Optimizing Oil Detachment from Silica Surfaces Using Gemini Surfactants and Functionalized Silica Nanoparticles: A Combined Molecular Dynamics and Machine Learning Approach.

The decline in the exploration of new oil sites necessitates the development of efficient strategies to maximize recovery from existing reservoirs. This study employs a Molecular Dynamics (MD) approach to investigate oil detachment from silica surfaces of varying hydrophobicity using a combination of bis-cationic gemini surfactants (GS) and functionalized silica nanoparticles (SNPs). Density profiles and radial distribution function (rdf) plots revealed a multilayered oil adsorption model. A reduction in oil-silica interaction energy was observed with increasing surface hydrophobicity, highlighting the importance of polar interactions. Standard waterflooding studies, involving oil detachment solely with water, were conducted to assess baseline recovery efficiency. All the GS-SNP combinations outperformed standard waterflooding methods. SNPs significantly mitigated GS adsorption on reservoir beds, as evidenced by center-of-mass measurements. However, the effectiveness of the added injectants (GS-SNP) went downhill with increasing surface hydrophobicity, further validating the existence of a potential barrier for oil detachment, as known previously. Finally, supervised Machine Learning (ML) models were generated to predict the GS-SNP combination for a given silica surface, with MD generated descriptors. In most cases, boosting models, viz., XGBoost and AdaBoost yielded best correlation with the observed data. However, for the complex oil model, Ridge regression and Support Vector Regression (SVR) outperformed other ML models in SNP prediction, pointing to the existence of a simpler correlation between the descriptors and the output variable. With these findings, the study attempts to streamline the data-driven design of chemical injectants for Enhanced Oil Recovery purposes.