LLM-assisted approach bridges semantic gaps in collaborative systems engineering with SysML v2

Cross-organizational teams building complex systems often struggle to ensure their independently created models can work together seamlessly. A new research paper presents a method to tackle this problem using large language models (LLMs) and the latest version of the SysML modeling language.

The challenge of model alignment#

When multiple teams develop system models in isolation, subtle differences in meaning and structure can prevent clean integration. This semantic alignment problem is particularly thorny in Model-Based Systems Engineering (MBSE), where accurate, consistent modeling across organizations is crucial. The new SysML v2 standard offers stronger structural modularity and formal semantics that provide a better foundation for making models work together.

How LLMs help bridge the gap#

The proposed approach combines two powerful tools: the semantic understanding capabilities of GPT-based language models and the advanced features in SysML v2. Researchers developed a structured, prompt-driven method that guides an LLM through iterative steps to understand and align models. The process involves extracting key model elements, matching semantically equivalent concepts between systems, and verifying alignment accuracy.

Rather than forcing hard transformations of models, the technique implements what researchers call "soft alignment" — leveraging SysML v2 features like aliases, imports, and metadata extensions to create traceable links between models without permanently altering them.

Proof of concept#

The team demonstrated the approach using a GPT-based LLM in a practical example involving a measurement system. While the research shows promise, the authors acknowledge both the benefits and limitations of the method, suggesting this is an evolving area where human judgment and LLM assistance work together.