Neural network based pore flow field prediction in porous media using super resolution

Previous works have demonstrated using the geometry of the microstructure of porous media to predict the ow velocity fields therein based on neural networks. However, such schemes are purely based on geometric information without accounting for the physical constraints on the velocity fields such as that due to mass conservation. In this work, we propose using a super-resolution technique to enhance the velocity field prediction by utilizing coarse-mesh velocity fields, which are often available inexpensively but carry important physical constraints. We apply our method to predict velocity fields in complex porous media. The results demonstrate that incorporating the coarse-mesh flow field significantly improves the prediction accuracy of the fine-mesh flow field as compared to predictions that rely on geometric information alone. This study highlights the merits of including coarse-mesh flow field with physical constraints embedded in it.