Electric Bicycle Series Arc Fault Identification Method Based on Improved PCA and SVM.

Area of Science:

• Electrical Engineering
• Materials Science
• Safety Engineering

Background:

• Electric bicycles (e-bikes) present environmental benefits but pose fire risks due to series arc faults.
• Identifying these faults under varying operating conditions (SoC, torque, speed) is crucial for safety.

Purpose of the Study:

• To develop and validate a robust method for series arc fault identification in e-bikes.
• To address complex operating conditions and differentiate between normal, DC-side, and AC-side arc fault states.

Main Methods:

• Extraction of eight time-domain and frequency-domain features (RMS, STD, SK, KUR, CA, AFE, AFM, AFK).
• Application of an improved principal component analysis (PCA)-based method for feature fusion into a five-dimensional representation.
• Classification of fused features using a radial basis function (RBF)-support vector machine (SVM) model.

Main Results:

• Achieved 98.68% test accuracy, 0.9869 Macro-F1 score, and 0.9931 Macro-AUC.
• Demonstrated the method's effectiveness across various operating states and fault types.
• Analysis confirmed accuracy-cost tradeoff and deployment feasibility.

Conclusions:

• The proposed PCA-based feature fusion and RBF-SVM classification offers an interpretable and lightweight solution.
• This method enhances safety for e-bike controllers, battery management systems (BMSs), and onboard monitoring.