Enhanced Fault Diagnosis of Vertical Friction Torque Using Improved VGG-CNN Network

This paper presents a CNN-based fault diagnosis method utilizing Infrared Thermography (IRT) to improve the low diagnostic rates of friction torque faults in upright placements and to address bearing faults in a test bench. Employing non-destructive, non-contact infrared thermal imaging technology, the study conducts tests across six scenarios: undamaged, damaged inner ring, damaged outer ring, defective marble, insufficient lubrication, and damaged inner and outer ring bearings. This article implements a thermal image processing technique based on two-dimensional discrete wavelet transform, combining VGG Net model with CBAM attention mechanism to improve classification accuracy while reducing training time. Convolutional neural networks were then employed for fault classification and performance evaluation, demonstrating superior results compared to support vector machines. This approach effectively identifies bearing torque faults, achieving an impressive 99.80% accuracy in classifying faulty bearings, indicating its broad applicability.