Trend Feature Consistency Guided Deep Learning Method for Minor Fault Diagnosis.

Area of Science:

• Engineering
• Computer Science

Background:

• Deep learning enables fault diagnosis without precise mechanism models.
• Accurate diagnosis of minor faults is challenging due to limited training data size and noise pollution.

Purpose of the Study:

• To develop a robust deep neural network (DNN) learning mechanism for enhanced feature representation.
• To improve the accuracy and reliability of fault diagnosis, especially for minor faults, using DNNs.

Main Methods:

• Designed a novel loss function for DNNs to ensure consistency in trend features and fault direction.
• Implemented a new learning mechanism to enhance the feature representation capabilities of deep neural networks.

Main Results:

• The proposed method successfully trains DNNs with only 100 noisy samples for gearbox fault diagnosis.
• Achieved satisfactory fault diagnosis accuracy, outperforming traditional methods requiring over 1500 samples.

Conclusions:

• The novel learning mechanism and loss function establish a more robust and reliable DNN-based fault diagnosis model.
• This approach effectively discriminates faults with similar classifier membership values, a limitation of traditional methods.