摘要详情

ID / 提交时间

287 / 2025-06-16 11:13:28

标题

Fault Diagnosis Method for Harmonic Reducer Based on Federated Prototype Domain Adaptation

关键字

harmonic reducer, fault diagnosis, federated learning, domain adaptation, prototype

主题及专题

6、Artificial Intelligence for EOM

状态

全文待审

作者

Yulin Sun / Harbin University of Science and Technology

Shouqiang Kang / Harbin University of Science and Technology

Hongyuan Zhu / Heilongjiang Polytechnic

Yujing Wang / Harbin University of Science and Technology

Wenmin Lv / Harbin University of Science and Technology

摘要

To tackle the issue of significant variations in harmonic reducer data distribution across multiple users under different operating conditions, as well as the sparsity of labeled data under certain working conditions, a federated prototype domain adaptation approach is proposed for harmonic reducer fault diagnosis. Class prototypes are employed as carriers for feature alignment, and a prototype interaction mechanism is proposed in this method. It ensures efficient knowledge transfer across all users while preserving user data privacy. This enhances the model diagnostic performance across multiple operating conditions. Additionally, a prototype-guided training strategy is designed to optimize user local model training. Finally, multiple comparative experiments are conducted on a self-built experimental platform using harmonic reducer data. The results illustrate that the federated prototype domain adaptation improves diagnostic accuracy on target domain by at least 6.81% compared to existing federated domain adaptation methods.