EUVIMEDEUVIMED entdecken
Crossref

Spectral Clustering for Large‐Scale Design Structure Matrices

Heekun Roh, Johannes J. Norheim, Olivier L. de Weck

Journal of Mechanical Design · 19.11.2026

DOI: 10.1002/jomd.70010 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Mechanical Design, online erschienen am 19.11.2026. Verfasst von Heekun Roh und 2 weiteren Autor:innen. Die Arbeit nennt 65 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper proposes applying a matrix spectra‐based method for clustering large‐scale Design Structure Matrices (DSMs), envisioned as a diagnostic tool for Extreme Designs (XDs). DSMs represent connectivity between system components, functions, or organizations, where clustering often reveals the system's modular and hierarchical structure. Clustering on DSMs—which also connects to the inherent “coupled‐ness” of the large‐scale engineering design—is especially important in managing the high complexity and uncertainty associated with XD. However, detecting the system cluster structure for complex systems is challenging due to the problem's inherent combinatorial and non‐unique nature. Prior methods targeted for DSM typically relied on randomized search (e.g., the widely‐used Idicula–Gutierrez–Thebeau Algorithm [IGTA]) or guaranteed optimality at the cost of high computational load and limited scalability (e.g., Mixed‐Integer Linear Programming [MILP]‐based methods). We illustrate that the spectral method, widely adapted in the image segmentation community, can also serve for clustering engineering—oriented DSMs, characterized by hierarchical structure and limited connections per parts. We demonstrate the spectral clustering method on DSM's normalized Laplacian, which corresponds to a relaxed maximization of the normalized modularity metric. Consequently, the suggested algorithm performs reasonably close to optimality while being computationally efficient on regular DSMs (15.1% loss in modularity with −99.978% reduction in computation time, on average against the optimal MILP method, for the jet engine DSM example) while demonstrating scalability handling up to 100,000 elements for synthetic, XD‐oriented DSMs. We compare the scalability and performance of this approach against established methods, including IGTA and MILP, to demonstrate its potential.

Zitieren

APA

Roh, H., Norheim, J. J., & Weck, O. L. D. (2026). Spectral Clustering for Large‐Scale Design Structure Matrices. Journal of Mechanical Design. https://doi.org/10.1002/jomd.70010

Vancouver

Roh H, Norheim JJ, Weck OLD. Spectral Clustering for Large‐Scale Design Structure Matrices. Journal of Mechanical Design. 2026. doi:10.1002/jomd.70010

BibTeX
@article{Roh2026,
  author = {Roh, Heekun and Norheim, Johannes J. and Weck, Olivier L. de},
  title = {Spectral Clustering for Large‐Scale Design Structure Matrices},
  journal = {Journal of Mechanical Design},
  year = {2026},
  doi = {10.1002/jomd.70010}
}
Kostenlos starten →