Professor Artur SCHWEIFTMANN (TU Delft)At the CAPE-OPEN 2022 Annual Meeting, Professor Artur SCHWEIDTMANN of TU Delft presented on his research activities focusing on extracting information from flowsheets in order to develop the capability of artificial intelligence to build flowsheets.

Artificial intelligence led to tremendous success in automated decision-making in various domains such as computer vision or games even surpassing human performance [1]. However, in the field of chemical process design, the lack of structured big data has hindered breakthroughs of artificial intelligence [2]. A promising approach to overcome the challenge of missing data structures, is flowsheet mining and information representation through knowledge graphs [3,4]. I envision to establish the Chemical Engineering Knowledge Graph (ChemEngKG) that provides big open and linked chemical process information.

In order to structure existing chemical process information, I present the concept of “flowsheet mining” [3]. Flowsheet mining extracts process information from flowsheet images and process descriptions found in scientific literature and patents. The proposed technology combines a range of tools including data mining, computer vision, natural language processing, and semantic web technologies. In my contribution, I will elucidate the theoretical background of flowsheet mining, discuss previous literature, and show the immense future potential. I believe the availability of structured process data will enable breakthroughs in process design through artificial intelligence.

To illustrate the potential of my vision, I present a framework to automatically identify and extract thousands of flowsheet images from several million chemical engineering publications. These flowsheet images are automatically digitized using developed deep learning computer vision algorithms. This allows us to represent flowsheet topologies in a machine-readable graph format. In addition, information form chemical process simulation files such as Aspen or DWSIM is automatically added to the database. The extracted data is enriched with an chemical process ontology, inspired by OntoCAPE [5] building a large knowledge graph including several million nodes. Finally, we show the potential of artificial intelligence algorithms on the mined data to perform a number of process engineering tasks.


[1] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature521(7553), 436-444.

[2] Schweidtmann, A. M., Esche, E., Fischer, A., Kloft, M., Repke, J. U., Sager, S., & Mitsos, A. (2021). Machine learning in chemical engineering: A perspective. Chemie Ingenieur Technik93(12), 2029-2039.

[3] Schweidtmann, A. M. (2022). Flowsheet mining. In Preparation.

[4] Weber, J. M., Guo, Z., Zhang, C., Schweidtmann, A. M., & Lapkin, A. A. (2021). Chemical data intelligence for sustainable chemistry. Chemical Society Reviews.

[5] Morbach, J., Yang, A., & Marquardt, W. (2007). OntoCAPE—A large-scale ontology for chemical process engineering. Engineering applications of artificial intelligence20(2), 147-161.