Grounding fault localization in urban rail transit via enhanced multi-source domain adaptation under varying conditions

Abstract

With the prolonged operation of metro systems, the insulation performance of running rails in direct current traction power supply systems gradually degrades, leading to potential grounding faults. These faults increase rail potential (RP) and stray current, causing electrochemical corrosion and safety risks. To achieve accurate grounding fault localization under varying conditions, this paper proposes an enhanced multi-source domain adaptive framework (EMDAF). First, a dynamic RP computation platform and a laboratory-scale return system testbed are developed based on the overall topology of the urban rail transit system to simulate both normal and faulty insulation conditions. Then, an EMDAF is developed, consisting of a ResNet-based feature extractor, a multi-domain distribution alignment module, and a classifier discrepancy alignment module. The proposed model enables effective knowledge transfer among different domains and accurate fault identification across diverse operating conditions. Experimental results demonstrate that EMDAF outperforms conventional single-source and domain adaptation methods in transferability and diagnostic accuracy, confirming its robustness and practical applicability in urban rail grounding fault localization.