Automatic Clustering of Sequential Design Behaviors

Abstract

Design is essentially a decision-making process, and systems design decisions are sequentially made. In-depth understanding on human sequential decision-making patterns in design helps discover useful design heuristics to improve existing algorithms of computational design. In this paper, we develop a framework for clustering designers with similar sequential design patterns. We adopt the Function-Behavior-Structure based design process model to characterize designers’ action sequence logged by computer-aided design (CAD) software as a sequence of design process stages. Such a sequence reflects designers’ thinking and sequential decision making during the design process. Then, the Markov chain is used to quantify the transitions between design stages from which various clustering methods can be applied. Three different clustering methods are tested, including the K-means clustering, the hierarchical clustering and the network-based clustering. A verification approach based on variation of information is developed to evaluate the effectiveness of each method and to identify the clusters of designers who show strong behavioral similarities. The framework is applied in a solar energy systems design problem — energy-plus home design. The case study shows that the proposed framework can successfully cluster designers and identify their sequential decision-making similarities and dissimilarities. Our framework can support the studies on the correlation between potential factors (e.g., designers’ demographics) and certain design behavioral patterns, as well as the correlation between behavioral patterns and design quality to identify beneficial design heuristics.