Process-driven solvent screening for efficient extractive distillation using interpolative rational functions

• Rational functions were trained as VLE surrogate models with thermodynamic consistency. • Superstructure-based extractive distillation process synthesis was solved within a few seconds. • A multi-level solvent screening was performed for ethylbenzene/styrene separation. • C 2 H 2 Br 4 was found to reduce cost by 27.9 % compared to the benchmark sulfolane. When designing extractive distillation processes, using selectivity and capacity at infinite dilution alone is hard to identify the real optimal solvent with minimal process cost. To overcome this problem, a new process-driven solvent screening approach is developed. As simple and reliable surrogate models, rational functions (algebraic fractions such that the numerator and the denominator are polynomials) and multivariate polynomials (a subset of rational functions) are trained to interpolate vapor–liquid equilibria with thermodynamic consistency. The surrogate models can directly be embedded into superstructure-based extractive distillation process design to obtain optimal solutions within a few seconds. This enables to evaluate the real process performance of numerous solvents efficiently. Incorporating the accelerated process design strategy, a multi-level solvent screening framework is proposed and exemplified for the separation of a close-boiling mixture ethylbenzene/styrene. The solvent C 2 H 2 Br 4 ultimately enables a cost reduction of 27.9 % compared to the industrially used benchmark solvent sulfolane.