Augmenting Aerospace System Design Using Large Language Models

Abstract

The democratization of artificial intelligence, exemplified by the widespread availability of ChatGPT since late 2022, presents significant opportunities for innovation in  engineering system design. This paper explores how large language models (LLMs), particularly ChatGPT, can support and partially automate engineering design processes by generating system configuration rules and conceptual system architectures.   The methodology is demonstrated using case studies on hybrid-electric propulsion systems and actuation system design. LLMs are used to generate system configurations represented as UML component diagrams. Recognizing the non-deterministic behaviour of LLMs, a structured prompting methodology is proposed. This includes reusable prompt templates combined with embedded examples (micro templates), allowing increased reproducibility and specificity.   Furthermore, the study shows how LLMs can be used to generate Python scripts that act as configurators, producing system descriptions within specified design freedoms. These scripts enable iterative expansion, refining the architecture over time. The outputs are exportable to simulation environments, allowing further analysis and optimisation.   Integration with LLM APIs within engineering tools enables interactive expansion of system diagrams into subsystems and detailed components, streamlining complexity management. LLMs also demonstrate potential for embedding regulatory and functional requirements into generated designs, aiding compliance.   These findings highlight the potential of LLMs to transform aerospace system design by improving efficiency, traceability, and early-phase exploration of system architectures.