Multiple Fault Detection in Taper Roller Bearings Based on Sensor Fusion of Acoustic Emission and Vibration Data Using Deep Learning Techniques

Abstract In rotating machinery, taper roller bearings are essential parts whose failure can have fatal consequences. To ensure safe and effective operation, these bearings must undergo condition monitoring and fault diagnosis. This study uses sensor fusion of vibration and acoustic emission data to demonstrate an integrated approach for multiple fault detection in taper roller bearings. Most of the literature classified bearing faults in 3 or 4 classes. Classification for lesser faults is an easy task as compared to classification of multiple faults. The proposed experiment demonstrate that the suggested method is highly accurate in detecting multiple bearing faults, and the integration of acoustic emission and vibration data improves the fault detection performance compared to using either data source alone. The proposed method involves collecting acoustic and vibration data using a DAQ system for multiple bearing faults and converting that data into scalogram images. This study also focused on importance of data denoising and use of optimal hyperparameter. As per proposed method data pre-processing and setting ideal hyperparameter have significantly increased the accuracy of bearing fault classification. This study uses deep learning CNN model to classify the scalogram images of bearing faults. CNN automatically extract features from scalogram images and classify images based on their similarities. The proposed approach can be used to monitorthe conditions and diagnose various faults of taper roller bearings in various industrial applications.