Large Language Model-supported approach for requirements extraction from different sources of reference system elements

This paper introduces an approach for LLM-supported requirements extraction from different data sources within SGE – System Generation Engineering. Developing new mechatronic system generations involves gathering requirements from heterogeneous artifacts, such as specifications, standards, competitor materials. Manual extraction is often time-consuming and error-prone. The proposed approach, developed through industry collaboration and academic projects, incorporates a five-step workflow comprising source identification, preparation, prompting, pre-check, and final check, as well as methodological support, such as a standardized prompt template and INCOSE criteria, to enable requirements extraction. Key features include applicability across different data sources, traceability, domain adaptability, reproducibility, expert validation, and scalability. An initial exploratory evaluation during a semester-long, practice-oriented live-lab indicates the approach’s applicability and potential benefits, such as reduced manual effort, traceability, and the possibility of using it as sparring partner. However, the evaluation also highlights challenges, such as the importance of human involvement to ensure context awareness and requirements quality.

Content

DDC Class

006.3: Artificial Intelligence

620: Engineering