A comparative analysis of machine learning algorithms for fault classification in cylindrical roller bearings

With the growing demand for enhanced machinery reliability, the application of artificial intelligence (AI) in fault diagnosis is becoming increasingly important. However, limited research has systematically compared the performance of various machine learning algorithms for bearing fault diagnosis. This paper presents a comparative analysis of five widely-used machine learning algorithms-linear SVM, Gaussian SVM, SVM Kernel, Weighted K-Nearest Neighbors (WKNN), and Artificial Neural Networks (ANN). Faults were induced using wire EDM on test bearings, and vibration data were recorded using a National Instruments data acquisition system with LABVIEW. Results indicate that while weighted KNN demonstrated 100% accuracy in testing, ANN emerged as the most reliable, achieving 100% accuracy in both validation and testing phases. The trained ANN was further employed to predict bearing conditions across ten random datasets, affirming its potential for real-time condition monitoring of industrial machinery.