ABSTRACT In regulated domains such as aerospace or automotive software, compliance with standards like ECSS and ISO/IEC 26262 is mandatory to ensure safety and reliability. Artifacts, such as process or product models, lay the foundation for planning and managing development projects, as well as for measurement and evaluation tasks. Manually extracting such artifacts from PDF‐based standards is time‐consuming and error‐prone. To address this issue, we developed a large language model (LLM)‐based approach for automatically generating artifact models from standards. However, the evolution of AI models constitutes a challenge, since the extraction results might differ when updating the LLM to a newer version. In this article, we present an enhanced approach that strengthens the artifact‐generation process against such challenges. The approach generates machine‐readable artifact models that allow for automated measurement systems, for example, for monitoring compliance and controlling complex projects. The performance of the artifact model generation was evaluated using the ECSS standards from the aerospace domain and resulted in an average completeness of 100.00% and an average precision of 71.33% of the generated models.
Building similarity graph...
Analyzing shared references across papers
Loading...
Straub et al. (Sun,) studied this question.
www.synapsesocial.com/papers/69ba42ee4e9516ffd37a3a51 — DOI: https://doi.org/10.1002/smr.70093
Philipp Straub
Mustafa Bülbül
Robin Korfmann
Journal of Software Evolution and Process
Reutlingen University
TRUMPF (Germany)
Building similarity graph...
Analyzing shared references across papers
Loading...