基于子域自适应对抗网络的轴承故障诊断

Bearing fault diagnosis based on subdomain adaptive confrontation network

现有基于深度学习网络模型的故障诊断方法往往依赖大量有标签数据进行训练,在变工况条件下,模型的诊断精度会有所下降。针对此,为提高变工况条件下的故障诊断准确率,基于域自适应理论提出一种新颖的网络模型——子域自适应对抗网络。该网络模型不仅充分利用了动态卷积的特征提取能力,同时还借鉴了生成对抗网络的博弈思想,使特征生成器和分类器对抗学习,利用每个类别的决策边界对样本进行正确分类;此外,在对抗网络中引入局部最大平均差异,考虑每个类别的细粒度信息,以此来对齐源域和目标域相应的类空间,减小网络模型在决策边界附近的分类误差,从而提高模型对故障类别的识别精度。最终,通过两个数据集对所提出的方法进行试验验证,结果表明模型在变工况条件下具有较强的泛化性能与良好的故障识别精度。

Existing fault diagnosis methods based on deep learning network model often rely on a large number of labeled data for training,and the diagnosis accuracy of the model can decline under variable working conditions.Here,to improve the accuracy of fault diagnosis under variable working conditions,a novel network model called sub-domain adaptive confrontation network was proposed based on the domain adaptive theory.This network model could not only make full use of the feature extraction ability of dynamic convolution,but also draw lessons from the game idea of generating confrontation network to make the feature generator and classifier learn against each other,and correctly classify samples by using the decision boundary of each category.In addition,the local maximum average difference was introduced into the confrontation network,and the fine-grained information of each category was considered to align corresponding class spaces of the source domain and the target domain,reduce the classification error of the network model near the decision boundary,and improve the recognition accuracy of fault categories.Finally,two data sets were used to verify the proposed method.The results showed that the proposed model has stronger generalization performance and good fault identification accuracy under variable working conditions.