LLM-powered Cyber-Physical System Features Extraction

Guillaume Nguyen, Xavier Devroey

Track: Type E (Late-Breaking Abstracts)

Keywords: Verification, Requirements Engineering, Compliance, Legal, Regulatory, LLM, Feature Model

Abstract: Compliance assessment of Cyber-Physical Systems (CPSs) with legal requirements remains a complex and critical task in software engineering. Current approaches often lack traceability between source code, requirements, and legal basis. Furthermore, the abstraction gap makes the task challenging and requires mastery of the different abstraction levels. In this abstract, we propose a semi-automated method that bridges that gap by leveraging Large Language Models (LLMs) to extract features from CPS source code using a Feature Model specified in the Universal Variability Language (UVL). Our approach shows the promise of LLMs for traceable compliance verification while identifying key challenges in domain-specific expertise and legal-technical alignment.

Submission Number: 84