Petroleum Engineering · Texas A&M University
I build physical-AI surrogates that simulate subsurface flow in seconds instead of hours — neural operators for reservoir engineering.
Reservoir simulation is the bottleneck in subsurface engineering — a single history-matching study can demand thousands of runs, each hours long on a supercomputer. My work replaces that loop with a single neural operator that learns the physics of Darcy flow directly from data, then predicts pressure and saturation fields on unseen permeability, well locations, well controls, and well counts — including forward extrapolation in time. In practice the surrogate runs on the order of a thousand times faster than a conventional simulator while holding relative error under five percent, which opens the door to real-time history matching, well-placement optimization, CO₂-storage planning, and live reservoir digital twins.
U-FNO surrogates that learn solution operators for parametric PDEs in porous media.
Embedding governing equations into networks to solve and invert reservoir flow.
Two-phase oil–water and single-phase Darcy flow across heterogeneous fields.
Accelerating characterization and production optimization by orders of magnitude.
Fast forecasting of plume and pressure behavior for CO₂ sequestration.
Surrogates fast enough for online interaction between engineers and the subsurface.
Daniel Badawi, Eduardo Gildin
PREPRINT · arXiv:2407.09728
Daniel Badawi, Eduardo Gildin
PREPRINT · arXiv:2309.17345
Full list on Google Scholar.
Open to collaboration on scientific machine learning for the subsurface — surrogate modeling, neural operators, and reservoir digital twins.