Geological carbon storage is emerging as an alternative technology capable of mitigating the impact caused by the emission of greenhouse gasses (GHG), among them carbon dioxide CO2. The trapping mechanisms that keep the gas in the subsurface rock are categorized into different processes. The adsorption process is critical and is also directly present in two storage mechanisms, residual storage and carbon mineralisation. The molecule can be adsorbed onto the mineral surface of the pores, and this process can also become a storage mechanism when the pore space is fully occupied by CO2. Another important point is that storage capacity is a critical factor due to the huge quantities of CO2 that need to be stored. Due to the nature of the interactions, simulations at the molecular scale are essential to understand the mechanism and raise the processes involved at the fluid/mineral interfaces present in the geological site reservoir. However, the heterogeneous nature of geological formations poses a significant challenge in accurately estimating storage capacity through adsorption processes. The vast space of chemical compounds across mineral surfaces is an excellent opportunity for machine learning technologies to address the problem and lead to predictions of CO2 adsorption energies under different thermodynamic and environmental conditions. Our approach leverages a diverse dataset comprising structural and chemical properties of minerals, alongside corresponding adsorption energies obtained from first-principles calculations. To understand the relevant descriptors that capture the physics underlying the adsorption of gas molecules on mineral surfaces, research has been initiated into a case study using Density Functional Theory (DFT) to systematically evaluate the adsorption behaviour of the CO2 molecule on the surfaces of selected minerals from the feldspar class. To deal with the complexity posed by geological heterogeneity, we plan to employ graph neural networks as our machine-learning model. This approach holds immense importance as it allows us to capture intricate spatial relationships within heterogeneous formations and helps to shed some light on significant interactions at complex CO2/mineral interfaces under reservoir conditions.
Jessica Santos Rego