Few-Shot Learning-Based Light-Weight WDCNN Model for Bearing Fault Diagnosis in Siamese Network.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence

Background:

• Traditional deep learning fault diagnosis requires extensive training data, which is often unavailable in dynamic manufacturing settings.
• Existing models can be complex and computationally expensive, limiting their efficiency in rapidly changing industrial environments.

Purpose of the Study:

• To develop a few-shot learning model for effective bearing fault diagnosis using limited data.
• To create a computationally efficient fault diagnosis model suitable for manufacturing sites.

Main Methods:

• Implemented a few-shot learning approach for data-efficient fault diagnosis.
• Integrated Depthwise Separable Convolution layers to reduce model parameters.
• Optimized hyperparameters and model architecture through systematic adjustments.

Main Results:

• Achieved higher accuracy in bearing fault diagnosis compared to existing models.
• Demonstrated a significant reduction in model parameters, enhancing computational efficiency.
• Validated the model's effectiveness with limited training data.

Conclusions:

• The proposed few-shot learning model offers a viable solution for bearing fault diagnosis with minimal data.
• The integration of Depthwise Separable Convolution layers contributes to a more efficient and accurate diagnostic tool.
• This approach addresses the practical challenges of data scarcity and computational cost in industrial fault diagnosis.