A fundamental challenge in the research of geologic subsurface flow is to upscale laboratory-measured, core-scale rock and flow properties to the field scale. We aim to combine physics with data analytics to address this fundamental challenge by developing novel machine-learning and data-assimilation based frameworks.
Super-resolution-assisted pore flow field prediction using neural networks
Recent neural-network-based pore flow prediction using only porous media geometry can cause ill-posedness and has poor extrapolation capability. We proposed incorporating a coarse velocity field in the input to effectively improve the prediction performance, especially for those with a large degree of heterogeneity.
X.-H. Zhou, J. E. McClure, C. Chen, and H. Xiao. Neural network--based pore flow field prediction in porous media using super resolution, Physical Review Fluids, 7, 074302 (2022).
Predicting the permeability from porous media images with convolutional neural networks
Calculating the permeability of porous media using direct pore-scale simulation is often expensive for realistic systems. We proposed using a physics-informed convolutional neural network to predict the permeability directly from porous media images.
J. Wu, X. Yin, and H. Xiao. Seeing permeability from images: Fast prediction with convolutional neural networks. Science Bulletin, 63, 1215 (2018).