Application of Generative Adversarial Network for Designer Assistance System

With the continual innovation of technology and product design methods, consumer needs change, and there is a gradual shift in product development from a function-oriented to consumer-oriented approach. The transformation of consumer perceptions into product design elements has become the focus of designers. This study develops an automated design generation system for product forms, which assists product designers to rapidly incorporate the consumer perceptions on the product form into the design process. It combines the Kansei engineering theory with the deep convolutional generative adversarial network (DCGAN), using the side-view contouring of cars as a case study. A questionnaire is given to the participants for selecting the top six subjective terms perceived by consumers in car styling. Further, 1006 car side views are scored on the six Kansei characteristics using a semantic differential scale. The results are clustered and analysed using fuzzy c-means. Subsequently, the clusters are applied as training datasets for the DCGAN, which generates antagonistic data to support the design process. The results of this study show that the data generated through the proposed method can facilitate designers in adjusting the product shape and realising an efficiently product shape design that matches the consumer's desired imagery.