From Documents to Models: Adopting MBSE Practices in Complex Embedded Systems and Leveraging Large Language Models

Abstract

This thesis investigates the transition from Document-Based Systems Engineering to Model-Based Systems Engineering while using Base Transceiver Station development at Nokia as a case study. Additionally, the thesis explores the potential of Large Language Models to facilitate this transition. The research includes a theoretical analysis through a literature review on MBSE and a case study of initial MBSE adoption efforts at Nokia. A proof-of-concept plugin for the MagicDraw modeling tool that integrates OpenAI’s GPT-4o to assist in systems engineering tasks is developed and evaluated. Multiple methodologies are employed in the thesis including literature review, case study analysis, proof-of-concept implementation, and experimental evaluation. The developed plugin performed well in the evaluation, achieving high accuracy in model understanding and model modification tasks, while minimizing token usage of the model representation. The results suggest that LLMs have potential in assisting in systems engineering tasks in modeling tools. Furthermore, MBSE adoption can benefit BTS development through improved system understanding and traceability, although organizations face challenges including high upfront investment and resistance to change. Identifying theoretical insights into MBSE adoption benefits and challenges as well as practical solutions through AI integration are the main contributions of this thesis, while also identifying some areas for future research in real-world applications of LLM assistance in systems engineering.