?
Molecular dynamics and machine learning insights into oil-brine interfacial tension: Uncovering roles of dissolved gases and salinity
Hypothesis:Accurate determination of interfacial tension values is crucial for many modeling applications in the oil and gas industry. However, many of the presently used empirical and theoretical models show poor prediction accuracy and do not allow for direct consideration of the concentrations of gases and salts dissolved in the system as well as the specific composition of the oil. Molecular dynamics methods allow one to directly model the oil-brine interface and may provide a pathway to build more precise and complete interfacial tension models.
Simulations:We adapt a multi-component molecular model consisting of 17 representative hydrocarbons to qualitatively match the experimental oil composition of a specific reservoir. We perform molecular dynamics simulations to assess the effect of temperature, pressure, salt concentration in water and gas contents in oil within experimentally observed ranges on the interfacial tension. Using the robust statistics collected over 10 μs of molecular dynamics simulations, a machine-learning model is trained to reproduce the interfacial tension with mean absolute percentage error of 0.9%.
Findings:We obtain linear dependencies of interfacial tension on the considered parameters and show that concentrations of salts and gases dissolved in the system exert the largest influence on the interfacial tension value. The asphaltenes contained within the oil phase act as surfactants, decreasing the interfacial tension, with the effect further enhanced by the concentration of the dissolved gases on the interface. The molecular model was verified against experimental data. The resulting ML model shows excellent agreement with simulation data. Our results provide a pathway towards building a scalable and physically grounded framework for predicting oil-brine interfacial tension across wide thermodynamic and compositional ranges to enable reliable reservoir-scale modeling.