Wiring Network Fault Diagnosis Based on Time-Domain Reflectometry and Gramian Angular Field Encoding with Residual Neural Networks

This paper introduces a novel fault diagnosis framework for wiring networks that integrates Time Domain Reflectometry (TDR) with Gramian Angular Field (GAF) representations and a deep residual neural network. The proposed methodology transforms TDR responses into GAF images, which are directly exploited by the residual neural network to enable robust feature extraction from complex reflectometry signals. To support supervised learning, a forward modeling strategy is employed to generate representative TDR responses under a wide range of fault scenarios. Theframeworkis designed to provide real-time fault detection, localization, and characterization, demonstrating high effectiveness on complex topologies such as the YY-shaped network. Numerical results demonstrate high diagnostic performance for hard faults, achieving an overall accuracy and macro-averaged sensitivity exceeding 99%, thereby highlighting the effectiveness and reliability of the proposed approach.