Complex Wired Network Fault Diagnosis Based on Distributed Reflectometry and Multi-Channel 1D-CNN

This article proposed a wired network fault diagnosis method based on 1-D convolutional neural network (1D-CNN) and distributed reflectometer. This article improved the distributed reflectometer technique by providing a universal reflectometer arrangement scheme, which effectively addresses the localization ambiguity caused by multibranch in wired networks and the signal attenuation issue of reflection signals. Compared to traditional data-driven methods, this article used a multi-channel 1D-CNN model to learn fault data, leading to an improvement in fault diagnosis accuracy. To address the common issue of existing methods relying on simulation datasets, this article designed an automated wired network fault introduction and collection system, which collects real fault data from RG58 coaxial cables, two-core RVV cables, and multicore RVV cables. Also, two simulation datasets for comparison were generated using LTSPICE, and the similarities and differences between the simulated fault data and real fault data were compared. Through testing on these five datasets, the effectiveness of the proposed method was verified. Finally, a transfer learning method based on fine-tuning was proposed, which achieves a high fault diagnosis accuracy while reducing the sample size of real datasets.