A Hybrid Deep Learning Approach for Bearing Fault Diagnosis Using Continuous Wavelet Transform and Attention-Enhanced...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing faults are critical in industrial machinery, leading to failures and downtime.
• Accurate fault diagnosis is essential for predictive maintenance and operational efficiency.
• Traditional methods struggle with complex, non-stationary vibration signals.

Purpose of the Study:

• To develop a robust hybrid deep learning model for accurate bearing fault diagnosis.
• To enhance feature extraction from non-stationary and nonlinear vibration signals.
• To validate the model's generalization capabilities on diverse bearing datasets.

Main Methods:

• Integration of Continuous Wavelet Transform (CWT) for time-frequency analysis.
• Development of an attention-enhanced spatiotemporal feature extraction framework.
• Utilizing Multi-Head Self-Attention (MHSA), Bidirectional Long Short-Term Memory (BiLSTM), and 1D Convolutional Residual Network (1D conv ResNet).

Main Results:

• The hybrid model effectively captures spatial and temporal dependencies in vibration signals.
• Demonstrated strong noise resilience and accurate feature extraction.
• Achieved high accuracy and clear separability between fault categories on laboratory and Paderborn datasets.

Conclusions:

• The proposed hybrid deep learning approach offers precise and reliable bearing fault diagnosis.
• The model exhibits strong generalization capabilities, suitable for diverse industrial conditions.
• Significant potential for real-time predictive maintenance applications in complex environments.