An Explainable Graph Transformer for Root-Cause Localization in Complex Industrial Networks
American Journal of Machine Learning · 2025
DOI: 10.71465/ajml3460 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Machine Learning, 2025. Verfasst von Qiang Li. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
The rapid proliferation of microservices architectures and Industrial Internet of Things (IIoT) devices has led to an exponential increase in the complexity of modern industrial networks. While this interconnectivity enhances operational efficiency and scalability, it simultaneously complicates the task of anomaly detection and root-cause localization (RCL). When a failure occurs, it often propagates through the network dependencies, generating a cascade of alarms that obscures the true origin of the fault. Existing data-driven approaches, particularly those based on deep learning, have demonstrated superior performance in fault detection but frequently suffer from a lack of interpretability. This "black box" nature hinders trust and delays human intervention in critical infrastructure. This paper proposes a novel Explainable Graph Transformer (XGT) framework specifically designed for root-cause localization in complex industrial topologies. By leveraging a spatial-temporal graph attention mechanism, the proposed model captures both the structural dependencies between network nodes and the temporal evolution of metric data. Crucially, we introduce a causal attention regularization term that forces the model to prioritize edges representing causal failure propagation paths. Extensive experiments on both synthetic and real-world datasets demonstrate that our approach not only outperforms state-of-the-art baselines in localization accuracy but also provides intuitive visualizations of fault propagation, significantly reducing the mean time to repair (MTTR).
Zitieren
APA
Li, Q. (2025). An Explainable Graph Transformer for Root-Cause Localization in Complex Industrial Networks. American Journal of Machine Learning. https://doi.org/10.71465/ajml3460
Vancouver
Li Q. An Explainable Graph Transformer for Root-Cause Localization in Complex Industrial Networks. American Journal of Machine Learning. 2025. doi:10.71465/ajml3460
BibTeX
@article{Li2025,
author = {Li, Qiang},
title = {An Explainable Graph Transformer for Root-Cause Localization in Complex Industrial Networks},
journal = {American Journal of Machine Learning},
year = {2025},
doi = {10.71465/ajml3460}
}
EUVIMED