Multi-scale attention network (MSAN) for track circuits fault diagnosis

Area of Science:

• Railroad Engineering
• Signal Systems
• Artificial Intelligence

Background:

• Track circuits are critical components of railroad signaling systems, essential for safe and efficient train operations.
• Prompt and accurate fault diagnosis is crucial to prevent operational disruptions and safety incidents.

Purpose of the Study:

• To develop an advanced fault diagnosis method for track circuits using a multi-scale attention network.
• To improve the accuracy and efficiency of identifying track circuit faults.

Main Methods:

• Utilized Gramian Angular Field (GAF) to convert 1D time-series data into 2D images for convolutional neural network (CNN) processing.
• Designed a novel feature fusion training structure incorporating spatial attention mechanisms for multi-scale feature extraction.

Main Results:

• Achieved a high fault diagnosis accuracy of 99.36% on real-world track circuit fault datasets.
• Demonstrated superior performance compared to existing classical and state-of-the-art fault diagnosis models.
• Ablation studies confirmed the significant contribution of each module within the proposed model.

Conclusions:

• The proposed multi-scale attention network offers a highly accurate and effective solution for track circuit fault diagnosis.
• The GAF transformation and feature fusion strategy are key to the model's success.
• This approach has the potential to significantly enhance railroad safety and operational efficiency.