变工况下动态卷积域对抗图神经网络故障诊断

Fault diagnosis of dynamic convolutional domain adversarial graph neural networks under variable working conditions

针对基于无监督域自适应故障诊断方法忽略了域间数据结构信息、传统域对齐平均最大差异法全局泛化能力差等问题,本文提出一种基于无监督域自适应理论的动态卷积域对抗图神经网络故障诊断模型,分别通过对数据的类别标签、域标签和数据结构信息进行建模。通过分类器和域鉴别器分别建模类别标签和域标签,通过图神经网络将数据结构信息嵌入到实例图节点中,利用高斯Wasserstein距离来度量不同领域的实例图之间的差异。本文对比了不同工况下共14种迁移任务在各模型下故障识别的准确率。实验结果表明:基于动态卷积的域对抗图神经网络模型在变工况下的故障诊断效果均优于其他对比模型,且模型性能稳定。

This paper proposes a dynamic convolutional domain adversarial graph neural network fault diagnosis model based on unsupervised domain adaptation theory.Traditional domain alignment methods,such as the average maximum difference method,often ignore inter-domain data structure information and exhibit poor global generalization ability.This model addresses these issues by separately modeling the category labels,domain labels,and data structure information.The approach embeds data structure information into instance graph nodes using a graph neural network.Category labels and domain labels are modeled using classifiers and domain discriminators,respectively.The Gaussian Wasserstein distance is used to measure the differences between instance graphs in different domains.This paper compares the accuracy of fault identification across 14 migration tasks using various models and under different operating conditions.Experimental results show that the domain adversarial graph neural network model,based on dynamic convolution,outperforms other comparative models in fault diagnosis under varying operating conditions and maintains stable performance.