Session: DAC-13-01 Geometric Modeling and Algorithms for Design and Manufacturing

Paper Number: 69544

Start Time: August 18, 10:00 AM

69544 - Design Concept Generation With Variational Deep Embedding Over Comprehensive Optimization 

This paper proposes a framework for generating design concepts through the loop of comprehensive exploitation and consequent exploration. Any sophisticated optimization such as topology optimization with diversely different conditions is utilized for the former. A variational deep embedding (VaDE), which is one of the deep learning techniques, is used for realizing the latter. In the process of design concept generation, first, a set of various possibilities of design entities is generated by exploitation through computational optimization. Second, they are learned by VaDE. This learning encodes the clusters of similar entities over the latent space with smaller dimensions. The clustering result reveals some voids where new design concepts are expected to be explored. Third, some possibilities of new design entities are generated by the decoder of the learned VaDE. Forth such new entities are examined, and relevant new conditions will be fed to the optimization. In this paper, this framework is implemented for and applied to the conceptual design problem of bridge structures. This application demonstrates that the framework can identify a void over the latent space and generate a possibility of a new concept. This paper is concluded with some discussion on the promises and possibilities of the proposed framework.

Presenting Author: Kikuo Fujita Osaka University

Authors:

Kikuo Fujita Osaka University
Kazuki Minowa Osaka University
Yutaka Nomaguchi Osaka University
Shintaro Yamasaki Osaka University
Kentaro Yaji Osaka University