计算机科学
断层(地质)
域适应
适应(眼睛)
方位(导航)
人工神经网络
过程(计算)
学习迁移
领域(数学分析)
数据挖掘
人工智能
机器学习
实时计算
数学分析
数学
地震学
分类器(UML)
地质学
物理
光学
操作系统
作者
Naiming Qi,Yijiao Liu,Jianfei Yang,Qiufan Yuan,Kang Sun,Mingying Huo
出处
期刊:IEEE Access
[Institute of Electrical and Electronics Engineers]
日期:2024-01-01
卷期号:12: 93771-93780
被引量:3
标识
DOI:10.1109/access.2024.3424476
摘要
Polytropic working conditions are prioritized by most intelligent adaptive methods. However, there are certain limitations in the existing domainadaptive methods for fault diagnosis that warrant improvement. First, the privacy and the inaccessibility of the source data during the transfer learning process are not considered. Additionally, the single-use limitation of the labeled source data for fault detection requires improvement. Moreover, the attention weights for each time point in the fault signals are also not addressed. Second, the contributions of different channels to label prediction during neural network training are not differentiated regarding importance. Intending to deal with the problems above, we put forward a mixed attention network (MAN) for source-free domain adaptation (SFDA) in bearing fault diagnosis work. Specifically, we fully utilize the only-once source fault information to generate a source model with strong anomaly detection capabilities. Mixed attention-dense convolution is essential in extracting the in-depth features of fault signals. Subsequently, reasonable prediction of target domain labels maximizes the transfer of source domain information. Additionally, a pseudo-learning strategy reduces label calculation deviation due to domain migration. Moreover, we reconstruct the joint loss function of the migration training process under a specific combination strategy. Keywords: Transfer learning(TL), Fault diagnostic, Deep neural network(DNN), Unsupervised domain adaptation(UDA), Deep learning(DL)
科研通智能强力驱动
Strongly Powered by AbleSci AI