Multiresolution Hypergraph Neural Network for Intelligent Fault Diagnosis

Intelligent fault diagnosis has made significant progress, thanks to machine learning, particularly deep-learning algorithms. However, most machine-learning algorithms treat samples as independent and ignore the correlations between samples that contain valuable information for creating discriminati...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement 2022, Vol.71, p.1-10
Main Authors: Yan, Xunshi, Liu, Yang, Zhang, Chen-An
Format: Article
Language:English
Subjects:
Algorithms
Convolution
Correlation
Deep learning
Fault diagnosis
graph convolutional network (GCN)
Graph neural networks
Graph theory
hypergraph
hypergraph neural network (HGNN)
Laplace equations
Machine learning
Machine learning algorithms
Machinery
multiresolution
Neural networks
Task analysis
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Intelligent fault diagnosis has made significant progress, thanks to machine learning, particularly deep-learning algorithms. However, most machine-learning algorithms treat samples as independent and ignore the correlations between samples that contain valuable information for creating discriminative features. In recent years, graph neural networks have increased diagnostic performance by capturing the correlation between samples according to establishing the inherent structure of data, but they also suffer from two shortcomings. First, a simple graph only represents pairwise relationships of samples and cannot depict complex higher-order structures. Second, the generated graph structure is insufficient to characterize the data without an explicit structure. To address the above two issues, this article proposes a multiresolution hypergraph neural network, a novel algorithm that can discover higher-order complex relationships between samples, and mine the structure hidden in data by establishing and fusing hypergraph structures at multiple resolutions. Experiments are conducted on three datasets to demonstrate the effectiveness of the proposed algorithm.
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2022.3212532