An automated functional decomposition method based on large language models
Author(s)
Type
Conference Paper
Abstract
Functional decomposition in conceptual design aims to break down the overall function of a complex product or system into smaller, implementable subfunctions. This process enables designers to systematically explore solution spaces, improve problem structuring, and manage design complexity. However, traditional functional decomposition presents substantial cognitive challenges, as designers often struggle to identify functionalities at suitable abstraction levels, adhere to structured syntax format requirements, and maintain solution neutrality. Such difficulties may result in inconsistencies in decomposition quality and inefficiencies throughout the design process. To mitigate these challenges, this research proposes an automated functional decomposition method leveraging a Large Language Model (LLM). A comprehensive function basis is constructed from a large-scale dataset of award-winning product and concept designs. This function basis serves as the vocabulary base for the automated decomposition process. The proposed method iteratively employs an LLM-powered functional decomposer to generate subfunctions while integrating a filtering mechanism to screen outputs and determine whether decomposition is complete. A comparative experiment evaluating designer-generated versus LLM-generated function trees demonstrates that the proposed method significantly enhances functional ideation breadth and improves decomposition quality and efficiency. The findings suggest that LLM-assisted decomposition can effectively support conceptual design, offering a scalable and systematic alternative to traditional manual approaches.
Date Issued
2025-10-27
Date Acceptance
2025-08-01
Citation
International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, 2025, Volume 3A: 51st Design Automation Conference (DAC), pp.1-10
ISBN
978-0-7918-8922-0
Publisher
American Society of Mechanical Engineers
Start Page
1
End Page
10
Journal / Book Title
International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
Volume
Volume 3A: 51st Design Automation Conference (DAC)
Copyright Statement
Copyright © 2025 by ASME; reuse license CC-BY 4.0 (http://creativecommons.org/licenses/by/4.0/)
License URL
Source
ASME 2025 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
Publication Status
Published
Start Date
2025-08-17
Finish Date
2025-08-20
Coverage Spatial
Anaheim, California, USA
Date Publish Online
2025-10-27