MAB-DrNet: Bearing Fault Diagnosis Method Based on an Improved Dilated Convolutional Neural Network

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearing fault diagnosis is critical for manufacturing equipment safety.
• Noisy environments degrade the performance of existing deep learning fault diagnosis methods.
• Nonlinear characteristics in bearing signals complicate accurate fault detection.

Purpose of the Study:

• To propose an improved deep learning model for bearing fault diagnosis in noisy environments.
• To enhance feature extraction and global information handling capabilities for better accuracy.
• To develop a noise-resilient method for reliable fault identification.

Main Methods:

• Designed a dilated residual network (DrNet) to expand the perceptual field for feature capture.
• Developed a max-average block (MAB) module to enhance feature extraction.
• Integrated a global residual block (GRB) module to improve global information processing.

Main Results:

• The proposed MAB-DrNet demonstrated strong noise immunity on the CWRU dataset.
• Achieved 95.57% accuracy with added Gaussian white noise at a -6 dB SNR.
• Outperformed existing advanced methods in noisy bearing fault diagnosis.

Conclusions:

• MAB-DrNet offers a highly accurate and noise-resilient solution for rolling bearing fault diagnosis.
• The model effectively handles complex noisy environments, improving equipment reliability.
• The integration of MAB and GRB modules significantly enhances diagnostic performance.