A Manufacturing-Oriented Diffusion Framework for Generative Product Design with Traditional Cultural Resources
Year: 2026
Type: article
Abstract: Traditional cultural resources contain rich symbolic and visual knowledge but remain difficult to incorporate into manufacturing-oriented generative design workflows because cultural semantics, visual characteristics, and pro-duction requirements are often represented separately. This study presents a manufacturing-oriented generative de-sign framework that integrates traceable cultural-resource preparation, structured cultural and manufacturing de-scriptions, and LoRA-based adaptation of Stable Diffusion XL (SDXL). A candidate cultural-resource inventory is examined through metadata auditing, while ResNet-50 features are used to characterize visual representations and to investigate the sensitivity of image features to common augmentation operations. Cultural semantic requirements and manufacturing-oriented requirements are organized as structured generation conditions for product-concept synthesis. In addition, a design-stage assessment framework is formulated in terms of structural feasibility, process compatibility, material adaptability, and visual consistency. Comparative experiments across four generation set-tings further indicate that the proposed framework achieves better performance in cultural consistency, design adaptability, constraint satisfaction, and expert-rated manufacturing feasibility. Although the current evaluation remains at the design stage rather than physical manufacturing validation, the proposed workflow provides a repro-ducible basis for connecting traditional cultural resources with manufacturing-oriented generative product design.
DOI: 10.4108/eetsis.14571
Authors: Yingru Ji, Jing Wu
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225664029
FWCI:
Cites:
Cited by:
Related to: