Benchmarking Machine Learning Algorithms for Bearing Fault Classification Using Vibration Data – A Deployment-Oriented Study

The paper presents a deployment-oriented study for bearing fault classification using vibration data collected from rotating machinery. The proposed framework aims to benchmark machine learning algorithms for identifying various fault types in bearing systems, a crucial task for predictive maintenance in industrial settings. The study leverages a Random Forest model for classification, trained on features extracted from vibration signals, such as statistical metrics (mean, standard deviation, skewness, kurtosis) and frequency-domain features (crest factor, root mean square). This paper also compares the performance of multiple algorithms, including XGBoost, MLP (Multilayer Perceptron), and CNN (Convolutional Neural Networks), based on accuracy and computational efficiency. The framework is implemented in a Flask web application that allows users to upload vibration data in CSV format, interact with the classification model, and receive predictions with confidence scores. MySQL is employed for user management and data storage. The system is designed to be scalable, user-friendly, and capable of being deployed for real-time fault detection in industrial applications. The study highlights the effectiveness of ensemble models such as Random Forest and XGBoost for fault classification, while also considering the practical aspects of model deployment and ease of use in industrial settings.

Keywords:
Bearing fault classification, vibration data, machine learning algorithms, Random Forest, XGBoost, MLP, CNN, Flask web application, predictive maintenance, MySQL, industrial applications.