Remote Fault Diagnosis Method for Traction Drive Systems Based on Domain-Adapted Knowledge Graphs

被引：0

作者：

Chen, Zhiwen ^{[1
]}

Mo, Siwen ^{[1
]}

Xu, Jiamin ^{[1
]}

Cheng, Zhenglin ^{[2
]}

Peng, Tao ^{[1
]}

Gui, Weihua ^{[1
]}

机构：

[1] Central South University, School of Automation, Changsha,410012, China

[2] Zhuzhou CRRC Times Electric Company, Ltd., Zhuzhou,412001, China

来源：

IEEE Transactions on Industrial Cyber-Physical Systems | 2024年 / 2卷

关键词：

Knowledge graph;

D O I：

10.1109/TICPS.2024.3477433

中图分类号：

学科分类号：

摘要：

As a vital component of railway cyber-physical systems (RCPSs), efficient fault diagnosis of traction drive systems (TDSs) is crucial for ensuring the safe and stable operation of rail trains. Traditional methods for diagnosing faults remotely often depend heavily on expert experience and underutilize unstructured knowledge, resulting in inefficiencies. To address these challenges, this paper develops a novel method based on domain-adapted knowledge graphs. Firstly, we propose a domain-adapted segmentation technique that constructs a domain lexicon, significantly reducing the misclassification of technical terms compared to general lexicons. Secondly, to counter the lack of interaction in traditional information extraction, we implement a dual BiLSTM approach, which promotes feature sharing between entity and relationship extraction, thereby enhancing their interconnectivity. Experiments utilizing fault logs from a specific train model confirm the effectiveness of our method. Based on these findings, a remote fault diagnostic system has been developed, providing a new solution for maintenance decision-making. © 2024 IEEE.

引用

页码：556 / 564