Surrogate modeling of phase equilibrium calculations using adaptive sampling

The choice of thermodynamic models for phase equilibrium calculations plays a central role in the context of process simulation. For highly non-ideal systems, equations of state as the Perturbed-Chain Statistical Associating Fluid Theory (PC-SAFT) are preferred for an accurate description of phase equilibria in a chemical process simulation. In order to implement such complex models, the thermodynamic model either is directly implemented into the process simulation software or it is outsourced into an external library, which is called by the process simulator. For PC-SAFT models, due to the iterative solution process, both options can lead to a high computational effort for the process simulation and optimization. This thesis suggests to use surrogate modeling - replacing a complex model by a more simple black box model - in order to reduce the computational effort for complex phase equilibrium calculations. Two methods are proposed: the indirect method applies surrogate models within explicit formulation of the phase equilibrium to efficiently solve the phase calculations during process simulation, while the direct method applies surrogate models to directly determine phase compositions. Prior to training the surrogate models, samples using original model calculations are drawn. In order to reduce the computational effort for sampling, an adaptive sampling method is proposed, which combines sampling and training of the models in an iterative manner. This method provides superior surrogate models for the same number of samples compared to a conventional space-filling sampling design, which is shown for different surrogate model types and systems. The trained surrogate models are applied to the simulation and optimization of the 1-dodecene hydroformylation process.