Railway Track Circuit Fault Diagnosis Using Recurrent Neural Networks.

Area of Science:

• Railway engineering
• Artificial intelligence
• Signal processing

Background:

• Railway track circuit faults pose significant risks to network safety and operational continuity.
• Accurate and timely fault detection is essential for maintaining reliable railway systems.
• Existing methods may not fully capture the complex spatial and temporal dependencies in track circuit data.

Purpose of the Study:

• To propose and evaluate a Long-Short-Term Memory (LSTM) recurrent neural network for detecting and identifying faults in railway track circuits.
• To leverage spatial and temporal dependencies from multiple track circuit signals for improved fault diagnosis.
• To demonstrate the capability of the LSTM network to learn these complex data relationships.

Main Methods:

• Utilized Long-Short-Term Memory (LSTM) recurrent neural networks for fault detection and identification.
• Analyzed spatial and temporal signal dependencies from multiple track circuits within a geographic area.
• Employed a generative model to illustrate the LSTM network's learning capabilities.
• Used t-Distributed Stochastic Neighbor Embedding (t-SNE) for network analysis and validation.
• Compared LSTM performance against a convolutional neural network (CNN) on the same task.

Main Results:

• The LSTM network achieved a 99.7% classification accuracy on test sequences.
• No false positive fault detections were recorded, indicating high precision.
• t-SNE analysis confirmed that the LSTM network effectively learned relevant spatial and temporal data dependencies.
• The LSTM architecture demonstrated superior performance compared to the convolutional network for this specific task.

Conclusions:

• The proposed LSTM network is highly effective for railway track circuit fault detection and identification.
• LSTM networks are better suited than convolutional networks for capturing the intricate dependencies in track circuit data.
• This approach enhances railway safety and network availability through reliable fault diagnosis.