Generative Models for Feature-Based Product Development as a Basis for Hybrid Decision-Making
DS 125: Proceedings of the 34th Symposium Design for X (DFX2023)
Year: 2023
Editor: Dieter Krause, Kristin Paetzold-Byhain, Sandro Wartzack
Author: Valentin Schemmann, Thorsten Schmidt, Niels Demke, Frank Mantwill
Series: DfX
Institution: Institute of Machine Elements and Computer Aided Product Design (MRP), Helmut Schmidt University Hamburg
Page(s): 173-182
DOI number: 10.35199/dfx2023.18
Abstract
This paper investigates the general possibility for applying generative models in the early phase of product development. For this purpose, the fundamentals of feature-based product development are introduced and related to the development methodology VDI 2221 alongside a brief overview of deep generative models. Based on this, a conceptual framework is developed that combines the methods and proposes a collaborative approach. In conclusion, a prototypical implementation is performed by training a StyleGAN2 based on vehicle profiles followed by executing a GANSpace principal component analysis. Finally, the various results are presented and the possibilities of manipulating the generated images based on identified features are discussed and transferred back into the product development process.
Keywords: Feature-Based Product Development, Deep Generative Models, Principal Component Analysis, StyleGAN2, GANSpace