Kronecker convolutional feature pyramid for fault diagnosis in rolling bearings

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are critical components in rotating machinery, and their failures cause significant operational disruptions and economic losses.
• Current manual fault diagnosis methods are time-consuming and unsuitable for routine maintenance.
• Existing deep learning models struggle with degradation problems and lack multi-scale feature extraction capabilities.

Purpose of the Study:

• To develop a novel deep learning model for autonomous and reliable fault diagnosis in rolling bearings.
• To address the limitations of existing models, including degradation issues and insufficient feature extraction.
• To enhance the efficiency and accuracy of fault diagnosis in industrial rotating machinery.

Main Methods:

• Introduction of a novel three-dimensional (3D) Kronecker convolution feature pyramid (KCFP) model.
• Utilizing 3D Kronecker convolution to replace single dilation rates for efficient data input without domain conversion or pixel loss.
• Employing 3D Feature Selection (3DFSC) for local feature learning.

Main Results:

• The KCFP model achieved 99.6% accuracy on the Paderborn University dataset, outperforming MFF-DRN (98.6%) and standard CNN (97.5%).
• On the MFPT dataset, KCFP achieved 97.6% accuracy, surpassing MFF-DRN (97.0%) and standard CNN (95.7%).
• The proposed model demonstrated enhanced feature representation and classification accuracy while mitigating degradation problems.

Conclusions:

• The novel KCFP model provides a promising solution for reliable and efficient rolling bearing fault diagnosis.
• The 3D Kronecker convolution approach effectively addresses limitations in existing deep learning models for this application.
• The KCFP model shows significant potential for practical implementation in industrial rotating machinery maintenance.