Duration: October 2022 - September 2025
Total project volume (university): 617,716 euros
Sponsored by: German Research Foundation (DFG)

Project in SPP 2353 "Daring to be more intelligent - design assistants in mechanics and dynamics"

The priority programme SPP2353 "Daring More Intelligence - Design Assistants in Mechanics and Dynamics" aims to develop an assistance system for the semi-automated design of technical systems, combining methods from the fields of optimisation, artificial intelligence, dynamics and mechanics.

As part of the HyM3 project, a flexible and adaptive, data-driven framework for the multi-criteria optimisation of complex multi-body systems is to be developed. The multi-criteria optimisation of multi-body systems involves a large number of model calls and evaluations, resulting in a conflict between the computing time of the design and the accuracy of the model: Accurate models, which are necessary for a good identification of the optimal design, are usually computationally expensive. Conversely, fast models, which would lead to an acceptable computational effort, are usually rather inaccurate. This problem is exacerbated by the fact that in multi-objective optimisation not a single optimum, but the entire set of optimal compromises (the Pareto set) must be calculated.

To counteract this problem, it should be possible to adapt the model used during the optimisation process. The physical model used is reduced in order to save computing time for model calls that do not require high accuracy. The inaccuracies resulting from the reduction are then corrected with the help of data-driven model components. The full physical model, which can also be improved by data-driven additions, is used for model calls where high accuracy is required.

In the project, the Chair of Dynamics and Mechatronics is mainly concerned with the adaptivity of the models used through the use of hybrid modelling. Building on this, the Data Science for Engineering specialist group is developing efficient, data-based multi-objective optimisation methods in order to reduce the number of expensive simulations as much as possible.