A Rolling Bearing Fault Diagnosis Based on Conditional Depth Convolution Countermeasure Generation Networks under...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Data Science

Background:

• Bearing fault diagnosis faces challenges due to insufficient samples and imbalanced data distributions, hindering accuracy.
• Existing generative adversarial networks (GANs) struggle with training instability and gradient issues in imbalanced datasets.

Purpose of the Study:

• To propose an efficient data augmentation method for rolling bearing fault diagnosis using conditional deep convolutional adversarial generative networks (C-DCGAN).
• To address limitations of insufficient samples and unbalanced data distribution in bearing fault diagnosis.
• To improve the accuracy and robustness of fault classification in rolling bearings.

Main Methods:

• Utilized conditional constraints within the GAN framework to guide sample generation for balanced, multi-category fault data expansion.
• Optimized the generative network structure with self-defined skip connections and spectral normalization.
• Employed Wasserstein distance with a penalty term as the loss function to enhance training stability and feature extraction.
• Integrated generated data with original datasets for training a one-dimensional convolution neural network (1D-CNN) for fault diagnosis.

Main Results:

• The C-DCGAN effectively generated simulation sample data closely resembling real data distributions.
• The proposed method successfully balanced multi-category fault data, overcoming limitations of small sample sizes.
• The fault diagnosis model incorporating augmented data demonstrated improved fault classification performance for rolling bearings.

Conclusions:

• The C-DCGAN-based data augmentation strategy significantly enhances rolling bearing fault diagnosis accuracy.
• The method provides a robust solution for handling insufficient and imbalanced datasets in condition monitoring.
• This approach offers a promising direction for improving the reliability of predictive maintenance in rotating machinery.