AI-Assisted SysML v2 Model Generation from Textual Requirements

Overview

PoC #3, presented by the MSE of RWTH Aachen University, demonstrates the automated generation of SysML v2 models from natural language requirements. Using a window opener system as a reference, this project illustrates how AI can extract both architectural structures and behavioral patterns (states/conditions) from simple text.

Methodology & Technical Stack

The core of this POC is a sophisticated integration of GPT 5.1 and a rule-based prompt engineering strategy. The process is divided into several analytical steps:

• Context Extraction: The AI identifies "Logical Parts", "Actions", and "States" within the requirements (e.g., recognizing an Emergency Stop System as a component with specific delay attributes).
• Multi-Layer Architecture: The generated SysML v2 code follows a strict ontology, covering the Functional Layer, Logical Layer (including state machines), and the Physical Layer (product structure).
• Few-Shot Prompting: To ensure syntactical correctness within the new SysML v2 standard, the system uses domain-specific examples and rigorous rules for performance requirements (e.g., correctly interpreting tolerances like "±10%").

Key Features

• Requirement Formalization: Numerical values and constraints are automatically translated into formal SysML attributes.
• Traceability: The generated model maintains direct links between functions and the requirements they satisfy ("satisfy" relationships).
• Behavioral Modeling: Beyond static structures, the tool derives complex State Machines to describe system dynamics, such as obstacle detection logic.

Impact

By automating the transition from text to formal model, KIMBA drastically reduces the manual effort for systems engineers and ensures that the resulting architecture is consistently aligned with the initial specifications from the very first draft.