Deep network fault diagnosis for imbalanced small-sized samples via a coupled adversarial autoencoder based on the...

Area of Science:

• Engineering
• Computer Science
• Machine Learning

Background:

• Deep learning fault diagnosis requires extensive labeled data.
• Insufficient or imbalanced samples degrade model performance and stability.

Purpose of the Study:

• To develop a novel data augmentation method for deep learning fault diagnosis.
• To address challenges of small-sized and imbalanced datasets.

Main Methods:

• Introduced a coupled adversarial autoencoder (CoAAE) utilizing Bayesian methods.
• CoAAE generates synthetic samples by capturing data distribution and adversarial training.
• A parallel coupled network addresses sample imbalance by learning joint distributions.

Main Results:

• CoAAE effectively augments imbalanced datasets for fault diagnosis.
• Experiments on a bearing dataset demonstrated superior performance over advanced methods.
• The method improved accuracy and stability of deep learning diagnosis models.

Conclusions:

• The proposed CoAAE method offers a robust solution for data augmentation in fault diagnosis.
• This approach enhances the reliability and accuracy of deep learning diagnostic models with limited data.