In process engineering, one very often has to deal with mixtures that are so complex that a complete analytical elucidation of the composition is not possible, e.g. fermentation broths from biotechnological processes. Such poorly specified mixtures pose a particular challenge for process design and optimization, since thermodynamic modeling is not possible using classical methods. At LTD, novel approaches for thermodynamic modeling of such mixtures are developed. These are based on fingerprinting of the mixtures with respect to functional groups by means of NMR spectroscopy as well as the quantitative definition of pseudo-components, and make use of machine learning algorithms to automatically evaluate information from multiple analyses. Thermodynamic modeling can then be performed using group contribution methods. Building on current preliminary work, these approaches will be further developed in this research project. On the one hand, the algorithms will be extended to analyze additional functional groups, which will also require the inclusion of additional NMR experiments. On the other hand, it will be investigated to what extent results from further analytical methods, e.g. FTIR spectroscopy, as well as partially available prior knowledge can be integrated in a suitable way. Furthermore, the developed approaches are to be brought into industrial application within current cooperations, among others in the context of winemaking and biotechnological production.

[back]