摘要
类比推理是指从熟悉的领域或先前的经验中提取元素,并将该元素运用到新的概念中,是设计师在产品外观造型设计阶段最常用的设计方法之一。以汽车外观设计作为载体,从设计的逻辑思维过程与三维生成式对抗模型特点入手,提出了基于类比推理设计的模型自动生成的思路与方法,将隐式场解码器(IM-NET)嵌入到生成对抗网络(GAN)得到IM-GAN,实现类推式快速建模。首先,运用公共数据集ShapeNet中的汽车、飞机模型,构建设计对象、启发源数据集。其次,使用IM-GAN对数据集中的模型进行训练。然后,在模型训练完毕后,通过在生成器的隐空间进行插值,得到任意两种已有生成物之间的任意程度的中间插值样式,探求产品三维造型的类推设计创新方案。最终,从设计师的角度,将隐函数表征与点云、体素表征方式进行比较,探讨不同的几何表征方式在辅助设计师进行造型设计时的优缺点。实验结果表明:基于隐函数表征的模型可以得到较好的辅助设计效果,并且启发设计师更多的产品造型创意,为其他工业产品的设计开发提供了经验参考。
Analogical reasoning refers to extracting elements from familiar fields or previous experience and applying them to new concepts.It is one of the most commonly used design methods by designers in the stage of product shape design.Taking automobile exterior design as the carrier,starting from the logical thinking process of design and the characteristics of 3D generative adversarial model,this paper proposes the idea and method of automatic model generation based on analogical inference design,combined with the IM-GAN obtained by embedding the implicit field decoder(IM-NET)into the generative adversarial network(GAN),to realize the analogous fast modeling.Firstly,the car and aircraft models in the public data set ShapeNet are used to build a design object and inspiration source data set.Then,after the model training is completed,by interpolating in the latent space of the generator,an intermediate interpolation style of any degree between any two existing products can be obtained,and an innovative scheme for analogy design of product 3D shape can be explored.Finally,from the designer's point of view,this paper compares the implicit function representation with point cloud and voxel representation,and discusses the advantages and disadvantages of different geometric representation methods in assisting designers in shape design.The experimental results show that the model based on implicit function representation can obtain better auxiliary design effects and inspire designers to have more product modeling ideas than the other ones.
作者
陈育苗
李雅婷
CHEN Yumiao;LI Yating(School of Art Design And Media,East China University of Science and Technology,Shanghai 200237,China)
出处
《机械设计与研究》
CSCD
北大核心
2024年第2期95-101,共7页
Machine Design And Research
基金
国家自然科学基金资助项目(51905175)
上海市2023年度"科技创新行动计划"软科学重点项目(23692104700)
教育部产学合作协同育人项目(220701710291027)
谢友柏设计科学研究基金资助项目(XYB-DS-202301)。
关键词
生成对抗网络
类比推理设计
汽车造型设计
生成设计系统
analogical reasoning
generative adversarial networks
car shape design
generative design system