Convolutional Neural Network-Based Pattern Recognition of Partial Discharge in High-Speed Electric-Multiple-Unit...

Area of Science:

• Electrical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Partial discharge detection is vital for assessing insulation integrity in high-speed electric multiple unit (EMU) cable terminals.
• Identifying specific defect types is crucial for preventing failures and ensuring operational safety.
• Existing neural network (NN) models have limitations in accurately classifying diverse discharge signals.

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Network (CNN) model for classifying partial discharge signals from defective EMU cable terminals.
• To compare the performance of the CNN model against traditional back-propagation NN and radial basis function NN models.
• To establish a reliable method for identifying four typical defects in high-speed EMU cable terminals.

Main Methods:

• Preparation of cable terminal samples with four distinct, typical defects from high-speed EMUs.
• Development of a cable discharge testing system employing high-frequency current sensing to capture discharge signals.
• Creation of comprehensive datasets for each defect type.
• Implementation and comparison of CNN, back-propagation NN, and radial basis function NN for signal classification.

Main Results:

• The CNN-based model demonstrated superior accuracy in classifying partial discharge signals corresponding to specific defects.
• The CNN model significantly outperformed both the back-propagation NN and radial basis function NN models in defect identification.
• Established datasets provide a valuable resource for further research in partial discharge analysis.

Conclusions:

• Convolutional Neural Networks (CNNs) offer a highly effective solution for the accurate classification of partial discharge signals in high-speed EMU cable terminals.
• The proposed CNN approach provides a significant advancement over existing neural network methods for defect diagnosis in critical rail infrastructure.
• This study validates the potential of AI-driven techniques for enhancing the safety and reliability of high-speed rail systems.