An improved re-parameterized visual geometry group network for rolling bearing fault diagnosis

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning models significantly advance data-driven fault diagnosis.
• Classical convolutional and multi-branch networks face challenges in computational efficiency and feature extraction.

Purpose of the Study:

• To propose an improved re-parameterized visual geometry group (RepVGG) network for enhanced rolling bearing fault diagnosis.
• To address limitations in computational complexity and feature extraction of existing deep learning models.

Main Methods:

• Data augmentation to increase dataset size.
• Conversion of 1D vibration signals to single-channel and then three-channel time-frequency images via Short-Time Fourier Transform and pseudo-coloring.
• Development of a RepVGG model with an embedded convolutional block attention mechanism for feature extraction and classification.

Main Results:

• The proposed RepVGG model demonstrates strong adaptability and improved performance in rolling bearing fault diagnosis.
• Effective extraction of defect features from processed time-frequency images.

Conclusions:

• The improved RepVGG network offers a robust solution for rolling bearing fault diagnosis.
• The integration of attention mechanisms enhances the model's ability to identify bearing defects accurately.