Upcoming Seminars

Machine-Learned Physical Properties for Molecular Design

Wednesday, May 8, 2024 | 9:00 AM
Location: 467 ISE Lab | Zoom

This talk gives an overview of our work on the predictions of physical properties and the use of these predictions for molecular design. We have been employing Graph Neural Networks and Generative Machine Learning. The calculation, estimation and prediction of physical properties is crucial for the design of chemical products and processes. As such chemical engineering has devoted substantial effort over decades, spanning experimental work, macroscopic semi-empirical thermodynamic models to a-priori predictive models. In the last years many research groups have employed modern techniques from machine learning. A particular challenge is the relatively small number of data. We present an overview of our employed and developed methods, the various physical properties predicted, and the effect of physics-informed machine learning. We also demonstrate how the predictions can be used for molecular design, with a focus on fuel synthesis from biomass and hydrogen.

View Flyer