Multi-Frequency Augmentation framework via information active capture for machinery intelligent fault diagnosis

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Deep neural networks are increasingly used for machinery fault diagnosis.
• Directly using raw machinery signals as model input presents limitations.
• Improving model generalization is crucial for accurate fault diagnosis.

Purpose of the Study:

• To propose a novel Multi-Frequency Augmentation framework for machinery fault diagnosis.
• To enhance the generalization ability of intelligent models by incorporating multi-frequency information.
• To address the limitations of learning directly from raw machinery signals.

Main Methods:

• A data augmentation method to create sample pairs from raw signals.
• Frequency Components Recombination for dataset augmentation, particularly in few-shot learning.
• A Multi-Frequency Capture Network designed for feature augmentation through sample pair learning.

Main Results:

• The Multi-Frequency Augmentation framework demonstrated effectiveness and compatibility through experimental validation.
• The proposed framework successfully improved machinery fault diagnosis performance.
• Competitive results were achieved compared to existing state-of-the-art methods.

Conclusions:

• The Multi-Frequency Augmentation framework offers a new perspective for analyzing models and datasets in fault diagnosis.
• The approach shows significant potential for practical applications in intelligent machinery monitoring.
• Enhancing models with multi-frequency information is a viable strategy for improving diagnostic accuracy.