Date of Award
6-2026
Document Type
Thesis
Degree Name
Master of Science in Astronautical Engineering
Department
Department of Aeronautics and Astronautics
First Advisor
David W. Meyer, PhD
Abstract
Modern defense systems continue to grow in complexity, placing increasing pressure on engineering workflows to be faster and more adaptable. While Model-Based Systems Engineering (MBSE) with the emerging SysML v2 standard provides a framework for capturing system behavior, its practical use is often limited by the expertise and time required for manual modeling. This research investigates whether large language models (LLMs) can help overcome that barrier by automatically generating SysML v2 state machines from Guidance, Navigation, and Control (GNC) textual inputs. Three LLM Flowise-based models were developed and evaluated: the Structured Transformation Model (STM), which uses a structured extraction and JSON coding process; the Example-Guided Structured Model (EGSM), which pairs a structured extraction with example-driven generation; and the Example-Guided Contextual Model (EGCM), which applies flexible parameter extractions and example-driven generation with SysMLv2 definitions. Each model was tested using technical documents of three GNC subsystems: an Electro-Optical seeker, an Inertial Stellar Compass, and a Hybrid Navigation System. Outputs were evaluated against ten metrics assessing syntactic correctness, semantic fidelity, and behavioral coherence. Local LLMs between 7b and 14b parameters were prioritized for their customized performance, privacy, and model autonomy, with comparisons made to the larger models GPT-4o and Claude 3.7. Results show that all approaches successfully generated valid, reusable SysML v2 models to different levels of fidelity. Local models, especially those using example-guided strategies, achieved solid performance across most metrics and required minimal correction. These findings demonstrate that LLMs can support automated SysML v2 model generation and significantly reduce the technical burden associated with MBSE, potentially enabling broader adoption in defense-related system development.
AFIT Designator
FY25-AFIT-ENY-MS-J-002
Recommended Citation
Ames, Andrea Louise, "Leveraging Large Language Models (LLMs) for Automated Generation of SysML v2 State Machines in Guidance, Navigation, and Control (GNC) Systems" (2026). Theses and Dissertations. 8348.
https://scholar.afit.edu/etd/8348
Comments
An embargo was observed for posting this thesis on AFIT Scholar.
Approved for public release, PA case number 88ABW-2025-0544.