Enhancing Micro-Nanoscale Pore Structure Reconstruction in Microfluidic Experiments Using Flow-Constrained Super-Resolution Algorithms

Abstract Microfluidic experiments have become indispensable in the study of fluid dynamics within porous media, especially for simulating water injection processes and multiphase flow in reservoirs. These experiments provide valuable insights into fluid distribution, interfacial dynamics, and enhanced oil recovery mechanisms. However, a major challenge lies in the limitations of traditional optical imaging techniques, which fail to capture the fine details of micro-nanoscale pore structures crucial for understanding flow behavior. This work proposes a novel approach that integrates fluid flow constraints into a super-resolution (SR) deep learning framework to achieve highly accurate reconstruction of micro-nanoscale pore structures in microfluidic chips. The proposed method leverages physical principles of fluid flow, such as capillary pressure equilibrium, flow continuity, and interfacial dynamics, as prior constraints within the deep learning architecture. By embedding these flow constraints, the model gains a deeper understanding of the underlying physics governing multiphase flow, enabling it to reconstruct high-resolution images from low-resolution data while preserving the physical realism of fluid dynamics. The algorithm is trained on a dataset consisting of paired low-resolution and high-resolution images of porous structures, augmented with synthetic flow patterns consistent with physical laws. This unique approach ensures that the model not only enhances spatial resolution but also accurately captures the fluid behavior within reconstructed pore geometries. Experimental results demonstrate that the flow-constrained SR algorithm significantly outperforms traditional SR methods and physics-unconstrained deep learning models in terms of reconstruction accuracy and fluid simulation fidelity. Key improvements include enhanced visualization of pore throats, better representation of wettability effects, and precise tracking of oil-water interfaces during dynamic flow experiments. These advancements enable more accurate analysis of microfluidic processes, such as evaluating the efficiency of waterflooding and chemical-enhanced oil recovery strategies. Beyond microfluidic experiments, this work highlights the broader potential of integrating physical laws with artificial intelligence to address challenges in various fields. Applications extend to optimizing microfluidic device designs, validating high-resolution reservoir models, and studying multiphase flows in energy, materials science, and biomedical research. By bridging the gap between physics-based modeling and data-driven approaches, this research sets a new benchmark for high-resolution imaging and opens up novel pathways for interdisciplinary advancements in fluid mechanics and AI-driven computational science.