Institute for Technologies and Management of Digital Transformation

Guided Visual Interactive Exploration and Labeling of Industrial Sensor Data

The dissertation of Dr.-Ing. Tristan Funken deals with the efficient exploration and labeling of industrial sensor data. An approach is presented with which the generation of high-quality, labeled data sets, which are required for machine learning algorithms, can be supported by a guidance system along the two process steps, exploration and labeling.

We interviewed Tristan about his dissertation:

What was the context of your dissertation? What projects or other factors particularly influenced your dissertation?

This dissertation contains the results of my research work at RWTH Aachen University and the University of Wuppertal. In various research projects that I have carried out together with industrial companies, I have found that the application of machine learning methods only accounts for a fraction of the project work time. Much more time is spent on processing the sensor data and understanding the underlying process. In contrast to familiar ML scenarios such as the recognition of road signs or the classification of dogs and cats, it is more difficult to understand the course of a production process or error case based on curves from pressure sensors attached to machines. This makes it more difficult to pre-process the data for the application of ML algorithms in such cases and also prevents this process from being scaled by, for example, Mechanical Turk approaches.

How does your work contribute to the field of research?

The idea for my dissertation arose from the problem described above of supporting the process of pre-processing data, in particular exploration and labeling, in order to be able to apply ML models more quickly in industrial use cases.  In my work, I first formalize the process and then present the approach for a guidance system that can be used to guide both process steps. The feasibility and efficiency of this approach were proven by evaluating a prototype implementation and by means of studies with real use cases.

Certificate handover with Prof. Meisen

What does the future hold for you and the topic?

Investigating the usability of expert knowledge and the combination of human knowledge and interactions with machine learning methods will remain an important topic in the coming years. Recently, a lot of attention has been paid to the use of Large Language Models. The topic could therefore develop further towards intelligent assistance systems. I am very pleased to be able to further advance AI research and teaching in the KI4BUW network at the University of Wuppertal.

Weitere Infos über #UniWuppertal: