EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Multiscale Multimodal Deep Federated Learning Framework for Collaborative Fault Diagnosis of Distributed Rotating Machinery in Industrial IoT

Peihan Guo, Weipeng Fan, Qiuyan Zhang

Shock and Vibration · 2026-01 · 2026 (1)

DOI: 10.1155/vib/5533758 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Shock and Vibration, online erschienen am 18.07.2026. Verfasst von Peihan Guo und 2 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Undetected or inaccurately diagnosed faults in distributed rotating machinery can gradually develop into serious safety hazards in Industrial IoT environments. Most existing data‐driven fault diagnosis methods rely on centralized training with sufficient labeled data. However, monitoring data collected by geographically distributed devices are often isolated and cannot be freely shared due to privacy concerns and commercial restrictions. To address this issue, we propose a federated multiscale attention network (MSAN)–based framework, termed Deep Federated Learning for Rotating Machinery (DeepFedRM), for multimodal fault diagnosis of distributed rotating machinery. At its core, DeepFedRM employs MSAN as the federated local backbone to enhance local multimodal representation learning for cross‐client collaboration. The proposed network adopts a two‐stage multimodal learning strategy, which first performs intramodal feature refinement and then conducts cross‐modal fusion. For each sensing modality, a modality‐aware multiscale feature extraction module with heterogeneous convolution kernels is introduced to capture fault‐related patterns under different temporal receptive fields. In addition, a channel attention mechanism is employed to enhance discriminative fault‐related responses while suppressing background noise. Experiments on a motor drive‐end bearing testbed dataset with eight representative health conditions show that DeepFedRM achieves near‐centralized diagnostic performance under independent and identically distributed (IID) federated settings, with an accuracy of 0.9951 versus 0.9953 for centralized learning, while still maintaining 0.9526 under Non‐IID conditions and 0.9518 under severe noise at SNR = 0.1 dB. Moreover, under the centralized setting, the proposed MSAN reaches an accuracy of 0.9957, outperforming MSCNN and MK‐ResCNN.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Guo, P., Fan, W., & Zhang, Q. (2026). A Multiscale Multimodal Deep Federated Learning Framework for Collaborative Fault Diagnosis of Distributed Rotating Machinery in Industrial IoT. Shock and Vibration, 2026(1). https://doi.org/10.1155/vib/5533758

Vancouver

Guo P, Fan W, Zhang Q. A Multiscale Multimodal Deep Federated Learning Framework for Collaborative Fault Diagnosis of Distributed Rotating Machinery in Industrial IoT. Shock and Vibration. 2026;2026(1). doi:10.1155/vib/5533758

BibTeX
@article{Guo2026,
  author = {Guo, Peihan and Fan, Weipeng and Zhang, Qiuyan},
  title = {A Multiscale Multimodal Deep Federated Learning Framework for Collaborative Fault Diagnosis of Distributed Rotating Machinery in Industrial IoT},
  journal = {Shock and Vibration},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/vib/5533758}
}
Kostenlos starten →