KnowTD: Translation of thermodynamic knowledge to computers

This project addresses one of the fundamental problems of artificial intelligence: the translation of human knowledge to computers. Our goal is to achieve this for the extensive and deep knowledge of thermodynamics. In preliminary work, we have shown a feasible way to do this, in which thermodynamic knowledge is represented by interacting graphs. However, we do not limit ourselves to the representation of knowledge, the system to be developed, KnowTD, will be able to build thermodynamic models of real world objects, analyze them, and autonomously answer questions about their thermodynamic behavior. Thus, thermodynamic knowledge will become operable on the computer, with or without human interaction.

The project is funded under the German Research Foundation's Priority Program 2331 "Machine Learning in Process Engineering: Knowledge Meets Data: Interpretability, Extrapolation, Reliability, Trust" as a tandem project with partners from the Computer Science Department of TU Kaiserslautern, whose work includes the development of a linguistic input/output for the system for KnowTD based on deep learning techniques. A close co-operation in the tandem is essential for the project’s success.

download job opening (German version)

[back]