A fault diagnosis method for aero-engine inter-shaft bearings based on 1DCNN-Transformer-BiGRU

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Aero-engine inter-shaft bearings face challenges in fault diagnosis due to strong noise and feature aliasing.
• Accurate fault feature extraction is critical for ensuring operational safety and reliability.

Purpose of the Study:

• To develop an advanced fusion diagnostic method for robust fault detection in aero-engine inter-shaft bearings under severe noise.
• To enhance the accuracy and reliability of bearing condition monitoring in demanding operational environments.

Main Methods:

• A three-stage progressive network architecture integrating a dual-scale one-dimensional convolutional neural network (1DCNN), a multi-head self-attention Transformer, and a bidirectional gated recurrent unit (BiGRU).
• 1DCNN for local temporal feature extraction, Transformer for modeling feature dependencies, and BiGRU for bidirectional temporal analysis.
• Techniques like batch normalization and dropout were employed to stabilize training and prevent overfitting.

Main Results:

• The proposed fusion diagnostic model achieved 97% diagnostic accuracy on the Harbin Institute of Technology aero-engine inter-shaft bearing dataset under extreme noise conditions (SNR = -5 dB).
• Demonstrated superior performance compared to existing methods in maintaining diagnostic accuracy under controlled noise.

Conclusions:

• The integrated network effectively extracts fault-sensitive features and captures complex temporal dependencies, outperforming traditional methods.
• The proposed method offers a robust solution for reliable aero-engine inter-shaft bearing fault diagnosis in noisy environments.