Data-Driven Modeling of General Fluid Density Under Subcritical and Supercritical Conditions

Abstract

Advanced propulsion and power-generation systems often operate under extreme conditions, where thermophysical properties of the working fluids undergo complex variations in a wide range of fluid states, where empirical cubic equations of state could yield substantial errors in density prediction. The present work develops data-driven models for accurate density estimation of general fluids across all thermodynamic regimes. The model starts with the cubic equation of state, whose alpha function is calibrated in a data-driven manner with statistical correction accounting for inherent correlations among training data samples. The developed models are examined for the representative pure substances in aerospace propulsion systems, including oxygen, nitrogen, carbon dioxide, and hydrocarbon fuels. Results show that the model with pressure and temperature as input variables provides consistently superior accuracy over wide ranges of temperatures and pressures, especially in the compressed-liquid region, where the Peng–Robinson equation of state significantly underperforms. The corresponding absolute average relative deviation for the studied substances is below 0.65% at different pressures, compared to 7.16% by the Peng–Robinson equation of state. The model is also extended to examine the density calculations of the selected binary and ternary mixtures, and the consistent result is obtained. The data-driven approach can be adopted to evaluate other thermodynamic properties of fluids and fluid mixtures and characteristics of vapor–liquid equilibrium, and further incorporated into large-scale multiphysics simulations where nonideal gas behavior occurs in the future.