Measurement and Control Engineering – Mechatronics Working Group

The core topic is learning models from data. These models can then be used for a wide variety of applications, such as the optimization or design of control, regulation, and diagnostic systems. We work with both static and dynamic models, although the latter present significant challenges in terms of both excitation signal design and the representation of dynamics within the models.

We see particular potential in so-called gray-box modeling, where physical engineering understanding is combined with data to create a robust and adaptive model. The automatic selection of relevant input variables and metamodeling (a model of a model) also address this frontier area.

The quality of the data distribution is a bottleneck for the creation of data-driven models. Information not contained in the data cannot be described by the model. Therefore, our research also focuses heavily on new methods of design of experiments (DoE) and extends to active learning.