Accurate predictions of thermodynamic properties are of central importance for designing and optimising chemical processes, since experimental data is often very limited. Established physical models are increasingly reaching their limits in terms of both predictive accuracy and scope of application. Modern machine learning (ML) methods offer promising new possibilities in this area. However, ML-based approaches generally do not yield consistent and physically interpretable results, which is problematic for safety-critical systems and other applications.

Consequently, my research focuses on integrating thermodynamic knowledge with ML methods to develop hybrid models. The aim is to improve the predictive accuracy and range of applications of the models while ensuring thermodynamic consistency. My research focuses on developing models to predict the properties of complex mixtures.

Lecture Support

• Thermodynamics of Electrolyte Solutions (WS 2026/27)

Office Hours

By arrangement

Publications

• N. Hayer, T. Wendel, S. Mandt, H. Hasse, F. Jirasek: Advancing thermodynamic group-contribution methods by machine learning: UNIFAC 2.0, Chemical Engineering Journal 504 (2025) 158667. [DOI]

Presentations

Posters