KGG4SE: A Knowledge Graph Generation Framework for Systems Engineering

In this paper we introduce the Knowledge Graph Generation Framework for Systems Engineering (KGG4SE). Based on the GENIAL! Basic Ontology (GBO), a variety of large language models and prompt engineering, we generate numerous knowledge graphs from a diverse set of input sources. One of the key features of the framework is the reasoning-in-the-loop integration. The generated classes are structurally consistency checked and inconsistent classes are removed. Further features include the generation of research articles, technology videos and datasheets. Also quality control prompts are used and the framework is integrated into a system engineering tool (SysMD) with frontend and backend. This makes content of the knowledge graph accessible to users in MBSE (Model-Based Systems Engineering) ecosystems. Finally we outline results of the generation process and content of the graphs as well as the reasoning process with disjoint axioms. The results show an improved graph quality and structure in comparison to existing approaches. In terms of succinctness and conciseness, we remove an overall of 67.4% of classes that do not adhere to the ontological entailment or domain. Although graph quality is sometimes difficult to qualify and quantify and the implementation needs further work, we believe the methodology of this framework is a good way forward to produce better quality graphs, which are scalable.