Related Experiment Video
Updated: Jun 27, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
A Safe-Domain Generative Adversarial Network with Swin Transformer for Noisy Imbalanced Fault Diagnosis
Xiao Lai1,2, Xiaohan Zhang3, Zhiqi Xie1
1School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China.
Sensors (Basel, Switzerland)
|June 26, 2026
Summary
This study introduces a novel Safe Domain Generative Adversarial Network with Swin Transformer (SDGAN-ST) to improve intelligent fault diagnosis. The method effectively handles imbalanced data and label noise, achieving high accuracy in industrial applications.
Area of Science:
- Machine Learning
- Artificial Intelligence
- Industrial Fault Diagnosis
Background:
- Data-driven fault diagnosis faces challenges with insufficient fault data, leading to imbalanced datasets.
- Label noise from manual recording and measurement errors degrades diagnostic performance.
Purpose of the Study:
- To propose a robust method for intelligent fault diagnosis that addresses data imbalance and label noise.
- To enhance the accuracy and reliability of fault diagnosis in industrial settings.
Main Methods:
- A Safe Domain Generative Adversarial Network (SDGAN) is used to eliminate noisy samples and generate high-quality minority class data.
- A Swin Transformer classifier is employed to capture global information for accurate fault sample classification.
- The proposed SDGAN-ST method integrates safe domain selection with advanced generative and classification models.
Main Results:
- SDGAN-ST achieved high accuracy on the CWRU dataset under severe data imbalance ratios (up to 98.88%).
- The method demonstrated 100% accuracy on a real-world oxygen compressor bearing dataset across all imbalance ratios.
- SDGAN-ST exhibited robust and stable diagnostic performance even with significant label noise (20-40%).
Conclusions:
- The proposed SDGAN-ST method effectively overcomes data imbalance and label noise issues in intelligent fault diagnosis.
- SDGAN-ST offers superior robustness and diagnostic accuracy compared to traditional methods like WGAN-GP.
- This approach significantly advances the reliability of industrial fault diagnosis systems.
