Duration: 01.09.22 - 31.08.25
Total project volume (university): 900,000 euros
Sponsored by: Federal Ministry of Education and Research

Motivation
Modelling complex dynamic systems is a challenge in many disciplines such as engineering, natural and social sciences. Corresponding mathematical models are the basis for many applications such as process monitoring and control of technical systems or the prediction of pandemics. Expert-based approaches reproduce the system behaviour in a robust and interpretable way, but often require a lot of time and human resources. For this reason, there has been a strong trend towards data-based modelling of dynamic systems using machine learning. These black-box models can be generated quickly with available software tools and without significant prior knowledge. However, their mode of operation and problem-solving strategy are difficult to understand and explain.

Aims and approach
The aim of the "ML-Expert" junior research group is to develop hybrid modelling of dynamic systems that incorporates both data and expert knowledge. Among other things, the group is investigating how a priori defined model structures and system-theoretical model properties can be integrated into the modelling process at different levels of abstraction. Based on this, automatable method and software packages are to be developed, which include data generation, the actual modelling process and the final validation. This should significantly accelerate and improve the modelling quality in terms of accuracy, robustness and complexity for various application domains.

Innovations and prospects
The solutions developed will be made available free of charge as open source. Efficient and resource-oriented data generation, modelling and validation will enable fast development cycles in the future, which is particularly relevant for industrial applications, e.g. in the automotive, energy or automation sectors. However, the planned work is cross-domain orientated.