Simulation of thermal phase changes in compressible flows using machine learning in OpenFOAM

Modeling of the phase change and heat transfer behavior of a high pressure steam cleaning process in a superheated steam dishwasher has been recently investigated in our research work. CFD simulations are performed using the two-phase VOF solver interThermalPhaseChangeFoam. This solver assumes constant material properties of the two phases and only considers the compressibility due to the phase change. To improve the temperature and flow field predictions, the solver is extended to handle compressible flows. The polynomial approximation of the temperature and pressure dependent material properties as well as the compressibility effects are integrated into the new solver. The polynomial coefficients are obtained by fitting experimental data from the CoolProp open-source thermophysical library. In addition, the energy equation is solved in terms of enthalpy, and the temperature field is subsequently updated with a partitioned neural network model trained with thermophysical data from the same library and integrated with pybind11 in OpenFOAM. With this data-based approach, an explicit equation of state model is no longer required to close the system of transport equations. The integration of thermophysical libraries and machine learning methods is a promising approach for advanced CFD calculations.