Seminars and Events
Deep neural networks for modeling fluid flow in underground rocks: applications to enhanced oil recovery and CO2 sequestration
Event Details
Location: CR#1135 Conference Room ISI-MDR
Speaker: Birendra Jha, USC
Abstract: Fluid flow in subsurface reservoirs—relevant to enhanced oil recovery, groundwater remediation, waste disposal, and CO₂ sequestration—is governed by complex hydrodynamic and geomechanical interactions. Viscosity and density contrasts between fluids cause viscous fingering, affecting flow, mixing, and reactions, while fluid injection can induce stress, deformation, and even earthquakes, leading to leakage and reduced storage efficiency. These processes lower recovery rates, limit CO₂ capacity, and increase costs. Physics-based simulations of such coupled systems remain computationally expensive. To address this, we developed deep learning surrogate models to predict viscous fingering, CO₂ leakage, and ground deformation using architectures such as the Fourier Neural Operator (FNO), DeepONet, Vision Transformer (ViT), and U-Net. Trained on physics-based simulation data, these models aim to rapidly and accurately extrapolate to new initial and boundary conditions and rock–fluid properties where traditional simulators are too slow or unreliable.
Webinar Passcode: 152680
Webinar ID: 935 7017 8243