Clustering design structure matrices: A comparison of methods using minimum description length

Understanding interactions between components is fundamental in the design of products. Design Structure Matrices (DSMs) are often used to represent the relationships between every component or subsystem in a product. The complex network of interactions can then be clustered into subassemblies and other hierarchies, aiding designers in making critical decisions that will impact assembly, maintenance, and end-of-life disposal. This paper explores three methods for clustering components in a DSM to create a modular product architecture: (1) genetic algorithm, (2) hierarchical clustering, and (3) divisive clustering using a graph. A discussion on each algorithm is followed by an industrial example. This paper leads to the conclusion that genetic algorithm is better at identifying complex structures like bus module, 3D structure and overlapping cluster whereas hierarchical and divisive clustering are computationally inexpensive and are able to find optimal DSMs faster than the genetic algorithm.