(311c) Superstructure Optimization Using the Extended Sfiles Representation of Flowsheets Combining Domain Knowledge, Machine Learning, and Symbolic AI

We present a novel superstructure representation framework for process flowsheets using an extended SFILES (eSFILES) representation that combines domain knowledge and artificial intelligence. The eSFILES framework proposed by Mann et al. [1] hierarchically represents process flowsheets, starting from a concise text-based flowsheet representation (Tula et al. [2]) at the base level (level 0) and progressively incorporating increasing degrees of process context and knowledge. At level 1, the SFILES grammar syntax rules are used to generate a flowsheet hypergraph explicitly capturing connectivity information. At level 2, specifications for simple mass and energy balance calculations are included as hypergraph annotations. Finally, at level 3, a process ontology is integrated with the hypergraph to incorporate design and operational parameters for rigorous simulations.

Building upon this multi-level eSFILES framework, we have developed a superstructure representation that enables the efficient enumeration and optimization of process alternatives. The superstructure represents using a hypergraph-based representation the various process alternatives generated based on provided flowsheet context and a knowledge based of process atoms. It captures the search space of feasible process alternatives while incorporating structural constraints to ensure syntactic correctness. The superstructure representation leverages the mathematical foundation of hypergraphs, allowing seamless integration with existing optimization formulations and solution strategies.

We demonstrate the application of the eSFILES-based superstructure representation using the hydrodealkylation (HDA) process as a case study. In principle, our eSFILES superstructure representation can handle all flowsheets with reaction followed by separation steps. The optimization formulation, including the objective function and constraints, is automatically derived from the superstructure representation and the embedded process knowledge. The resulting GAMS code is generated to solve the superstructure optimization problem, considering simple models for mass and energy balances. As a future extension, we plan to include level 4 to incorporate process safety, control, life cycle assessment considerations, and so on, along with applications in process intensification.

The eSFILES-based superstructure representation provides a powerful framework for process synthesis and optimization. It combines the advantages of concise and standardized flowsheet representation, explicit capture of process knowledge, and integration with state-of-the-art optimization techniques. This enables the systematic exploration of the design space, identification of innovative process alternatives, and data-driven decision-making.

REFERENCES

1. Mann, Vipul, et al. "eSFILES: Intelligent process flowsheet synthesis using process knowledge, symbolic AI, and machine learning." Computers & Chemical Engineering181 (2024): 108505.
2. Tula, Anjan Kumar, Mario R. Eden, and Rafiqul Gani. "Process synthesis, design and analysis using a process-group contribution method." Computers & Chemical Engineering 81 (2015): 245-259.