Generative graph machine learning for integrated molecule and process design

• Applicants: Professor Dr. Martin Grohe; Professor Alexander Mitsos, Ph.D.
• Subject Area: Chemical and Thermal Process Engineering ; Theoretical Computer Science
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 466417970

We will develop a framework for the integrated design of molecules and processes with generative graph machine learning (ML). The framework couples the molecule scale with the process scale of Chemical Engineering in four modules: On the molecule scale, (1) graph neural networks (GNNs) provide thermodynamic-consistent predictions for molecular and mixture properties. Further, (2) generative graph ML enables automated computer-aided design of molecules with desired properties. The GNNs and generative graph ML for molecules are integrated with the process scale for (3) model equation-based process design and (4) reinforcement learning-based process design. The methodological advances in each module and their combination will enable the integrated design of chemical processes and corresponding molecules, i.e., working fluids, solvents, and products. The framework will therefore advance and accelerate the discovery of novel molecules and chemical processes with higher efficiency and sustainability through graph ML. The framework and integrated molecular and process design case studies will be made available as open source as part of SPP 2331, allowing for active use and extension.

• DFG Programme: Priority Programmes
• Subproject of: SPP 2331: Machine Learning in Chemical Engineering. Knowledge Meets Data: Interpretability, Extrapolation, Reliability, Trust