Domain Adaptation for Intelligent Fault Diagnosis from Relation Perspective. Issue 1 (1st March 2022)

• Record Type: Journal Article
• Authors: Tang, Lv; Zhang, Qing; Xuan, Jianping
• Is Part Of: Journal of physics. Volume 2184:Issue 1(2022)
• Publication Date: 2022-03-01
• Subjects: Bearing fault diagnosis -- Domain adaptation -- Variable working condition -- Knowledge transfer; Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/2184/1/012014
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: A novel domain adaptation technology for fault diagnosis is introduced to solve the variable working condition problem. As the relationships between things are widespread according to human visual and cognitive logic, the network also learns the relation and shows its sense to classify. Thence, transfer knowledge beyond the layer features are constructed to learn a relation invariant representation. The hierarchical relation and category relation are considered, while the connection between relation alignment and feature alignment is uncovered using exceptional cases. The hierarchical relation captures the behavior of information flow in adjacent layers, and the category relation captures the similarity between bearing faults. Empirical evidence provided by the experiment with variable speed test demonstrates that the proposed model outperforms the existing method.