Research on the physical layer diagnosis of an Ethernet-based train communication network

Abstract Ethernet technology is widely applied in train communication networks (TCNs), serving as a crucial foundation for the enhancement of train intelligence. However, with its extensive deployment, some reliability issues have been exposed, particularly those at the physical layer. Certain faults have significantly impacted the daily operations and services of trains. Focusing on the diagnosis of the health status of the Ethernet physical layer, this paper proposes a window-voting Support Vector Machine (SVM) classification method based on multi-feature fusion. It aims to identify various fault conditions in TCNs and to detect potential communication issues in advance. Initially, the specific problems in TCNs are analysed, examining data waveform characteristics under four health statuses: normal, interference, aging and fault. Subsequently, the weights of the waveform features are calculated using the Fuzzy Analytic Hierarchy Process and the Grey Relational Analysis method, and a window-voting SVM classifier is then constructed to categorize the data waveforms. Finally, a test system is set up in the laboratory to simulate different health statuses of the Ethernet physical layer, and to acquire experimental data for validating the effectiveness of the proposed method. The results show that the accuracy of recognizing the health status of the Ethernet physical layer exceeds $95\%$.