Thermodynamic modeling of water-oil-gas separation systems with focus on subsea separation: a critical review and assessment of equations-of-state

In offshore production, crude oil is initially separated on the platform topside, and a subsea system intended to replace this step must be highly reliable and require minimal maintenance. Understanding phase equilibrium under subsea conditions, which involve high pressures and nearly adiabatic processes, is essential for efficient separation. Thermodynamic modeling of such systems generally follows two strategies: using volumetric equations of state (EOS) for all phases or applying hybrid approaches that combine an EOS with an activity coefficient model for the liquid phase. Modeling brine–oil–gas mixtures is challenging because it requires representing ion–ion, ion–molecule, and molecule–molecule interactions. For ionic solutions, the Pitzer model is one of the most established, combining long-range Debye–Hückel theory with a short-range virial term, while eNRTL is also widely used and provides a robust framework for high-salinity produced water. Approaches based on excess Gibbs energy can describe short-range interactions and be coupled with long-range terms by separating the residual Helmholtz energy into hard-sphere, attractive, and Debye–Hückel contributions. Tests were performed with PR-Pitzer and ePC-SAFT, showing accurate predictions of CO₂ solubility up to 100 bar and, for electrolyte mixtures, up to 70 bar, with ePC-SAFT performing better at higher pressures. For the non-parameterized Peng–Robinson, PC-SAFT, and SAFT-VR Mie models, density predictions were obtained with satisfactory accuracy. The systems analyzed were selected to show how these EOS can produce comparable results and to illustrate the influence of temperature and pressure, including light hydrocarbons, water, and brine; for mixtures containing ions, electrolyte versions such as ePC-SAFT and SAFT-VRe Mie were used. In conclusion, thermodynamic models are strongly affected by pressure and temperature, and no single model can represent all conditions reliably.