TU Ilmenau graduate Theo Käufer gained worldwide recognition with a novel approach that combines real 3D temperature measurement data and physically informed AI for the first time. Today, he is researching turbulent flows and their prediction using artificial intelligence at the renowned Massachusetts Institute of Technology (MIT). He believes that one of the reasons for his success is the well-founded and practical training at TU Ilmenau - and the early confidence to implement his own ideas.
Hello Theo, you are currently a postdoc at the Massachusetts Institute of Technology (MIT), one of the most renowned technical universities in the world - but your path into science began at TU Ilmenau.
That’s right, I studied mechanical engineering at TU Ilmenau for both my Bachelor’s and Master’s degrees and then went on to do my doctorate in the Group of Engineering Thermodynamics. The reason why I chose the university back then was the Basic Engineering School. I made my decision to study mechanical engineering quite on the short term and the Basic Engineering School meant that I didn’t have to do a basic practical training, as this was integrated into the course. Ilmenau quickly won me over: small university, close supervision, lots of freedom.
What particularly characterized your time at TU Ilmenau?
Definitely the opportunity to implement my own ideas early on. I have been working as a student assistant on research projects in the Group of Engineering Thermodynamics since my third semester. After my bachelor’s degree, my current doctoral supervisor, Christian Cierpka, arranged a research stay for me at the von Karman Institute for Fluid Mechanics in Brussels. My time there had a big impact on me and I decided to do a PhD. I was then able to conduct independent research during my master’s degree in Ilmenau and published my first scientific article. When I received the offer to do a PhD, I gratefully accepted. And now I’m doing postdoctoral research at MIT. But I want to emphasize that none of this would have been possible without the support of many wonderful people, both personally and professionally.For me, the Group of Engineering Thermodynamics has proven to be an ideal learning and research environment. Christian (Prof. Cierpka, editor’s note) was always open to new ideas and supported them, even if they were costly at first. There was a time when I was jokingly greeted by him with the words "How much will it cost this time?" when I entered his office. Apart from the financial support, here I am particularly grateful to the German Research Foundation and the Carl Zeiss Foundation, he was always on hand with help and advice and made it possible for me to present my research at international conferences right from the start. But also the colleagues at the institute with whom I was allowed to work are fantastic and there is an open and friendly atmosphere, characterized by cooperation and collaboration instead of competition. All of this allowed me to enjoy doing research - even outside of my PhD topic, such as on a quantum computing project for which we won second prize in the Fujitsu Quantum Simulator Challenge 2025. I can only recommend the work and research at the Institute of Thermo and Fluid Dynamics, even if the subjects are often not my classic favorites.
In your PhD, you investigated turbulent heat flows - and incorporated newly developed 3D measurement technology.
I developed a measurement technique that allows you to simultaneously measure the temperature and velocity of a fluid, such as water, in three dimensions - with high spatial and temporal resolution. To do this, I used particles of so-called thermochromic liquid crystals, which change color depending on the temperature. Using several cameras, I was then able to track the movement of these particles in space and at the same time derive the temperature from the particle color. This data then formed the basis for the next step: we trained a physically informed neural network that not only works with the data, but also with physical equations - e.g. the Navier-Stokes equations. The aim was to extract additional information from the data - such as predicting the temperature from velocity data alone. With the measurement technique I developed, this approach could be applied and validated purely on experimental data for the first time. It worked pretty well and was a real innovation.
Was your PhD also the door opener for MIT?
Yes, indirectly in any case. We published the results on measurement technology. The research group from Brown University, which invented these physically informed neural networks, then contacted me - they had a suitable AI approach and were interested in a collaboration. This collaboration resulted in our joint paper , which we published in the journal "Science Advances". This connection also brought us into contact with MIT, as our cooperation partner was involved in a project there. I asked him if he knew anyone who was looking for someone - and that’s how I ended up at MIT.What is the article in "Science Advances" about?
The article is about how to better understand and analyze thermal turbulence using a combination of experimental measurement techniques and artificial intelligence. The core is that we use so-called physics-informed neural networks, i.e. neural networks that take physical laws into account in addition to the measurement data. I provided the measurement data: Speed and temperature at thousands of points simultaneously and spatially distributed, something that didn’t exist before. Brown University then developed a neural network that can predict the temperature, for example - even in places where none was measured. This works because the network learns the physical relationships. Our paper shows that this approach also works with real data for the first time - not just in simulations.You just mentioned the collaboration with Brown University. How did the collaboration between TU Ilmenau and the American university work?
Very organically. After Brown University contacted me, we quickly realized that it would work well together - our colleagues had the expertise in machine learning, and we had the expertise in measurement technology and the experimental data that simply didn’t exist before. So we combined our approaches: Brown developed the algorithm, we provided the data and brought in the physical background.
What does your day-to-day research at MIT look like?
I research turbulence behind moving objects. We have a so-called drag tank in which we move objects through still water in a controlled manner and measure the currents. We use this to develop new machine learning models. The research group is very interdisciplinary: other teams analyze fish movements, work with satellite images or build underwater robots. I am responsible for turbulence.How does the research at MIT differ from that in Ilmenau?
The research here is very large, visible and strongly networked. Delegations from politics and industry come here all the time. There are enormous opportunities here for me to develop and broaden my horizons. Boston and Cambridge in particular are great places to live. At the same time, I miss nature. Ilmenau is quieter, greener - but you also had more time for in-depth work.Does that mean you want to come back?
I’m keeping everything open. Maybe I’ll stay in science, maybe I’ll go into industry. A return to Germany is definitely conceivable. And I wouldn’t rule out Ilmenau either. I learned a lot there - and the creative freedom was worth its weight in gold.You can find the paper in the journal "Science Advances" here:
www.science.org/doi/10.1126/sciadv.ads5236
