Denoising Diffusion Implicit Model Combined with TransNet for Rolling Bearing Fault Diagnosis Under Imbalanced Data.

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Data imbalances critically impair intelligent fault diagnosis accuracy, risking equipment reliability and safety.
• Existing methods struggle with imbalanced datasets, leading to suboptimal diagnostic performance.

Purpose of the Study:

• To propose a novel fault diagnosis method addressing data imbalance issues.
• To enhance diagnostic precision and generalization capabilities for intelligent systems.

Main Methods:

• Generating 2D images using Gramian angular difference field (GADF).
• Augmenting data with the denoising diffusion implicit model (DDIM).
• Developing a TransNet model combining convolutional neural networks (CNNs) and Transformers for global data processing and self-attention.

Main Results:

• Achieved over 99% recognition accuracy on the CWRU bearing and Nanchang Railway Bureau datasets.
• Demonstrated superior generalization performance compared to existing fault diagnosis methods.
• Successfully mitigated the impact of data imbalance on diagnostic precision.

Conclusions:

• The proposed DDIM-augmented TransNet method offers a robust solution for intelligent fault diagnosis with imbalanced data.
• This approach significantly improves diagnostic accuracy, reliability, and safety in industrial applications.
• The method shows strong potential for real-world deployment in equipment health monitoring.