Session: DAC-04-01-Data-Driven Design

Paper Number: 88548

88548 - Classification-Directed Conceptual Structure Design Based on Topology Optimization, Deep Clustering, and Logistic Regression 

The structure design of mechanical parts and components has been a major theme and a benchmark of design automation. Today, topology optimization techniques have become applied to various engineering applications under critical situations. As each topology optimization technique can effectively generate the optimal geometry in detail under the given design conditions and parameter settings, the paradigm has become recognized to be relevant for the embodiment design of structures. However, such an optimal geometry must be restricted within a certain type through the design conditions, parameters, and optimization setting. When viewing structure design under conceptual design, the identification of the optimal type is essentially important rather than any precious details of structures. The type means here configuration, morphology, form, or shape rather than geometry. It is a matter of conceptual design. This paper proposes a method for conceptual structure design for overcoming the above shortcoming. The method considers that conceptual design is a result of the classification of potentially possible geometries into types and identification of the best appropriate type from those types. In its details, a topology optimization technique generates diverse optimal geometries under various settings of conditions and parameters, a deep clustering technique, i.e., the variational deep embedding, clusters them into several categories, a logistic regression technique retrieves the criteria that distinct respective categories one by one. Further, the interaction with a designer identifies the relevant set of criteria  that can potentially lead the optimal geometries against the design requirement with simultaneously revealing and refining those criteria. This paper finally applies the proposed method to a simple bridge design problem for demonstrating its validity and associated possibilities.

Presenting Author: Ryo Tsumoto Osaka University

Presenting Author Biography: He is now a master course student of Osaka University

Authors:

Ryo Tsumoto Osaka University
Kikuo Fujita Osaka University
Yutaka Nomaguchi Osaka University
Shintaro Yamasaki Osaka University
Kentaro Yaji Osaka University