A variable-speed-condition bearing fault diagnosis methodology with recurrence plot coding and MobileNet-v3 model

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing condition monitoring is crucial for industrial machinery reliability.
• Traditional fault diagnosis methods struggle with non-stationary vibration data from variable-speed conditions.
• Developing robust and accurate fault diagnosis techniques is essential.

Purpose of the Study:

• To propose an advanced bearing fault diagnosis approach using Recurrence Plot (RP) coding and MobileNet-v3.
• To improve the characterization of non-stationary vibration features for variable-speed conditions.
• To achieve higher diagnostic accuracy and efficiency compared to existing methods.

Main Methods:

• Generated 3500 Recurrence Plot (RP) images representing seven fault modes using angular domain resampling and RP coding.
• Employed the MobileNet-v3 model for bearing fault diagnosis based on the generated RP images.
• Conducted experimental validation using bearing vibration data.

Main Results:

• The RP image coding method achieved a test accuracy of 99.99%, outperforming Gramian Angular Difference Fields (96.88%), Gramian Angular Summation Fields (90.20%), and Markov Transition Fields (72.51%).
• The proposed RP+MobileNet-v3 model demonstrated superior performance in diagnosis accuracy, parameter count, and GPU usage compared to other advanced models (MobileNet-v3 small/large, ResNet-18, DenseNet121) and state-of-the-art approaches.
• The method effectively overcame overfitting and enhanced anti-noise performance.

Conclusions:

• The RP+MobileNet-v3 model is highly effective for bearing fault diagnosis under variable-speed conditions.
• This approach offers a lighter model with higher diagnostic accuracy and fewer parameters.
• Recurrence Plot coding is a suitable method for characterizing variable-speed fault features.