A Fault Diagnosis Method Based on Graph Convolutional Attention Networks for Complex Coupled Systems

Large-scale units, transmission lines, and distributed integrated equipment in Industry 5.0 exhibit complex characteristics, including multiple components, extensive sensor networks, and strong signal coupling. Conventional fault diagnosis methods typically rely exclusively on time-series and time-frequency features, overlooking the physical topological relationships among devices. This limitation leads to challenges in identifying complex faults, reduced accuracy in fault tracing, and poor interference resistance. To overcome these issues, this paper introduces a combined graph convolutional attention-based fault diagnosis method (CGCA), which synergizes Graph Convolutional Networks (GCN) with Graph Attention Networks (GAT). Initially, the industrial physical system is modeled as a weighted topological graph, with components and sensors acting as nodes. By constructing a weighted adjacency matrix based on equipment connections and signal coupling strengths, prior knowledge of the system topology is effectively embedded. Subsequently, a graph convolutional network aggregates global node coupling features, which are further processed by a multi-head graph attention network to adaptively assign node weights. This mechanism not only enhances the representation of fault source features but also enables simultaneous fault classification and localization. Extensive comparative experiments and noise robustness tests are performed on the CWRU bearing dataset and the PU gearbox dataset. The results indicate that, relative to benchmark algorithms, the proposed method delivers superior performance in fault classification and source localization under typical industrial operating conditions, consistently surpassing reference models in both classification and localization accuracy. Quantitatively, on the CWRU bearing dataset, CGCA reaches 98.6% classification accuracy, 98.2% F1-score, and 96.9%/99.1% Top-1/Top-2 fault-source localization; on the PU gearbox dataset it reaches 97.8% accuracy and 95.8%/98.7% Top-1/Top-2 localization and retains 82.5% accuracy at 0 dB noise.

Authors

Institutions

Publication Details

Journal
Algorithms
Published
2026-11-19
DOI
https://doi.org/10.3390/a19111018
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Fault Diagnosis Method Based on Graph Convolutional Attention Networks for Complex Coupled Systems

Xuguang Chai, Ying Li, Zetong Sun, Xiaojing Cao et al.
Algorithms
Machine Fault Diagnosis Techniques
article

A Fault Diagnosis Method Based on Graph Convolutional Attention Networks for Complex Coupled Systems

Xuguang Chai, Ying Li, Zetong Sun, Xiaojing Cao, Kaihe You
article en

Abstract

Large-scale units, transmission lines, and distributed integrated equipment in Industry 5.0 exhibit complex characteristics, including multiple components, extensive sensor networks, and strong signal coupling. Conventional fault diagnosis methods typically rely exclusively on time-series and time-frequency features, overlooking the physical topological relationships among devices. This limitation leads to challenges in identifying complex faults, reduced accuracy in fault tracing, and poor interference resistance. To overcome these issues, this paper introduces a combined graph convolutional attention-based fault diagnosis method (CGCA), which synergizes Graph Convolutional Networks (GCN) with Graph Attention Networks (GAT). Initially, the industrial physical system is modeled as a weighted topological graph, with components and sensors acting as nodes. By constructing a weighted adjacency matrix based on equipment connections and signal coupling strengths, prior knowledge of the system topology is effectively embedded. Subsequently, a graph convolutional network aggregates global node coupling features, which are further processed by a multi-head graph attention network to adaptively assign node weights. This mechanism not only enhances the representation of fault source features but also enables simultaneous fault classification and localization. Extensive comparative experiments and noise robustness tests are performed on the CWRU bearing dataset and the PU gearbox dataset. The results indicate that, relative to benchmark algorithms, the proposed method delivers superior performance in fault classification and source localization under typical industrial operating conditions, consistently surpassing reference models in both classification and localization accuracy. Quantitatively, on the CWRU bearing dataset, CGCA reaches 98.6% classification accuracy, 98.2% F1-score, and 96.9%/99.1% Top-1/Top-2 fault-source localization; on the PU gearbox dataset it reaches 97.8% accuracy and 95.8%/98.7% Top-1/Top-2 localization and retains 82.5% accuracy at 0 dB noise.

AlgorithmsVol. 19(11)
Jinan University (CN), Guangdong University Of Finances and Economics (CN), Guangdong University of Finance (CN), Hebei Vocational University of Technology and Engineering (CN)
Machine Fault Diagnosis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon