针对滚动轴承在变工况环境中网络特征提取能力不足的问题,提出了一种域对抗图卷积注意力迁移学习的故障诊断方法(DAGRESL).首先,通过残差神经网络(residual network,Resnet)提取输入的轴承故障信息特征并通过Simam注意力模块增强Resnet的特征表达能力;其次,利用图生成层学习Resnet的特征数据并挖掘样本结构特征之间的关系来构造实例图;然后,利用图卷积网络(graph convolutional network,GCN)对实例图进行建模;最后,利用域判别器和局部最大平均差异(local maximum mean discrepancy,LMMD)对齐子域和全局域之间的分布并通过标签分类网络完成故障分类.通过在SQI-MFS轴承数据集的实验结果证明了所提出的DAGRESL模型能够精准地区分变工况轴承故障类型,有效解决了滚动轴承在变工况环境中网络特征提取能力不足的问题.
Domain Adversarial Graph Convolutional Attention Network for Fault Diagnosis Under Variable Working Conditions
In order to solve the problem that the network feature extraction ability of rolling bearings is in-sufficient in variable working conditions,a fault diagnosis method(DAGRESL)based on domain antago-nism graph convolutional attention Transfer learning is proposed.Firstly,the residual neural network(Res-net)is used to extract the input bearing fault information features,and the Simam attention module is used to enhance the feature expression ability of Resnet;Secondly,use the graph generation layer to learn Resnet's feature data and mine the relationships between sample structure features to construct an instance graph;Then,model the instance graph using graph convolutional network(GCN);Finally,the domain discrimina-tor and local maximum mean difference(LMMD)are used to distribute the distribution between the homo-geneous sub domain and the global domain,and the fault classification is completed through the label Class-ful network.The experimental results on the SQI-MFS bearing dataset demonstrate that the proposed DAGRESL model can accurately distinguish the types of bearing faults under variable operating conditions,effectively solving the problem of insufficient network feature extraction ability for rolling bearings in varia-ble operating conditions.
fault diagnosisvariable working conditionsconvolutional attention modulegraph convolu-tional
NETL
NSTL
万方数据
