(161g) Combining Process Knowledge and Machine Learning for Efficient Process Flowsheet Synthesis

We present a hybrid-AI framework for fast, efficient, and reliable flowsheet synthesis and design that takes into account underlying process systems engineering knowledge and concepts. A hierarchical extended SFILES representation (eSFILES) of process flowsheets is presented that is able to represent a wide range of process flowsheets, involving multiple reaction-separation sections, solvent-based separations, and many more. This representation is characterized by text-based flowsheet representations, a novel flowsheet grammar containing formal SFILES syntax rules, a novel hypergraph representation, and a process ontology for design and operation parameters. The flowsheet grammar defines the symbolic structure of each symbol (process atom, streams, numeric symbols) and the connectivity pattern embedded in the text-based flowsheet representation. The defined syntax rules are always applicable for syntactically correct SFILES representations irrespective of the complexity of the process flowsheets. We show that the text-based SFILES representation alone does not capture all the important details of a process. For instance, additional information about the number of chemicals and their behavior, direction of flow-paths for participating chemicals, a symbolic AI-based intelligent system for determining the number of feasible alternatives for each task, and so on are needed. Development of the above is necessary to correctly and consistently represent process flowsheets using the new eSFILES representation and allow easy integration with existing computer-aided process synthesis, design, and simulations methods and tools.

We demonstrate the efficacy of the eSFILES representation on complex process flowsheets with multiple reactors, separators, and recycle streams. We highlight aspects of process synthesis by generating different alternative flowsheets, including intensified and hybrid schemes from a reference. We also present other aspects of integration of process synthesis and design with commercial simulation tools such as Pro II and Aspen using a motivating example highlighting the smooth transfer of data to the simulators for purposes of verification by simulation. The eSFILES-based framework for process synthesis and design incorporates a combination of artificial intelligence-based methods and well-known chemical engineering knowledge incorporated through an intelligent system facilitating fast, correct, and consistent decision-making related to process synthesis and design [1-8].

References:

1. d'Anterroches, Loïc. Process Flowsheet Generation & Design through a Group Contribution Approach. [CAPEC], Department of Chemical Engineering, Technical University of Denmark, 2005.

2. Bommareddy, Susilpa, Mario R. Eden, and Rafiqul Gani. "Computer aided flowsheet design using group contribution methods." Computer aided chemical engineering. Vol. 29. Elsevier, 2011. 321-325.

3. 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.

4. Vogel, Gabriel, Lukas Schulze Balhorn, and Artur M. Schweidtmann. "Learning from flowsheets: A generative transformer model for autocompletion of flowsheets." Computers & Chemical Engineering 171 (2023): 108162.

5. Mann, Vipul, and Venkat Venkatasubramanian. "AI-driven hypergraph network of organic chemistry: network statistics and applications in reaction classification." Reaction Chemistry & Engineering (2023).

6. Mann, Vipul, and Venkat Venkatasubramanian. "Predicting chemical reaction outcomes: A grammar ontology‐based transformer framework." AIChE Journal 67.3 (2021): e17190.

7. Mann, Vipul, and Venkat Venkatasubramanian. "Retrosynthesis prediction using grammar-based neural machine translation: An information-theoretic approach." Computers & Chemical Engineering 155 (2021): 107533.

8. Mann, Vipul, Rafiqul Gani, and Venkat Venkatasubramanian. "Group contribution-based property modeling for chemical product design: A perspective in the AI era." Fluid Phase Equilibria (2023): 113734.