BUP47 - STRIKE-FS

In order to remain competitive with the rapid development in the field of mobility and production technologies, efficient and innovative approaches in system analysis and optimization must be developed. The V-model of Model-Based Systems Engineering promotes multi-disciplinary model views and supports system analysis with special methods of validation and verification. Nevertheless, there are still deficits in the explanatory capability of the system analysis, which are to be remedied in the project using a Large Language Model (LLM). To this end, digital models of a FORMULA student racing car are generated automatically using a machine-executable V-model. The scientific challenge in the project is to explain the implicitly coded couplings in these machine-generated information representations using an LLM in order to optimally support humans in system analysis.

Aim

The aim of the project is to integrate an adaptive language model into design languages. This model should be able to identify key influencing factors in the design process (design drivers) and classify them into supporting (protagonists) and inhibiting (antagonists) factors.

The model should answer natural language queries about formal representations and identify suitable information measures as evaluation metrics. An example of this would be: "How does a material change of the wishbones affect the design and what influence does this have on the vehicle as a whole?" The model should understand this question and define the technically necessary changes. Corresponding information measures as evaluation metrics can be the vehicle mass and aerodynamic drag as well as the associated influence on fuel consumption and lap time.

Approach

• Integration of an adaptive language model (LLM) in design languages
• Identification of design drivers and classification into protagonists and antagonists
• Answering natural language queries about formal representations such as class diagrams and activity diagrams
• Identification of suitable information measures as evaluation metrics for the outputs

Benefit

The integration of an adaptive language model in design languages enables future users to efficiently identify and classify design drivers. The natural language queries make complex formal representations easier to understand, leading to improved communication and more effective decision-making. The use of suitable information measures as evaluation metrics enables a precise evaluation of the outputs, which ultimately leads to higher quality and efficiency in the design work. For the future vehicle owner, this means that motors will be better, more reliable and more efficient.

Further information

• Research area design theory and similarity mechanics at the Institute of Aircraft Design