两级混淆对抗域自适应网络轴承故障诊断

Bearing fault diagnosis based on two-level confusion adversarial domain adaptation network

在变工况轴承故障诊断任务中,领域自适应方法仅仅对两个域进行全局对齐,而未进行相应类别的对齐,为解决上述问题,提出了一种两级混淆对抗域自适应网络.该网络由一个特征生成器、两个标签分类器及一个辅助分类器组成.网络使用源域样本帮助两个任务分类器学习,同时在辅助分类器上构造了基于两级域混淆损失的对抗性学习目标函数,通过对抗训练,驱动特征生成器生成类别对齐的特征.两个公共轴承数据集的实验结果表明,该方法的平均诊断准确率远远高于传统深度学习算法和其它四种域自适应算法.

In the variable condition bearing fault diagnosis task,the domain adaptation method only aligns two domains globally without considering category level alignment.To solve the above problem,a two-level confusion adversarial domain adaptation network is proposed.The network consists of a feature generator,two task classifiers,and an auxiliary classifier.We use source domain samples to assist two task classifier learning.Meanwhile,a novel adversarial learning objective based on twolevel domain confusion strategy is implemented on the auxiliary classifier.Through adversarial training,the features generated by the feature generator can align at category point.The results on two bearing datasets show that the diagnostic accuracy of our method is higher than the traditional deep learning algorithm and the other four domain adaptation algorithms.