Trade-offs Among System Architecture Modularity Criteria
DS 85-1: Proceedings of NordDesign 2016, Volume 1, Trondheim, Norway, 10th - 12th August 2016
Year: 2016
Editor: Boks, Casper; Sigurjonsson, Johannes; Steinert, Martin; Vis, Carlijn; Wulvik, Andreas
Author: Sanaei, Roozbeh; Otto, Kevin; Wood, Kristin; Hölttä-Otto, Katja
Series: NordDESIGN
Institution: 1: Aalto University, Finland; 2: Singapore University of Technology and Design, Singapore
Section: Product Architectures & Modularity
Page(s): 352-359
ISBN: 978-1-904670-80-3
Abstract
In last decade, there has been considerable research in product modularity, measuring the level of modularity and various procedures for searching for ideal modular architectures. Different manual heuristics and computer clustering algorithms have been developed to search more ideal modular architectures by optimizing a modularity metric. However, the different criteria can be in conflict and improving one criteria may drive another infeasible without a compromising effect on another. We pose here the research question as to how to visualize architectural design criteria and trade-offs in the early conceptual design phase. We analyze correlations between different system architecture modularity criteria provided in the research field, namely the intra-cluster, extra-cluster costs, number of modules and the variance in the size of modules. We demonstrate that these criteria trade-off with each other, and therefore one cannot be improved without affecting the other. We also show that several of these metrics are directly correlated, for example the variance in the size of modules can be controlled through the intra-cluster cost. Finally, we observe that although architectures that optimize extra-cluster cost for different number of modules can construct a hierarchy, this is an exception. Overall, architectures which optimize these modularity metrics are not hierarchical, and therefore well-sized modules cannot be obtained through typically proposed agglomerative or divisive clustering algorithms such as K-means clustering or similar.
Keywords: Modularity, Clustering algorithms, Trade-off analysis