An Effective Measure to Identify Meaningful Concepts in Engineering Design optimization

Abstract

Identifying similar solutions during an engineering design process and organizing the design data set into several concepts has substantial benefits. A concept is an abstract representation of design solutions that share comparable properties and behavior. Inspecting such concepts facilitates an increase of knowledge about the structure of the design problem. Concepts also allow for the selection of archetypal representatives which can be used as prototypes for further processing. Each prototype represents a different part of the design domain and can, for example, be effectively used to initialize the starting population of a design optimization leading to meaningful variations towards increased design performance. However, identifying meaningful concepts in a large engineering design data set and objectively quantifying the quality of the identified set of concepts is a challenging task. Existing measures to evaluate concepts of design solutions exhibit substantial drawbacks as they do not consider the simultaneous existence and interactions of multiple concepts on one design data set thoroughly. Therefore, we propose a new measure for objectively quantifying the quality of a set of concepts which explicitly takes the overlap and sizes of the concepts into account. We show the benefits of our measure by comparing it to state-of-the art measures in an automated optimization-based concept identification approach for a real-world inspired engineering design data set.