Fault Diagnosis of the Rolling Bearing by a Multi-Task Deep Learning Method Based on a Classifier Generative...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Accurate fault diagnosis is critical for the reliable operation of rotating machinery.
• Traditional deep learning methods for fault diagnosis often rely solely on supervised learning, leading to limitations in feature extraction and interpretability.
• Existing methods struggle with effectively utilizing small convolution kernels for feature extraction, hindering controllability and understanding.

Purpose of the Study:

• To propose an innovative semi-supervised learning method for bearing fault diagnosis.
• To enhance the extraction of both local and global features from rotating machinery data.
• To improve the generalization ability and interpretability of fault diagnosis models.

Main Methods:

• Designed multi-scale dilated convolution squeeze-and-excitation residual blocks for feature extraction.
• Employed a classifier generative adversarial network for simultaneous unsupervised and supervised multi-task learning.
• Utilized supervised learning for fine-tuning the model, incorporating implicit data augmentation.

Main Results:

• The proposed method effectively extracts multi-scale local and global features.
• Simultaneous unsupervised and supervised learning improved the model's generalization capabilities.
• Experiments on two datasets demonstrated the superiority of the developed semi-supervised learning approach.

Conclusions:

• The novel semi-supervised learning method offers a significant advancement in bearing fault diagnosis.
• The integration of multi-scale feature extraction and multi-task learning enhances diagnostic performance.
• The approach provides a more interpretable and controllable solution compared to traditional methods.