Semantic-Aware Change Management for SysML v2 Using Large Language Models

Systems engineering models are growing in size and complexity, while SysML v2’s textual notation and API enable automation and scalable operations. We present a methodology for semantic-aware change management in SysML v2 that interprets high-level natural-language requests and translates them into controlled, verifiable CRUD operations over the SysML v2 API. The approach combines targeted context extraction via vector-based retrieval with scoped graph expansion to minimize token/cost budgets, LLM-guided change synthesis using structured tool calls, and layered local validation that separates syntactic conformance from semantic safeguards (e.g., duplication and intent preservation). A proof of concept on the open SysML v2 Pilot API demonstrates feasibility on a coffee machine model. The system proposes ordered edits (create, update, delete), enforces type- and relationship-aware checks, and commits traceable diffs while refreshing the retrieval index. First, we show how to systematically extract only the necessary model context via the SysML v2 API using vector-based retrieval with scoped graph expansion. Second, we verify LLM-proposed changes to the SysML v2 model through layered validation that combines generic and type-specific checks with optional LLM-assisted semantic review. We discuss limitations (large type specifications, repeated context reconstruction for related edits, server-assigned identifiers) and outline future work on adaptive retrieval, richer validators, multi-turn sessions, and governance. Together, these contributions provide a practical foundation for reliable, cost-aware AI support in MBSE change management with SysML v2.