Immiscible fluid displacement in porous media is governed by pore-scale behaviors, which can be manipulated by chemical additives to engineer the process toward a greater turn-out. Although recent advances in nanofluids have been reported to influence such a process, their interfacial phenomena are likely controversial and need independent cross-examinations. As non-energetically interfacial responsive nanoparticles, silica cores adorned with polyvinylpyrrolidone were examined for their direct contribution to crude oil displacement performance at relatively low concentrations, ranging from 10 to 500 ppm, in the current study. The crude oil displacement was experimented via water-wet borosilicate micromodel and visualized to elucidate pore-scale interfacial phenomena involved. Concentration-dependent property of nanofluids was found, evidenced by different pore-scale mechanisms observed. At low concentrations (10 and 50 ppm), wetting layer flow controlled the oil displacement and led to swelling into pore space, inducing snap-off events and hence high oil ganglia trapped (> 300). At higher concentrations (100 ppm), nanoparticle self-arrangement at the water wedge was more effective, which induced oscillatory structural disjoining pressures between the oil-aqueous and solid-aqueous interfaces leading to narrow nanofluid spreading. Hence, the spatiotemporal displacement performed differently at high concentrations (displacement efficiencies were 36.8% at 100 ppm and 43.1% at 500 ppm), with snap-off hardly observed. At 500 ppm, more stable and stronger nanofilm spreading was developed due to meniscus expansion, obtaining faster-displacing dynamics (54.9% per pore volume injected) and additional oil displaced (+5.7%) after breakthrough time. The findings amplify nanofluid contribution and emphasize its concentration dependence on immiscible fluid flow in porous media, a potential applicability to various fields including enhanced oil recovery and CO₂ geological storage.

To achieve net-zero carbon emission, securely and permanently sequestrating CO₂ into deep underground is internationally assured as a robust solution, although a few technical challenges on complex in-situ storage process are yet to be overcome. Despite researchers are increasingly familiar with laboratory-scale CO₂-brine displacement and how to characterize and improve the process, field implementation is not that simple and of great challenge. In this article, an opportunity on an approach that utilizes fluid-fluid displacement fundamentals is discussed to predict CO₂ sequestration using artificial intelligence. A concept of machine learning is introduced, where computer programs can learn and improve automatically via previous experiences. With machine learning model, fluid displacement behaviors that are spontaneously monitored are emphasized to predict the displacement result, which is readily adjusted if needed while training the model from real-time CO₂ injection response. Such an approach is a real-time autonomous adjusting process, consisting of three main stages: Selection of first appraisal fluid for trial injection, real-time machine learning from in-situ injection response, and fluid adjustment if needed or continuation on the same injection until achieving a maximum CO₂ storage. This approach could play a vital role in the carbon capture and storage industry to develop CO₂ storage effectively with adequate resources, and yet has a potential to substitute a conventional design or fluid screening approach for subsurface fluid injection, including underground hydrogen storage and hydrocarbon recovery.