An Anti-Noise Convolutional Neural Network for Bearing Fault Diagnosis Based on Multi-Channel Data.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Industrial bearings face variable working conditions and inevitable equipment noise.
• Existing neural network models struggle with noise, impacting bearing fault diagnosis accuracy.

Purpose of the Study:

• To enhance the anti-noise performance of neural networks for bearing fault diagnosis.
• To develop a robust method for identifying bearing faults in real-world, noisy industrial settings.

Main Methods:

• A multi-channel sample representation using envelope time-frequency spectrum and 3D filtering for fault feature extraction.
• A multi-channel data fusion neural network (MCFNN) incorporating dropout for training.
• Utilizing a dataset with diverse rotation speeds and loads under varying noise conditions.

Main Results:

• Achieved 99.00% fault classification accuracy in a noise-free environment.
• Demonstrated superior performance in noisy conditions, outperforming other methods by 11.80% at 0 dB SNR and 32.89% at -4 dB SNR.
• The proposed MCFNN effectively extracts fault features and classifies bearing conditions accurately despite environmental noise.

Conclusions:

• The proposed MCFNN and multi-channel sample representation method significantly improve bearing fault diagnosis in noisy environments.
• The method offers a robust solution for real-world industrial applications requiring reliable bearing health monitoring.
• This approach enhances the anti-noise capabilities of diagnostic systems, crucial for industrial machinery maintenance.