2025, Masterarbeit, Sarthak Sali
 

Betreuung durch Dr.-Ing. Damun Mollahassani

This thesis explores how to support engineers in keeping SysML v2 models complete and consistent over time. In practice, SysML diagrams can silently drift away from the intended system because of human or machine error. Small mismatches accumulate, and important elements or links simply go missing without being detected during ongoing modeling work. Also, some links and components that should be present in the SysML model according to the domain knowledge are not added at all or effectively become hidden.

This thesis tackles this problem by investigating whether a domain ontology can serve as an external reference point for SysML v2 models. A domain ontology captures the domain's key concepts and relationships and can therefore help reveal such issues in the SysML model. In the approach designed in this thesis, both the model and the ontology are first translated into a common triple-based representation of subjects, predicates, and objects. In addition, a pipeline based on natural language processing (NLP) methods and a large language model (LLM) compares the two graphs, aligns names and relations, and highlights potential mismatches between “what the ontology says should exist” and “what actually appears in the SysML model.”

A Tkinter-based GUI, along with the approach, has been developed to present these findings to the user. It lists matches in tables, shows similarity scores and local graph context, and can both generate paste-ready SysML v2 snippets and directly insert missing elements at plausible positions in the existing model hierarchy. The approach is evaluated on an Electric Bike architecture case study, where it recovers meaningful missing components and requirements and suggests sensible insertion positions. Overall, the thesis shows how combining ontologies and SysML v2 models can help engineers use domain knowledge to improve model completeness and traceability.