Advancements in Bearing Defect Diagnosis: Deep Learning-based Signal Processing and Real-time Fault Detection

Bearings assume a pivotal role in the operation of rotating machinery, particularly in large motor systems, underscoring the imperative of early detection of bearing faults for effective machine health monitoring. This research introduces an innovative AI-driven approach for the identification and assessment of bearing anomalies, particularly in correlation with diverse shaft speeds. Vibration data are acquired in the time domain using piezoelectric accelerometers for bearings exhibiting varying degrees of health and fault conditions. The proposed methodology employs spectrograms to represent vibration signals, incorporating pre-processing techniques such as continuous wave transformation to generate scalograms, converting one-dimensional vibration signals into two-dimensional images. Feature extraction and health status classification are executed utilizing AlexNet, a convolutional neural network. Experimental evaluation conducted on a bearing test rig demonstrates prediction accuracies of 95.23%, 100%, and 98.43% for fault-related anomalies, respectively, supported by the Case Western Reserve University bearing dataset and Vishwakarma Institute of Technology vibration laboratory dataset. These findings affirm the robustness of the proposed approach, highlighting its efficacy in accurately detecting various manifestations of rolling bearing faults through deep learning applied to observational data.