed models, a major focus of the research is how GPUs in high performance computers, such as those at the Texas Advanced Computing Center, can be more efficiently used to process data used to create fracture models. Conventionally, high performance computing has relied on leveraging the power of thousands of CPUs, which process code in a serial manner. GPUs offer more even more processing efficiency by processing code in parallel.

Wheeler says that the research team is working with Dr. Joe Eaton, a software developer at NVIDIA and former Ph.D. student of Wheeler’s, to investigate how GPUs and the parallel algorithms that run on them can boost computing efficiency.

“You really need heterogeneous computers,” Wheeler said. “We need more efficient ways of doing computing and Joe’s methods are really state-of-the-art.”

The fracturing models Wheeler and the research team are developing will rely on a mix of experimental data and data taken from real fracturing sites. InSAR (Interferometric Synthetic Aperture Radar) satellite, can give researchers a general idea of a site’s natural fractures. Wheeler utilizes probability software to predict the location of natural fractures that imaging techniques could not accurately account for. From there, she and her team can introduce phase-field fracture modeling into the field and simulate the interactions between an induced fracture and naturally occurring ones. Probability and uncertainty quantification algorithms are additional tools used to validate the accuracy of the model.

In addition to accounting for the interactions between fractures, the research will also seek to model how seismic waves propagate in fractured media—an important factor that can help pinpoint where rock destabilizations, and potential earthquakes, could take place, Wheeler said.

“The geophysicists will help us characterize the subsurface, and then we’ll do flow, mechanics, thermal,” Wheeler said. “There has to be a coupling between these two, because their work depends on the parameters we see, while we in turn will be looking in their characterization of the subsurface.”