Porous media flows are ubiquitous in natural and engineered systems, where pore-scale dynamics fundamentally govern macro-scale transport behavior. Accurate flow field prediction is essential for evaluating, optimizing, and controlling such systems. However, realistic porous structures feature multiscale heterogeneity spanning several orders of magnitude, rendering high-fidelity pore-scale simulations computationally intractable for system-scale applications. Moreover, conventional upscaling approaches struggle in heterogeneous media where representative elementary volumes are ill-defined. To address these challenges, we propose a hybrid-coupled multiscale framework that explicitly models the dynamic coupling between pore-scale and Darcy-scale processes. Within this framework, a lightweight deep neural network is used to rapidly predict pore-scale flow fields under evolving boundary conditions. The results show that the proposed deep learning model effectively learns high-fidelity, physically consistent mappings from geometry and boundary conditions to multi-physics flow fields, enabling robust generalization across a wide range of flow scenarios. The multiscale method supports fast, reliable upscaling of pore-scale information in heterogeneous structures and accurately predicts complex flow patterns within 1 s, achieving over 640 times speedup and 97% memory savings compared to conventional pore-scale methods. This work bridges the gap between pore-scale insights and macro-scale applicability, establishing a powerful and efficient paradigm for porous media analysis.
Xie et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: