A Fault Diagnosis Model for Railway Track Circuits Using Deep Learning

The railway track circuit is critical for train detection and signaling safety. Conventional diagnosis methods, reliant on manual checks and threshold-based alarms, are inefficient and prone to errors. This paper presents a deep learning model for intelligent fault diagnosis in track circuits. Real-time voltage and current waveforms under various conditions—normal operation, shunt faults, broken rails, and insulation degradation—are collected to form a high-dimensional time-series dataset. A hybrid neural network combining 1D Convolutional Neural Networks and Long Short-Term Memory networks is designed to automatically extract spatiotemporal features from raw signals. The model performs end-to-end diagnosis, identifying both fault type and severity. Validated using field data from a heavy-haul railway, the model achieves an overall classification accuracy of 98.7%, surpassing traditional threshold and Support Vector Machine methods. Notably, it attains a 95.3% recall rate for early-stage insulation degradation, a fault notoriously difficult for conventional approaches to detect. An integrated attention mechanism enhances interpretability by highlighting signal segments most relevant to the fault. This research demonstrates that deep learning offers a viable pathway toward predictive and intelligent maintenance of track circuits, with significant potential to reduce unplanned downtime and strengthen the safety of railway signaling systems.