(235b) Data-Driven Integrated Design of Solvents and Extractive Distillation Processes

In extractive distillation (ED), a suitable solvent is added to increase the relative volatility of closely boiling or azeotropic mixtures, thereby facilitating their separation. The feasibility and efficiency of this separation depend not only on the selection of the solvent, but also on the process operating conditions. Computer-aided molecular and process design (CAMPD) approaches have been proposed to optimally design solvents and processes for a given application. As advanced molecular property models and detailed process models with many variables needed to be combined in CAMPD, such problems are usually computationally expensive optimization tasks.

In this work, we propose an efficient integrated solvent and process design approach for ED processes based on a data-driven modeling strategy. The latter uses process models built by using artificial neural networks (ANNs) from which the key performance indicators of an ED process are calculated based on the solvent’s physical properties and the process parameters (Fig. 1a). The ANNs serve as surrogates for the first principle models normally used in CAMPD. On this basis, multi-objective optimization is performed with a genetic algorithm to maximize the product purity while minimizing the energy consumption (Fig. 1b), through which optimal solvent properties and corresponding optimal process parameters are obtained. Subsequently, real solvents that approximate the optimal property values are identified from a large solvent database. The performance of the optimal solvent and corresponding process parameters is evaluated by rigorous simulations of the ED process. Additional chemical hazard assessment and economic evaluation are performed to support the final decision making.

The proposed data-driven integrated solvent and process design methodology is demonstrated for the separation of 1-butene from 1,3-butadiene. For this separation task, three real solvents are identified, showing around 11% lower energy consumption than the benchmark solvent n-methyl-2-pyrrolidone (NMP), while maintaining high product purity (Fig. 1c). Supported by chemical hazard assessment and economic evaluation, acetylacetone is finally determined as a potentially suitable solvent. Overall, the solvents obtained from the integrated solvent and process design are technologically feasible and economically attractive. The integrated CAMPD is solved in less than 1 minute, which indicates that the data-driven approach can substantially reduce the optimization complexity and enable efficient CAMPD.

Fig. 1. (a) Schematic diagram of the data-driven integrated solvent and process design method for extractive distillation, (b) multi-objective optimization of the integrated design problem in which two objective functions are defined based on C4H8 purity and reboiler heat duty, and (c) performance evaluation of the ED process for the identified optimal solvents and corresponding process parameters.