A Novel Fault Diagnosis Method of High-Speed Train Based on Few-Shot Learning

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Real-time monitoring of high-speed train suspension systems is crucial for safety and stability.
• Machine learning (ML) for fault diagnosis requires large, labeled datasets, which are often unavailable in practice.
• Limited labeled data leads to overfitting in traditional ML models, hindering effective fault diagnosis.

Purpose of the Study:

• To develop a novel few-shot learning method for high-speed train fault diagnosis.
• To address the challenge of insufficient labeled data in real-world industrial applications.
• To enhance the accuracy of fault detection in suspension systems.

Main Methods:

• Proposed a few-shot learning approach integrating sensor-perturbation injection.
• Incorporated meta-confidence learning to improve model robustness and accuracy.
• Conducted experiments to evaluate the method's performance against existing techniques.

Main Results:

• The proposed method demonstrated superior performance compared to existing fault diagnosis methods.
• Analysis confirmed the effectiveness of sensor perturbation in improving fault detection.
• The impact of perturbation effects and varying class numbers on detection accuracy was analyzed.

Conclusions:

• The novel few-shot learning strategy effectively enhances high-speed train fault diagnosis with limited data.
• Sensor perturbation and meta-confidence learning are key components for improving detection accuracy.
• The findings validate the proposed learning strategy for practical industrial applications.