Assessment of large language models for use in generative design of model based spacecraft system architectures

This paper investigates using Large Language Models (LLMs) within Model Based Systems Engineering (MBSE) as the basis for generative design tools for spacecraft. This study has developed tooling for automatically generating system architecture, functions, modes, and components from an initial requirement set. Specifically, a Python tool was developed to couple the Capella MBSE tool to an LLM in order to facilitate a rapid generative design process. The approach was tested by application to three system design tasks: a European Space Agency Earth observation mission, a CubeSat payload design, and a masters’ degree level group design for an Earth observation spacecraft. For each, generated outputs were evaluated against those produced by technical designers. It was found that the generation of system modes and components was of good quality, providing high traceability and alignment against requirements, and also providing generated architectures that in some areas were more detailed than human-generated equivalents. Further development could provide spacecraft system engineers with an ‘AI design assistant’, as human input is still at the centre of the process and appears necessary to ensure a high-quality output.