Detection of Bearing Faults from Vibration Signals

Bearings are critical mechanical components in rotating machinery, playing a vital role in system safety and operational continuity. In this study, the Case Western Reserve University (CWRU) bearing dataset is used to perform fault classification using four machine learning algorithms: Random Forest, XGBoost, Support Vector Machine (SVM), and Naive Bayes. Based on statistical features extracted in the time domain, the performance of each model is evaluated using accuracy, precision, recall, and F1-score metrics. The results reveal that Random Forest and XGBoost algorithms achieved superior performance with 95.73% accuracy and 96% in precision, recall, and F1-score. The SVM model, with 93.73% accuracy, stands out as a robust alternative, while the Naive Bayes algorithm shows relatively lower performance with 92.40% accuracy. Additionally, an individual feature-based classification analysis indicates that standard deviation (sd) and root mean square (RMS) features contribute most significantly to model performance. This study provides a comprehensive performance analysis of traditional machine learning algorithms, offering a valuable reference for early and accurate detection of bearing faults.