Coupling of intrusive and non-intrusive locally decomposed model order reduction techniques for rapid simulations of road systems

On the one hand, the Digital Twin Road of the Road System is based on multi-physical models, which are developed in different subprojects of the SFB/TRR 339. On the other hand, other SP provide pure data or data correlations (e.g. using sensor measurements), which the Digital Twin uses to assess the current state of the road system and for further decision-making. A major technical challenge arises from the fact that the coupling of the physics- and data-based models must be achieved in such a way that a very fast analysis and controlling of the coupled system becomes possible. In the long term, real-time simulations are envisaged.

A promising and novel approach to tackle this problem, pursued by SP B05, is based on the combination of so-called intrusive and non-intrusive model order reduction (MOR). While intrusive MOR is based on the availability of physical models and is performed in SP B05 by a further development of already established methods in the field (e.g. Proper Orthogonal Decomposition (POD) and Gappy POD, see the figure on the right for an exemplary illustration), this is not the case for non-intrusive MOR. For the latter, state-of-the-art data-driven methods as well as machine learning concepts are used and extended. As a final result, purely data-based input-output correlations can be linked to physics-based models.

The new method is used to create a substitute model for the detailed tire-road system, making real-time simulations within the Digital Twin of the Road System feasible. Furthermore, the developments in SP B05 are important to enable fast simulations of the road system considering data uncertainties.