Cross-device fault diagnosis method based on graph convolution and multi-sensor fusion
SUN Yuanshuai
KONG Fanqin
NIE Xiaoyin
XIE Gang
Abstract：[Objective]To address the problems of difficulty in obtaining labeled fault data for mechanical equipment and low diagnosis accuracy caused by different probability distributions of cross-device data in actual production,a cross-device fault diagnosis method based on graph convolution and multi-sensor fusion,named convolutional domain graph convolution network(CDGCN),is proposed,realizing the unified modeling of class labels,domain labels,and data feature structures.[Methods]Firstly,a convolutional neural network(CNN)was utilized to extract preliminary features from raw signals.Secondly,an instance graph was constructed by mining the feature structural relations among samples through a graph generation layer,and a multi-receptive field graph convolutional network(MRF-GCN)was employed for modeling to extract more expressive node features.Meanwhile,a high-level feature fusion method was proposed to achieve multi-sensor information integration.Finally,let the maximum mean discrepancy(MMD)metric,the classifier and the domain discriminator work synergistically to achieve domain adaptation(DA)through a minimax game.[Results]Test results show that the average accuracy of CDGCN reaches 75.33%,which is improved by 29.23,30.35,15.20 and 12.70 percentage points compared with the domain-adversarial neural network(DANN),conditional domain adversarial network(CDAN),joint adaptation network(JAN),and deep adaptation network(DAN)method,respectively.Ablation test verifies the effectiveness of multi-receptive field feature extraction,data feature structure modeling,and multi-sensor information fusion in improving transfer diagnosis accuracy.
Keywords：Graph convolutional neural networkMulti-sensorCross-deviceDomain adaptationFault diagnosis
Publication Date：2026-02-15
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：10( 21-30 )