Machine Learning for an Articulated Design Space

Research Title: Application of an Automatic Shape Clustering Method into Generative and Design Optimization Systems

Abstract: Despite their prevalence and extensive applications, generative and design optimization systems lack effective organizational methods of the excessive number of design options they produce, which is problematic for designers’ interaction. Ideally, a diverse and organized set of designs can mediate successful designers’ evaluation and exploration of the design space. Cluster analysis, a big-data management strategy, offers a solution. Yet, there is a need for investigating appropriate methods for applying cluster-analysis to a dataset of geometric shapes. Therefore, we have recently developed and published a new approach, the Shape Clustering using K-Medoids (SC-KM) method as an articulation mechanism in generative systems. The method involves shape description, shape difference measure calculation, and implementation of the K-Medoids clustering algorithm. The focus of this work is on incorporating the method into a generative system with parametric building shape generation and design optimization. The method organizes a dataset of shapes into clusters where shapes within the cluster share similarities yet differ from other clusters, and each cluster is signified by one representative shape. The SC-KM method contributes to an organized design presentation and facilitates designers’ examination of their designs’ geometric qualities.

%d bloggers like this: