EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Topology Regularized Contrastive Learning Diagnoses Multisensor Faults with Sparse Sensor Graphs

Svenja Vogel, Julien Rey, Tatsuya Hayashi

International Journal of Computational and Biological Sciences · 15.06.2026 · 3 (2) · 135-145

DOI: 10.66238/ijcbs92 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Computational and Biological Sciences, online erschienen am 15.06.2026. Verfasst von Svenja Vogel und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid expansion of the Industrial Internet of Things has facilitated the deployment of massive multisensor networks for equipment monitoring and fault diagnosis. However, practical industrial environments frequently suffer from sensor malfunctions, communication latency, and severe bandwidth limitations, resulting in highly sparse and asynchronous data streams. Traditional deep learning paradigms struggle to extract robust diagnostic features under such severe data sparsity, often leading to catastrophic degradation in diagnostic accuracy. To address this fundamental challenge, this paper proposes a novel framework based on topology-regularized contrastive learning specifically designed for sparse multisensor fault diagnosis streams. By leveraging the inherent spatial and functional topology of the sensor network, the proposed method constructs an inductive bias that guides the representation learning process even when temporal data points are missing. We introduce a contrastive learning objective that maximizes the agreement between augmented views of sensor data while simultaneously enforcing a topological penalty. This penalty ensures that sensors physically or functionally linked in the network maintain structurally consistent feature embeddings, effectively bridging the gaps caused by sparse sampling. Comprehensive experiments conducted on large-scale industrial datasets demonstrate that the proposed framework significantly outperforms conventional representation learning techniques, maintaining high diagnostic reliability even when a substantial percentage of the sensor stream is dropped or corrupted.

Zitieren

APA

Vogel, S., Rey, J., & Hayashi, T. (2026). Topology Regularized Contrastive Learning Diagnoses Multisensor Faults with Sparse Sensor Graphs. International Journal of Computational and Biological Sciences, 3(2), 135-145. https://doi.org/10.66238/ijcbs92

Vancouver

Vogel S, Rey J, Hayashi T. Topology Regularized Contrastive Learning Diagnoses Multisensor Faults with Sparse Sensor Graphs. International Journal of Computational and Biological Sciences. 2026;3(2):135-145. doi:10.66238/ijcbs92

BibTeX
@article{Vogel2026,
  author = {Vogel, Svenja and Rey, Julien and Hayashi, Tatsuya},
  title = {Topology Regularized Contrastive Learning Diagnoses Multisensor Faults with Sparse Sensor Graphs},
  journal = {International Journal of Computational and Biological Sciences},
  year = {2026},
  volume = {3},
  number = {2},
  pages = {135--145},
  doi = {10.66238/ijcbs92}
}
Kostenlos starten →