Enhancing Gas Turbine Fault Diagnosis Using a Multi-Scale Dilated Graph Variational Autoencoder Model

This paper proposes a Multi-scale Dilated Variational Graph Convolutional Autoencoder (MG-VAE) model for gas turbine fault diagnosis. The model integrates a multi-scale dilated convolutional attention mechanism to extract features across different scales, enhancing its ability to represent complex d...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE access Ročník 12; s. 104818 - 104832
Hlavní autoři: Kun, Zhang, Hongren, Li, Xin, Wang, Daxing, Xie, Shuai, Yang
Médium: Journal Article
Jazyk:angličtina
Vydáno: Piscataway IEEE 2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Ablation
Accuracy
Artificial neural networks
Convolutional neural networks
Data mining
Data models
Fault diagnosis
Feature extraction
Gas turbine
Gas turbines
Industrial applications
MG-VAE
Monitoring
noise robustness
Real gases
Turbines
ISSN:2169-3536, 2169-3536
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:This paper proposes a Multi-scale Dilated Variational Graph Convolutional Autoencoder (MG-VAE) model for gas turbine fault diagnosis. The model integrates a multi-scale dilated convolutional attention mechanism to extract features across different scales, enhancing its ability to represent complex data and improving robustness in noisy environments. Additionally, a graph convolution module captures correlations between sensors, further enhancing diagnostic accuracy. Experimental results demonstrate the model's effectiveness, achieving high diagnostic accuracy in both gear fault simulation and real gas turbine fault datasets. Ablation experiments show that the integration of the graph convolutional network and the multi-scale dilated convolutional attention mechanism significantly improves accuracy, highlighting the model's potential for practical industrial applications in gas turbine fault diagnosis.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3434708