A Siamese Vision Transformer for Bearings Fault Diagnosis

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning fault diagnosis methods face challenges due to limited training data and complex operating conditions.
• Intelligent diagnostic systems require robust feature extraction and classification capabilities for reliable performance.

Purpose of the Study:

• To propose an intelligent bearing fault diagnosis method, the Siamese Vision Transformer (SVT), designed for scenarios with limited training data and complex work conditions.
• To enhance the performance and data diversity of deep learning models for bearing fault diagnosis.

Main Methods:

• The proposed Siamese Vision Transformer (SVT) integrates Siamese networks and Vision Transformers for high-level feature vector extraction and fault classification.
• A novel bidirectional Kullback-Leibler divergence loss function is introduced to improve model performance.
• A random mask training strategy is employed to increase input data diversity.

Main Results:

• The SVT method achieved reasonably high accuracy on the Case Western Reserve University and Paderborn bearing datasets, even with limited data.
• The model demonstrated satisfactory generation capability for cross-domain tasks, indicating robustness and adaptability.
• Comparative tests confirmed the effectiveness of the proposed method over existing approaches.

Conclusions:

• The Siamese Vision Transformer is a promising approach for intelligent bearing fault diagnosis, particularly in data-scarce and complex environments.
• The integration of novel loss functions and training strategies enhances the model's diagnostic accuracy and generalization ability.
• This research contributes to the advancement of reliable and efficient condition monitoring systems for rotating machinery.