Microfluidic investigation on CO2 self-storage behavior in submarine sediment using artificial intelligence image recognition

Current research on CO2 storage technologies predominantly focused on core-scale experiments, which overlooked the effect of intricate pore structure and multiphase fluid behaviors at the pore scale. This study investigated CO2 self-storage behavior using an integrated microfluidics platform combined with CCD imaging and in situ Raman spectroscopy. An artificial intelligence image recognition model was developed and trained on over 1000 annotated images to enhance image analysis. Experiments were conducted in both homogeneous and heterogeneous microfluidic chips to observe key processes including CO2-pore water displacement, hydrate nucleation and decomposition, as well as the sealing performance of the hydrate cap. CO2 hydrate phase transition analysis revealed distinct formation pathways: porous hydrates nucleated at liquid CO2-water interfaces with inward growth, while opaque hydrates developed via aqueous-phase crystallization from CO2-saturated solutions. Thermal stability assessments demonstrated superior performance of porous hydrates across scales (ΔT = -2.69∼-1K) compared to opaque hydrates (ΔT = -1.69∼0K). Breakthrough test quantified critical thresholds, showing the homogeneous chip with a pore size of 61.5 μm achieved 12,000 kPa seal integrity at 13.52 % hydrate saturation. Hydrate cap stability increased with hydrate saturation (+28.43 %), whereas the heterogeneous chip with an average pore size of 125.82 μm exhibited premature failure. This work confirms the feasibility of CO2 marine geological storage and proposes a multimodal methodology that establishes a framework for in situ hydrate characterization, offering quantitative predictors for seal capacity optimization.