Organizing a design space of disparate component topologies

Abstract

In a previous DCC paper, the authors presented an approach to generate a large space of conceptual designs by a set of grammar rules. These results indicated a large number of topologically unique solutions that could be created from a single black box that consists of a simple description of the function of the product and the input and output flows. The problem remains as to how to efficiently organize and search this space to find the best design for a given set of user preferences. In this paper we present new results that organize the candidate space using clustering methods such as K-means algorithm that group a large number of points in space based on a certain spatial property. Candidate component topologies, referred to as Component Flow Graphs (CFGs), are categorized using this method based on properties that physically distinguish them. From a theoretical and computational standpoint, this is an open research question as the CFGs may be vastly different graph topologies and the nodes and arcs of the graph may represent many different types of components and component connections. This paper details an experiment wherein ten products are designed from function structure to CFG. A space of over 8000 candidate solutions is developed. From this large set, clustering algorithms are employed to organize the space, and eventually aid an automated or interactive search algorithm that can find a best candidate solution for a particular user. A vast space of clustered concepts would allow an interactive process to query the user about particular CFGs and gauge whether the user would like to see more similar CFGs (i.e. from the same cluster) or is more interested in different ones (i.e. ones from other clusters). Such an interactive tool would be useful in mass customization.