EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

To Quantize or Not to Quantize: Effects on Generative Models for Topology Optimization Problems

Arthur Drake, Qiuyi Chen, Jun Wang, Ardalan Nejat, James K. Guest, Mark Fuge

Journal of Mechanical Design · 01.10.2026 · 148 (10)

DOI: 10.1115/1.4071440 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Mechanical Design, online erschienen am 22.04.2026. Verfasst von Arthur Drake und 5 weiteren Autor:innen. Die Arbeit nennt 77 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Advanced Research Projects Agency - Energy, National Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In topology optimization (TO) and related engineering applications, physics-constrained simulations are often used to optimize candidate designs given some set of design conditions. Such models can be computationally expensive and do not guarantee convergence to a desired result, given the frequent nonconvexity of design optimization problems. Creating data-driven approaches to warm-start these models—or even replace them entirely—has thus been an active area for engineering design researchers. In this article, we propose an augmented vector-quantized GAN (VQGAN) that allows for effective compression of TO designs within a discrete latent space, known as a codebook, while preserving high reconstruction quality. Further, we present a new dataset of two-dimensional heat sink designs optimized via multiphysics topology optimization (MTO) to train and evaluate this model. To concretely assess the benefits of the VQGAN quantization process, we conduct a latent analysis of its codebook as compared to the continuous latent space of a deep autoencoder (AE). We find that VQGAN can more effectively learn topological connections despite a high rate of data compression. Finally, we leverage the VQGAN codebook to train a small GPT-2 model that is capable of rapid generation of thermally performant heat sink designs. We show that the transformer-based approach is more effective than using a Wasserstein GAN with gradient penalty (WGAN-GP) due to its superior preservation of topological connections in MTO and similar applications.

Förderung

  • Advanced Research Projects Agency - Energy
  • National Science Foundation

Zitieren

APA

Drake, A., Chen, Q., Wang, J., Nejat, A., Guest, J. K., & Fuge, M. (2026). To Quantize or Not to Quantize: Effects on Generative Models for Topology Optimization Problems. Journal of Mechanical Design, 148(10). https://doi.org/10.1115/1.4071440

Vancouver

Drake A, Chen Q, Wang J, Nejat A, Guest JK, Fuge M. To Quantize or Not to Quantize: Effects on Generative Models for Topology Optimization Problems. Journal of Mechanical Design. 2026;148(10). doi:10.1115/1.4071440

BibTeX
@article{Drake2026,
  author = {Drake, Arthur and Chen, Qiuyi and Wang, Jun and Nejat, Ardalan and Guest, James K. and Fuge, Mark},
  title = {To Quantize or Not to Quantize: Effects on Generative Models for Topology Optimization Problems},
  journal = {Journal of Mechanical Design},
  year = {2026},
  volume = {148},
  number = {10},
  doi = {10.1115/1.4071440}
}
Kostenlos starten →