Visualization of Mechanical Kansei with deep learning approach

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K03824
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 18030:Design engineering-related
• Research Institution: Nara National College of Technology
• Principal Investigator: Hira Toshio 奈良工業高等専門学校, 機械工学科, 教授 (60280426)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 力学的感性 / 深層学習

Outline of Final Research Achievements

The capability to make unconscious and intuitive judgments for the mechanics aspect of an object is called Mechanical Kansei, and there are attempts to elucidate how this capability is utilized in the realization of structural rationality within the design process. In this research, we assume the ability to intuitively recall the force flow at the level that a novice learner acquires, and simulate this ability with a network obtained by deep learning. By considering the topology-optimized shapes, determined under various structural boundary conditions, as a representation of the force flow, these shapes were used to train a variational autoencoder. We showed that these shapes could be reconstructed in a low-dimensional latent variable space. We also confirmed that the trained network represents a mapping from structural boundary conditions to shapes without the mechanics model inside the network.

Free Research Field

設計工学

Academic Significance and Societal Importance of the Research Achievements

一般に，デザインという行為は設計者の審美眼的な感性的側面が強調される．その一方で機能美の概念のように，構造物にはその力学的な合理性が美しさとして表出するとの考え方も古くから論じられている．これらのことは，デザインにおける感性的側面と，力学的合理性のような客観的側面とが不可分であることを示している．本研究は，力学的感性の概念を手がかりに，その能力を深層学習によって再現することで，デザインに対して体系的なアプローチをしようとする試みの一部であり，得られた知見が人の創造的行為のさらなる理解につながるという意義を持つ．