AI
LLMs and AI-Powered Graphs Show Promise for Automating System Design Analysis
A new study explores how language models and knowledge graphs can generate design matrices for complex engineering systems like satellites and power tools.
Researchers have investigated whether AI techniques can automate the creation of Design Structure Matrices (DSMs)—visual frameworks that map how components in a system depend on and interact with each other. The work applies large language models (LLMs), retrieval-augmented generation (RAG), and graph-based retrieval methods to tackle this engineering challenge.
The Challenge of System Design Documentation#
Understanding the internal architecture of engineered products is critical for design, maintenance, and improvement. A DSM is a square matrix that captures these relationships systematically, but building one typically requires manual effort from engineers who must analyze component dependencies. The researchers wondered whether AI could streamline this process.
Testing Real-World Products#
The team evaluated their approach using two concrete examples:
- A power screwdriver — a familiar consumer product with well-understood design documentation
- A CubeSat — a standardized small satellite platform with publicly available architectural references
They assessed the AI methods on two tiers of difficulty: first, determining how predefined components connect to each other, and second, the harder task of identifying what components exist in the first place and then mapping their relationships.
Results and Future Directions#
Despite encountering design and computational hurdles, the researchers identified genuine opportunities for automating DSM generation. Performance varied depending on the task's complexity, with the simpler component-relationship mapping showing more promise than the autonomous component discovery phase.
The team has released all code publicly to enable reproducibility and welcome feedback from domain experts in engineering and systems design. This open approach aims to accelerate refinement of these AI-driven methods for practical use in product development workflows.
