(390ap) Superstructure-Based Synthesis of Extractive Distillation Sequences Considering Preconcentration

Approximately 10% of global energy consumption is attributed to chemical production. Among various unit operations in the chemical industry, distillation is the most widely used, accounting for over 40% of the total energy consumption in chemical processes. Furthermore, azeotropes are commonly encountered in the production of chemicals, petrochemicals, biofuels, and pharmaceuticals, as well as in environmental engineering and waste treatment. Separating azeotropic mixtures by extractive distillation typically requires the use of entrainers, which further increases energy consumption. Preconcentration can significantly reduce the energy demand of extractive distillation processes for azeotrope separation. However, incorporating preconcentration into process synthesis introduces additional complexity. Under increasingly stringent carbon dioxide emission regulations, there is an urgent need to develop advanced process design method for the synthesis and optimization of extractive distillation sequences with preconcentration for multicomponent azeotropic mixtures.

In this presentation, we propose a systematic method to address the synthesis and optimization of extractive distillation sequences considering preconcentration for multicomponent azeotropic mixtures, which can be used for mixtures containing any number of binary azeotropes. First, the sequence search algorithm (SSA) identifies feasible sequences based on the separation matrix (SM) and the composition information matrix (CIM). The SM stores possible top/bottom products during the separation of the mixture, while the CIM stores the composition of these products. To enable automated sequence generation, a novel numerical method is introduced to automatically update the SM and the CIM, enabling the exploration of all extractive distillation sequences considering preconcentration to obtain a more comprehensive solution space. After that, all feasible distillation configurations are modeled using a superstructure approach. One issue that arises when using superstructure is that the amount of entrainer is treated as an optimization variable, and its amount shows significant change, especially when preconcentration is considered. Therefore, a revised superstructure and rigorous tray-by-tray model are used to construct accurate model for potential distillation sequences. The superstructure model is developed within process simulation software. To enhance computational efficiency, repetitive separation tasks without extrainer are optimized separately, enabling some distillation sequences to be implicitly embedded in the model, while the amount of entrainer added is considered as an optimization variable. The revised superstructure allows for the simultaneous simulation of all distillation sequences and improves the computational efficiency of the optimization algorithm.

We propose a hybrid optimization algorithm (HOA) to optimize the revised superstructure for simultaneous optimization of the distillation sequence and columns. The HOA consists of two parts: improved genetic algorithm (IGA) and the topology optimization algorithm (TOA). In this work, the genetic algorithm (GA) is modified to enable it to achieve the combined selection of good-performed sequences over multiple individuals, and integrates it with the TOA to optimize the revised superstructure, obtaining accurate results with a more efficient parallel computation algorithm. Two case studies are used to illustrate the applicability of the proposed method. The first case involves an azeotropic mixture with nonlinear distillation boundary and a better solution is obtained compared to existing literature. The second case includes multiple binary azeotropes and has shown energetic and economic benefits for applying preconcentration techniques, presenting the effectiveness of the proposed method in handling complex problems and developing more efficient, energy-saving distillation processes. Moreover, the HOA reduces computational time by over 70% compared to the method of enumerating all sequences and optimizing them individually using the GA.