A modified transformer based on adaptive frequency enhanced attention, large kernel convolution, and multiscale...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing fault diagnosis is crucial for rotating machinery health monitoring.
• Deep learning (DL) models excel at feature extraction but struggle with noisy industrial data.
• Robust fault diagnosis methods are needed for real-world applications.

Purpose of the Study:

• To develop a noise-robust deep learning model for bearing fault diagnosis.
• To enhance the performance of DL models under strong noise interference.
• To improve the reliability of machinery health monitoring systems.

Main Methods:

• A novel attention-enhanced Transformer model integrating large-kernel convolution and multiscale CNNs was proposed.
• The framework combines spatiotemporal feature modeling with adaptive frequency-domain enhancement.
• The model was evaluated on the Paderborn University and Case Western Reserve University datasets.

Main Results:

• The proposed method achieved superior recognition accuracy across various signal-to-noise ratios.
• It outperformed several state-of-the-art models in noisy conditions.
• Ablation studies and visualization confirmed the model's effectiveness and noise suppression capabilities.

Conclusions:

• The attention-enhanced Transformer model offers a robust solution for bearing fault diagnosis.
• The integration of large-kernel convolution and multiscale CNNs enhances noise resilience.
• The developed framework significantly advances machinery health monitoring technology.