@inproceedings{Wang:2023,
author = {Wang, B and Zuo, H and Cai, Z and Yin, Y and Childs, P and Sun, L and Chen, L},
title = {A task-decomposed AI-aided approach for generative conceptual design},
year = {2023}
}

Download

TY  - CPAPER
AB  - Generative algorithm-based conceptual design has beeninnovatively applied as an emerging digital design paradigm forearly-stage design ideation. With powerful large languagemodels (LLMs), designers can enter an initial prompt as a designrequirement to generate using machine reasoning capabilitydescriptive natural language content. The machine-generatedoutput can be used as stimuli to inspire designers during designideation. However, the lack of transparency and insufficientcontrollability of LLMs can limit their effectiveness whenassisting humans on a generative conceptual design task. Thisgeneration process lacks theoretical guidance and acomprehensive understanding of design requirements, whichmay potentially lead to generated concepts that are mismatchedor lack creativity. Inspired by the Function-Behavior-Structure(FBS) model, this paper proposes a task-decomposed AI-aidedapproach for generative conceptual design. We decompose aconceptual design task into three sub-tasks including functionalreasoning, behavioral reasoning, and structural reasoning.Prompt templates and specification signifiers are specified fordifferent steps to guide the LLMs to generate reasonable results,controllably. The output of each step becomes the input of thenext, aiding in aggregating gains per step and embedding theselection preferences of human designers at each stage. Aconceptual design experiment is conducted, and the results showthat the conceptual design ideation with our method are morereasonable and creative in comparison to a baseline.
AU  - Wang,B
AU  - Zuo,H
AU  - Cai,Z
AU  - Yin,Y
AU  - Childs,P
AU  - Sun,L
AU  - Chen,L
PY  - 2023///
TI  - A task-decomposed AI-aided approach for generative conceptual design
ER  - 

Download