Accelerating Multiphase Simulations With Denoising Diffusion Model Driven Initializations

• Published in: Journal of geophysical research. Machine learning and computation Vol. 1; no. 4
• Main Authors: Chung, Jaehong, Marcato, Agnese, Guiltinan, Eric J., Mukerji, Tapan, Viswanathan, Hari, Lin, Yen Ting, Santos, Javier E.
• Format: Journal Article
• Language: English
• Published: United States American Geophysical Union (AGU) 01.12.2024; Wiley
• Subjects: diffusion models; generative modeling; lattice‐Boltzmann; physics; pore‐scale
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2024JH000293

This study introduces a hybrid fluid simulation approach that integrates generative diffusion models with physics‐based simulations, aiming at reducing the computational costs of flow simulations while still honoring all the physical properties of interest. Pore‐scale simulations enhance our understanding of applications such as assessing hydrogen and CO2 ${\text{CO}}_{2}$ storage efficiency in underground reservoirs. Nevertheless, they are computationally expensive and the presence of non‐unique solutions can require multiple simulations within a single geometry. To overcome the computational cost hurdle, we propose a method that couples generative diffusion models and physics‐based simulations. While training the data‐driven model, we simultaneously generate initial conditions and perform physics‐based simulations using these. This integrated approach enables us to receive real‐time feedback on a single compute node equipped with both CPUs and GPUs. By efficiently managing these processes within a single compute node, we can continuously monitor performance and halt training once the model meets the specified criteria. To test our model, we generate realizations in a real Berea sandstone fracture which shows that our technique is up to 4.4 times faster than commonly used flow simulation initializations. Plain Language Summary Pore‐scale simulations enhance our understanding of how fluid moves in the subsurface, they have recently been used for evaluating hydrogen storage and carbon dioxide sequestration projects in underground reservoirs. However, these simulations are often limited by their high demand for computational resources, posing a challenge to efficient execution. To address this, we propose a method that integrates artificial intelligence (AI) with physics‐based simulators. After training our AI model with synthetic data, we tested our approach using a real reservoir sample obtained through X‐ray imaging. This allowed us to showcase the computational efficiency of our method. Key Points We introduce a method that integrates generative AI with multiphase simulations improving efficiency while maintaining physical accuracy We propose simulation metrics to evaluate model performance, ensuring alignment with the targeted physical phenomena We demonstrate a significant speed‐up over traditional initialization methods in synthetic and real (x‐ray) fracture data sets