A few-shot sample augmentation algorithm based on SCAM and DEPS for pump fault diagnosis.

Area of Science:

• Engineering
• Machine Learning
• Data Science

Background:

• Pumps are vital in agriculture, industry, and military applications, requiring robust fault diagnosis.
• Existing fault classification models heavily rely on large datasets, which are difficult to acquire in real-world scenarios.
• The performance of fault diagnosis models often correlates with the volume of training data.

Purpose of the Study:

• To develop an effective few-shot learning fault diagnosis method for pump systems.
• To address the challenge of limited real-world data in fault classification.
• To enhance the accuracy and reliability of pump fault diagnosis.

Main Methods:

• Integration of domain migration theory and sample expansion techniques.
• Development of a few-shot learning framework for fault diagnosis.
• Utilisation of the T-SNE visualization algorithm to validate the self-calibration attention mechanism (SCAM) and distribution edge prediction strategy (DEPS).

Main Results:

• The proposed algorithm successfully maps expanded sample spaces, preventing feature aliasing among similar categories.
• Significant enhancement in both the quality and quantity of training samples was achieved.
• The methodology demonstrated a marked increase in few-shot task accuracy, achieving 72% on the 9-way-15-shot task, outperforming other methods by approximately 30%.

Conclusions:

• The developed few-shot learning method effectively improves pump fault diagnosis accuracy with limited data.
• The approach shows strong potential for applicability in various few-shot diagnosis scenarios.
• This work offers a viable solution for data-scarce fault diagnosis challenges.