Abstract
Proper lubrication is critical for ensuring the reliability and longevity of mechanical systems, yet its degradation due to factors like contamination or insufficient lubricant often leads to equipment failure. This study proposes a novel approach integrating Continuous Wavelet Transform (CWT) and Convolutional Neural Networks (CNN) for robust lubrication state identification. Vibration signals were collected from a pin-disk tribological system under three lubrication states: normal (NL), insufficient (IL), and contaminated (LC). CWT was applied to convert raw signals into time-frequency diagrams, which were preprocessed and input into a CNN model. The CNN architecture, comprising three convolutional layers, max pooling, and fully connected layers, was trained using the Adam optimizer with early stopping to prevent overfitting. Results demonstrated exceptional performance: the model achieved 99.8% training accuracy and 100% test accuracy, significantly outperforming traditional methods (RMS + CNN: 63.8%; PSD + CNN: 73.4%; CWT + SVM: 76.3%). t-SNE visualization confirmed distinct feature separation among lubrication states, and the confusion matrix revealed flawless classification on the test set. The method's ability to capture time-frequency characteristics via CWT and leverage CNN's deep feature learning offers a highly accurate and reliable solution for real-world lubrication monitoring.