Autocompletion of engineering diagrams

Inspired by text autocompletion, we develop generative AI models that suggest how to complete a flowsheet while an engineer is drawing it.

Challenge

Process synthesis is still largely a manual activity. Engineers draw flowsheets in simulation or drawing software, drawing on experience and on patterns that recur across many processes. Software tools do not yet use this collective design knowledge to make suggestions during interactive flowsheet synthesis.

Our approach

We represent flowsheets as strings using the text-based SFILES 2.0 notation and train a transformer-based language model to learn the grammatical structure of the SFILES 2.0 language and common patterns in flowsheets. The model is pre-trained on synthetically generated flowsheets to learn the flowsheet grammar and then fine-tuned on real flowsheet topologies by transfer learning. The trained model is used for causal language modeling to autocomplete flowsheets. The results show the potential of this approach for AI-assisted process synthesis, where the model provides chemical engineers with recommendations while they design a process. Data augmentation of flowsheets further improves model performance when data are scarce.

From research to practice

Key publications

  • Vogel, G., Schulze Balhorn, L., & Schweidtmann, A. M. (2023). Learning from flowsheets: A generative transformer model for autocompletion of flowsheets. Computers & Chemical Engineering, 171, 108162. doi:10.1016/j.compchemeng.2023.108162 · Details
  • Vogel, G., Hirtreiter, E., Schulze Balhorn, L., & Schweidtmann, A. M. (2023). SFILES 2.0: An extended text-based flowsheet representation. Optimization and Engineering, 24, 2911–2933. doi:10.1007/s11081-023-09798-9 · Details
  • Schulze Balhorn, L., Hirtreiter, E., Luderer, L., & Schweidtmann, A. M. (2023). Data augmentation for machine learning of chemical process flowsheets. Computer Aided Chemical Engineering, 52, 2011–2016. doi:10.1016/B978-0-443-15274-0.50320-6 · Details
← All projects