基于条件对抗图卷积网络的轴承故障诊断方法

打开文本图片集

中图分类号：TH133.3 文献标识码：A 文章编号：2095-2953（2025）03-0060-09

Conditional Adversarial Graph Convolutional Network-Based Fault Diagnosis Method for Bearings'

ZHANG Han-qi， XIA Peng *，XIE Shao-zhang， FANG Yi-kai （CollgeofElectromechanicalEngineering，NortheastForestryUniversity，HarbinHeilongjiang15O40，China）

Abstract：Thereliabilityof bearings isof paramount importance for the eficient and safe operation offorestry machinery.In the fieldof unsupervised domainadaptation（UDA），the integrationof classlabels，domain labels，and data structures is crucial forestablishing a connection between the labeled sourcedomainand theunlabeled target domain.The majorityof existing methodologiesconcentrate solelyonthefirsttwoaspects，thereby neglecting the significance of data structure.This ultimatelyleads to the extraction of incomplete feature information bydeep networks.To address this issue，this paper proposes the Conditional Adversarial Graph Convolutional Network （CDAGCN），which models these three types of information and performs UDA in aunified framework.The firsttwo types of information are processed through the clasifiers and domain discriminators.In terms of data structure，the first step is to use a convolutional neural network（CNN）to accurately extract the features of the input signals.These are then fed intoa graph generation layer，where graphs are constructed through relationship mining and instance graph construction. Subsequently，the instance graphs are analyzed using graph convolutional networks，and the correlation alignment （CORAL）strategy is applied to estimate the structural differences of instance graphs in different domains.The synthesized strategy provides an innovative and eective approach to addressing real UDA problems.

Key words：bearing； fault diagnosis；graph convolutional network ；conditional adversarial

轴承在林业机械中扮演着至关重要的角色，包括伐木设备、运输车和加工机等。（剩余12795字）