(60u) Optimization-Based Multi-Technology Separation Network Synthesis

In most chemical processes, a sequence of separation tasks is performed to recover unconverted raw materials and purify potentially many products. These tasks are typically energy-intensive and/or capital-intensive; hence, decisions on the separation network made in the preliminary design phase will determine, for the most part, the final overall cost. Most studies in separation network synthesis revolve around distillation, and, consequently, the benefits of having multiple technologies in the separation are often overlooked. Integrating different technologies in the separation network synthesis problem expands the search space allowing a potentially better network to be identified.

The separation network synthesis problem has been addressed using two main approaches: sequential and simultaneous. In sequential approaches, the problem is decomposed into subproblems and solved by fixing some decisions and using heuristic rules to determine the remaining [1]. Evolutionary algorithms can also be employed to sequentially improve an initial structure. Due to the use of heuristics, these approaches often lead to suboptimal systems. Alternatively, in simultaneous approaches, useful units and their relevant interconnections are embedded into a superstructure, and the final separation network is obtained by solving the resulting optimization problem [2]. Most studies in this area consider only distillation, whereas the available study which considers different technologies explicitly requires complete information on the feed stream [3]. Ryu et al. (2020) discussed a generalized problem statement that allows the seamless integration of the separation network and reactor network synthesis.

Accordingly, we propose a superstructure-based approach for separation network synthesis with multiple technologies where the feed streams may be variable (unknown), so that the method can be readily coupled with reactor network synthesis. The approach employs a superstructure that encompasses numerous promising configurations comprising multiple separation technologies. Since every technology exploits the difference, across components, in one or more key properties, we use a component ranking system based on the key properties to generate possible separation splits, which can be different across technologies. Consequently, we construct a separation matrix [4] for each technology based on the component ranking and identify all possible separations and connections using these matrices. We use the concepts of nodes and arcs to represent the superstructure where the nodes represent the network inlets (source nodes), network outlets (sink nodes), and separation blocks (mixture nodes), whereas the arcs represent streams between nodes. The mixture nodes are essentially the elements of the constructed separation matrices. We employ arcs to represent streams from network inlets to the mixture nodes, the mixture nodes to the network outlets, and the products of separation blocks to other mixture nodes.

In a mixture node, separation is performed in a separation block resulting in a top and a bottom product streams whose compositions depend on the selected technology. A separation block here represents a set of operations that enables a sharp split separation. In solvent extraction, for instance, a separation block consists of an extraction column and the solvent recovery system. Similarly, in reverse osmosis, we utilize multi-stage membrane units to achieve a sharp split separation. To maintain the tractability of the final optimization model, the separation block models are generated via offline (block-specific detailed) calculations. The final model is a mixed integer nonlinear programming model where discrete decisions such as the activation of nodes and arcs, the selection of technologies, and the separation splits are represented by binary variables. We demonstrate the applicability of our approach via two examples.

References

[1] A. Gomez M and J. D. Seader, “Separation sequence synthesis by a predictor based ordered search,” AIChE journal, vol. 22, no. 6, pp. 970–979, 1976.

[2] R. W. H. Sargent and K. Gaminibandara, “Introduction: approaches to chemical process synthesis,” Optimization in action, 1976.

[3] I. Heckl, Z. Kovács, F. Friedler, L. T. Fan, and J. Liu, “Algorithmic synthesis of an optimal separation network comprising separators of different classes,” Chemical Engineering and Processing: Process Intensification, vol. 46, no. 7, pp. 656–665, 2007.

[4] V. H. Shah and R. Agrawal, “A matrix method for multicomponent distillation sequences,” AIChE journal, vol. 56, no. 7, pp. 1759–1775, 2010.