Machine Learning-Based Phase Resolved Partial Discharge Pattern Classification using HOG and PCA

Partial Discharge (PD) detection in rotating machine insulation is crucial for preventing failures in electrical systems. This paper proposes a machine-learning method for classifying Phase-Resolved Partial Discharge (PRPD) patterns using six machine-learning classifiers, Support Vector Machine (SVM), Bagging Tree (BaggTree), K- Nearest Neighbors (KNN), Logistic Regression (LogReg), Decision Tree (DecTree), and Random Forest (RandForest). The Histogram of Oriented Gradients (HOG) is used as the feature extraction method, while Principal Component Analysis (PCA) is applied for dimensionality reduction. The models are evaluated based on their classification performance and execution time. In terms of classification accuracy, Bagging Tree achieves the highest accuracy of $\mathbf{8 9. 7 9 \%}$, followed by Random Forest and SVM with $\mathbf{8 7. 7 2 \%}$ and 87.10%. Furthermore, KNN has the fastest execution time of $\mathbf{3 0 1. 8 2 0}$ seconds, followed by Bagging Tree with 409.741 seconds. This work also investigates performance metrics such as accuracy, precision, recall, F1-score, specificity, and balanced accuracy, demonstrating the effectiveness of combining HOG and PCA with machine learning algorithms to improve PD detection. This study demonstrates the effectiveness of machine learning in enhancing PD detection, contributing to improved predictive maintenance of rotating machines.