基于深度学习的智能化工业产品外形设计

Intelligent Industrial Product Design Based on Deep Learning

传统的工业产品外观设计流程,往往依赖手工建模,存在成本较高、以往模型内部几何形状、关系等知识未得到重用的问题。文章提出了一种基于深度学习的三维建模流程,利用人工智能提供的理解、推理能力,学习模型库中三维模型蕴含的先验知识,用于后续的建模设计工作中。该流程通过图自动编码器学习以往三维模型的布局、尺寸特征,首先生成三维物体的零部件布局特征,接着合成零部件的细节部分,由粗到精的渐进式完成三维生成任务。实验结果表明,该流程能够得到合理的三维模型,生成的模型保留了可识别的外形特征以及几何特征。

Traditional industrial product appearance design often relies on manual modeling,which has the disadvantages of high cost and no reused knowledge of the previous model's internal geometry and relationships.This paper proposes a three-dimensional modeling process based on deep learning.With the understanding and reasoning capabilities provided by artificial intelligence,it learns from the prior knowledge contained in the three-dimensional models in the library for subsequent modeling and design work.This process learns the layout and size characteristics of the previous 3D model with a Graph Auto-encoder.First,it generates component layout features of the 3D object,then synthesizes the details of the component,and completes the 3D generation task to fineness.The experimental results show that this process can generate reasonable 3D models with recognizable shape and geometric features.