DC Series Arc Fault Detection in Electric Vehicle Charging Systems Using a Temporal Convolution and Sparse...

Area of Science:

• Electrical Engineering
• Automotive Safety
• Artificial Intelligence

Background:

• DC series arc faults are a significant fire hazard in electric vehicle (EV) charging systems.
• These faults are difficult to detect and can compromise charging system safety.
• Existing detection methods may struggle with the complex electrical noise present during EV charging.

Purpose of the Study:

• To investigate the characteristics of DC series arc faults in EV charging systems.
• To develop an advanced arc fault detection algorithm for enhanced EV safety.
• To ensure reliable and real-time fault detection even in noisy charging environments.

Main Methods:

• Developed an improved hybrid arc fault model in Simulink for simulation.
• Conducted experiments on a real EV charging platform to analyze arc fault behavior.
• Designed and implemented a deep neural network algorithm (Arc_TCNsformer) utilizing Temporal Convolutional Networks and Transformers.
• Performed end-to-end fault recognition directly from current signal samples without manual feature engineering.

Main Results:

• Arc faults generate high-frequency noise affecting charger output and battery voltage quality.
• Real-world arc faults lack alarm indications and cause significant current signal disturbances.
• Normal charging stages (startup, pre-charge) exhibit current characteristics mimicking arc faults.
• The Arc_TCNsformer algorithm demonstrated high detection accuracy and robustness in complex noise environments.
• The algorithm achieved reliable real-time performance on embedded edge computing platforms.

Conclusions:

• DC series arc faults pose a critical safety risk in EV charging.
• The proposed Arc_TCNsformer algorithm offers a robust and accurate solution for real-time arc fault detection.
• This AI-driven approach enhances the safety of electric vehicle charging infrastructure.