Grant: Discovery Grants Program - Individual

The transition to net zero requires carbon storage (CS) with 95% of global carbon expected to be stored geologically as it is seen as permanent, less risky, and technically more advanced and economical. Subsurface energy storage can be a safe option for temporary hydrogen (H2) and long-term storage of CO2. Our challenge is to predict and reduce uncertainty of its fate. CO2 is trapped via structural, residual, solubility, and mineral trapping. H2 storage is often considered in salt domes, but geological formations (ie depleted oil/gas fields, saline aquifers) could be targets for both H2/CO2 storage. There are many uncertainties around injectivity and storage capacity required including parameters related to geologic heterogeneity, rock characterization, in-situ fluid properties (density, viscosity of live fluids, solubility, interfacial tension (IFT)), and rock-fluid interactions/reactive transport (reservoir wettability/contact angle, capillary pressure, relative permeability, adsorption and reactive transport); the same uncertainties that oil and gas has been facing. Uncertainty comes from property measurement and prediction with the largest uncertainty being the ability to upscale properties from pore to core to field due to heterogeneities at all scales. We propose a novel hybrid data-driven molecular to macro upscaling framework to predict properties and reduce storage and injectivity uncertainty. This approach is based on our multidisciplinary, multiscale research, from screening EOR mechanisms phenomenologically in the lab to investigating molecular/thermodynamic models to using machine learning to predict and optimize at all scales. This insight suggests that "upscaling" may be approached differently. Pore scale modeling was a hopeful approach but is fraught with expensive in-situ monitoring and computational time to solve the continuum model even at the pore scale. New methods suggest machine learning (ML) to upscale pore scale models however they still do not capture the molecular level interactions that underpin the ultimate behavior captured through phenomenological models (Darcy's law, capillary pressure, relative permeability). The proposed hybrid data-driven molecular to macro upscaling framework will skip pore scale continuum modelling and will use molecular dynamic (MD) simulations and ML to predict and upscale fluid/rock properties from basic data using MD and thermodynamic models. Once validated with experimentally measured properties (at high temp, high pressure (HPHT)), statistical ML methods will be used to predict phenomenological parameters (relative permeability, injectivity) that will again be validated with lab results. The novel framework will be tested for both CO2/H2 in different compositions of saline brine and rock mineralogy. The measured HPHT properties are important outputs themselves due to lack of publicly available data. The proposed framework could impact how we upscale transport in porous media.