Fault Classification in Rolling Element Bearing Based on Vibration Signature Using Artificial Neural Network

In mechanical systems, the role of rolling element bearings is pivotal, as their vulnerability to various faults emphasizes the potential for performance decline, increased maintenance costs, and potentially catastrophic failures. To address these issues, this study focuses on leveraging Artificial Neural Networks (ANN) to improve the identification of faults in rolling element bearings through the analysis of vibration signatures. The vibration signals have been acquired from test bearings (both healthy and faulty) using the developed experimental setup. Statistical parameters have been extracted from the vibration data to create a comprehensive dataset for training, validation, and testing the ANN. By training the ANN with this dataset, the study establishes a reliable and automated system for the classification of faults in rolling element bearings. The outcomes are presented using confusion matrices, depicting correctly classified and misclassified instances. The training confusion matrix highlights a 90.5% accurate classification rate, while the test confusion matrix demonstrates 88.9% accuracy. The overall confusion matrix culminates in an 86.7% accurate classification. Finally, the developed ANN model has been validated by predicting the bearing conditions with unseen vibration data, achieving a 100% accurate prediction. This study contributes to advancing dependable and efficient fault detection systems, underscoring the potential of ANN in enhancing monitoring and maintenance strategies for rolling element bearings.