A fault diagnosis method based on an improved diffusion model under limited sample conditions.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are crucial for mechanical system stability and safety.
• Limited sample sizes hinder accurate fault diagnosis in practical applications.
• Existing methods often struggle with suboptimal diagnostic accuracy due to data constraints.

Purpose of the Study:

• To develop an advanced rolling bearing fault diagnosis method addressing small sample size limitations.
• To improve the generalization capability and accuracy of fault diagnostic models.
• To provide a novel approach for fault detection and prevention in industrial settings.

Main Methods:

• Utilizing an improved denoising diffusion probability model (DDPM) to generate one-dimensional vibration data.
• Incorporating feature differences between original and generated data into the loss function for targeted data generation.
• Employing a one-dimensional convolutional neural network (1D-CNN) for effective fault feature extraction.

Main Results:

• The proposed method significantly enriches datasets, enhancing diagnostic model generalization.
• The innovative loss function ensures more directional and targeted data generation.
• Experimental results demonstrate improved accuracy and reliability in rolling bearing fault diagnosis.

Conclusions:

• The improved DDPM-based method effectively overcomes small sample size challenges in fault diagnosis.
• This approach enhances the accuracy and reliability of identifying rolling bearing faults.
• The study offers a promising solution for reducing system failures and maintenance costs in mechanical systems.