A Fusion Model for Intelligent Diagnosis of Gear Faults with Small Sample Sizes

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Gear faults are a common cause of rotating machinery failure.
• Current intelligent diagnosis models face challenges with data requirements and feature extraction.
• Shallow models need manual feature engineering, while deep models require extensive data.

Purpose of the Study:

• To propose a novel fusion model, CBAM-TCN-SVM, for intelligent gear fault diagnosis.
• To address limitations of existing models concerning data scale and prior knowledge dependency.
• To enhance the accuracy and feasibility of gear fault prediction.

Main Methods:

• Developed a fusion model combining Temporal Convolutional Network (TCN), Convolutional Block Attention Module (CBAM), and Support Vector Machine (SVM).
• Utilized CBAM-TCN to extract deep fault features from frequency-domain sequence data using attention mechanisms.
• Employed SVM for the final intelligent classification of gear fault types.

Main Results:

• The proposed CBAM-TCN-SVM model achieved a classification accuracy of 98.3%.
• Demonstrated effective deep fault feature extraction through multi-layer convolutions and attention mechanisms.
• Validated the model's feasibility and superior performance in gear fault prediction.

Conclusions:

• The CBAM-TCN-SVM fusion model successfully integrates deep learning and shallow classification advantages.
• The method overcomes the constraints of limited data and reliance on prior knowledge in gear fault diagnosis.
• This approach offers a robust and accurate solution for intelligent gear health monitoring and prediction.