EUVIMEDΑνακαλύψτε το EUVIMED
Crossref

Graph Neural Networks for Fault Diagnostics in Cyber-Physical Systems: A Survey of Taxonomy, Deployment Architectures and Failure Modes

Vaibhavi Tiwari, Ola Suaifan, Ramy Othman, Anand Gupta

Journal of Engineering Research and Sciences · 2026-06 · 5 (6) · 67-96

DOI: 10.55708/js0506006 ↗

At a glance

Journal article in Journal of Engineering Research and Sciences, 2026-06. Written by Vaibhavi Tiwari and 3 co-authors. 79 references.

ΠερίληψηΒιβλιογραφικά στοιχείαCiteRelated studies

Abstract (original language)

Graph Neural Networks (GNNs) have emerged as a promising approach for fault diagnosis in complex cyber-physical systems because they can model intercomponent relationships, fault propagation, and system-level anomalies across domains such as industrial automation, smart grids, transportation, and healthcare. This survey presents a multidimensional review of GNN-based fault diagnostics, organizing existing methods according to graph representation, learning paradigm, diagnostic objective, and deployment context. It examines commonly used benchmark datasets, evaluation protocols, and cloud, edge, hybrid, and federated deployment architectures, with particular attention to reproducibility and practical implementation. In addition to methodological limitations, the survey identifies operational failure modes, including cascading misdiagnosis, topology drift, noise amplification, open-set misclassification, adversarial vulnerability, and concept drift, and examines their implications for safety-critical systems. Emerging research directions, including physics-informed learning, multimodal fusion, dynamic graph modeling, and privacy-preserving federated GNNs, are discussed alongside their ethical and safety implications. Unlike reviews centered primarily on diagnostic tasks or application domains, this survey integrates methodological, architectural, and operational perspectives within a unified framework. The analysis indicates that although GNNs offer capabilities for dependency-aware fault diagnosis, their practical deployment remains constrained by inconsistent benchmarking, sensitivity to graph construction, computational requirements, limited interpretability, and insufficient validation under evolving operational conditions. Finally, practitioneroriented design guidelines are presented to support the development of reliable, robust, and deployable GNN-based diagnostic systems and to connect algorithmic advances with the operational reliability requirements of next-generation fault-tolerant infrastructures.

Cite

APA

Tiwari, V., Suaifan, O., Othman, R., & Gupta, A. (2026). Graph Neural Networks for Fault Diagnostics in Cyber-Physical Systems: A Survey of Taxonomy, Deployment Architectures and Failure Modes. Journal of Engineering Research and Sciences, 5(6), 67-96. https://doi.org/10.55708/js0506006

Vancouver

Tiwari V, Suaifan O, Othman R, Gupta A. Graph Neural Networks for Fault Diagnostics in Cyber-Physical Systems: A Survey of Taxonomy, Deployment Architectures and Failure Modes. Journal of Engineering Research and Sciences. 2026;5(6):67-96. doi:10.55708/js0506006

BibTeX
@article{Tiwari2026,
  author = {Tiwari, Vaibhavi and Suaifan, Ola and Othman, Ramy and Gupta, Anand},
  title = {Graph Neural Networks for Fault Diagnostics in Cyber-Physical Systems: A Survey of Taxonomy, Deployment Architectures and Failure Modes},
  journal = {Journal of Engineering Research and Sciences},
  year = {2026},
  volume = {5},
  number = {6},
  pages = {67--96},
  doi = {10.55708/js0506006}
}
Start for free →