Fault Detection and Severity Classification of Inter-turn Short-Circuit Faults in PMSMs Using a Combination of Projection-Based Support Vector Machine and Bayesian Neural Network

This paper presents a novel approach for detecting and classifying the severity of Inter-Turn Short Circuit (ITSC) faults in a Permanent Magnet Synchronous Machine (PMSM) using a combination of Dempster-Shafer (DS), Support Vector Machine (SVM), Projection Recurrent Neural Network (PRNN), and Bayesian Neural Network (BNN). The dataset comprises three-phase current signals gathered under healthy and faulty conditions at six severity levels. Feature extraction is performed using Discrete Wavelet Transform (DWT) and Power Spectral Density (PSD), followed by Kernel Principal Component Analysis (KPCA) for dimensionality reduction. For ITSC fault detection, an optimized classifier is developed by integrating DS, SVM, and PRNN to ensure high convergence speed and robust decision-making under uncertainty. For fault severity classification, a BNN is employed to provide probabilistic predictions and enhance reliability in distinguishing fault severity levels. The proposed method achieves 98.10% accuracy in fault detection and demonstrates a strong correlation between classification performance and increasing fault severity. The BNN model outperforms LSTM by achieving 97.33% accuracy at the highest severity level and ensures high precision, recall, and F1-score across all severity levels. The results confirm that the proposed approach provides a reliable and interpretable solution for ITSC fault detection and severity estimation in PMSMs.