Autoencoder-Multidimensional Taylor Network for Intelligent Fault Diagnosis Under Privacy Preservation

作     者：Yang, Li Wen, Chenglin Zhou, Zhe 

作者机构：Huzhou Univ Sch Engn Huzhou Key Lab Intelligent Sensing & Optimal Cont Huzhou 313000 Peoples R China Guangdong Univ Petrochem Technol Sch Automat Maoming 525000 Peoples R China 

出 版 物：《IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT》 (IEEE Trans. Instrum. Meas.)

年 卷 期：2024年第73卷

页      面：1页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0804[工学-仪器科学与技术] 

基　　金：National Natural Science Foundation of China [U22A2046, 62125307] Natural Science Foundation of Huzhou [2023YZ47] Huzhou Key Laboratory of Intelligent Sensing and Optimal Control for Industrial Systems [2022-17] 

主　　题：Fault diagnosis Federated learning Accuracy Mathematical models Feature extraction Encoding Servers Real-time systems Decoding Data models Autoencoder-multidimensional Taylor network (MTN) federation learning hybrid additive and multiplicative coding strategy intelligent fault diagnosis punishing eavesdroppers 

摘      要：Data-driven-based intelligent fault diagnosis has achieved significant success. However, the limited number of features and samples often hinders the accuracy of such diagnoses. On the one hand, addressing the issue of low accuracy in intelligent fault diagnosis due to limited features, and the additional complexity of intelligent fault diagnosis algorithms makes people worry about the rationality and interpretability of the decisions made by the model, benefiting from the Taylor series expansion theorem, we propose an intelligent fault diagnosis method based on an autoencoder-multidimensional Taylor network (MTN). On the other hand, to address the problem of low accuracy of intelligent fault diagnosis due to small samples, we introduce federated learning. In federated learning, multiclient data is pooled to improve the accuracy of intelligent fault diagnosis. However, the interaction information between the client and the federation center is easily stolen by eavesdroppers. For this reason, to protect the security of the information interaction between the federated centers and the clients, we propose a hybrid additive and multiplicative coding strategy based on punishing eavesdroppers. Finally, we validate the new method based on the actual flexible rotor and the open-bearing dataset of Western Reserve University. The experimental results show that the intelligent fault detection model proposed in this article can improve the accuracy of fault detection, while the proposed information protection strategy can realize the effect of punishing eavesdroppers. The proposed method in this article is further compared with the latest research methods to verify the advancement of the proposed method.