Synergetic combination of data-driven and model-based methods for application to control engineering problems

In the course of digitalization, artificial intelligence and machine learning are currently receiving high attention from science and industry. In the engineering sciences, in our case in control engineering, data-driven methods are already used, but mainly as an alternative to physical modeling of dynamic behavior respectively to specialized methods of control design or in a pragmatic simple combination.




Therefore, the goal of the junior research group "DART - Data Driven Methods in Control Engineering" is to develop novel hybrid methods for control engineering problems by combining the established physically motivated methods with modern data driven methods to achieve the highest possible performance in control design. These hybrid approaches go far beyond simple, pragmatic combinations, because they are based on structurally well-founded compositions of tailored methods that synergistically combine their advantages. Typical design steps such as modeling and parameter identification of the physical system, observer design, controller design and commissioning of a controller are addressed. This enables us to extend all aspects of classical control engineering by hybrid approaches with data-based methods. The results are then combined on a test bench, so that we can evaluate the hybrid procedures across methods.

Our research focuses are divided into the following areas