Shape Clustering Using K-Medoids in Architectural Form Finding

Abstract

As the number of design candidates in generative systems is often high, there is a need for an articulation mechanism that assists designers in exploring the generated design set. This research aims to condense the solution set yet enhance heterogeneity in generative design systems. Specifically, this work accomplishes the following: (1) introduces a new design articulation approach, a Shape Clustering using K-Medoids (SC-KM) method that is capable of grouping a dataset of shapes with similitude in one cluster and retrieving a representative for each cluster, and (2) incorporate the developed clustering method in architectural form finding. The articulated (condensed) set of shapes can be presented to designers to assist in their decision making. The research methods include formulating an algorithmic set with the implementation of K-Medoids and other algorithms. The results, visualized and discussed in the paper, show accurate clustering in comparison with the expected reference clustering sets.