Fault Diagnosis Method for Boost Chopper of High-Speed Maglev Train Based on Deep Time-Series Modeling

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Railway Systems

Background:

• Boost choppers (HS) are critical for 440 V grids in high-speed maglev trains, powering essential systems.
• HS faults present complex challenges due to long-period fluctuations and transient characteristics, hindering single-network diagnosis.
• Accurate fault diagnosis is vital for ensuring the stable operation and safety of maglev trains.

Purpose of the Study:

• To develop a multi-scale fault diagnosis method for boost choppers (HS) in maglev trains.
• To address the limitations of single networks in capturing diverse fault characteristics.
• To enhance the precision and reliability of HS fault identification.

Main Methods:

• Proposed a dual-branch architecture combining TimesNet and Convolutional Neural Networks (CNN).
• TimesNet branch utilizes Fast Fourier Transform (FFT) and Inception convolution for global periodicity analysis.
• CNN branch employs stacked kernels and downsampling for local high-frequency anomaly detection.
• Implemented feature fusion to synergistically combine global and local features.

Main Results:

• The proposed TimesNet-CNN method demonstrated superior performance over existing models (TimesNet, CNN, ResNet, Informer).
• Achieved high precision, recall, and F1-score on a real-world dataset with 11 system states.
• Validated the effectiveness of the dual-branch approach in capturing multi-scale fault features.

Conclusions:

• The dual-branch TimesNet-CNN architecture effectively identifies boost chopper (HS) faults in maglev trains.
• The method's ability to capture both global periodicity and local transiency leads to high-precision fault diagnosis.
• This approach significantly contributes to ensuring the operational stability and safety of high-speed maglev trains.