Fault Diagnosis Method for Mechanical Components Fusing RPM, DAM-IResNet, and Transfer Learning.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Machine fault diagnosis faces challenges with scarce data and poor generalization under varying conditions.
• Existing methods struggle to adapt effectively to new operational environments with limited samples.

Purpose of the Study:

• To develop a novel fault diagnosis method robust to data scarcity and working condition variations.
• To enhance feature representation and learning for improved diagnostic accuracy and generalization.

Main Methods:

• Integration of Relative Position Matrix (RPM) for signal-to-image conversion, a Downsampling Attention Module (DAM) for feature compression, and an Improved Residual Network (IResNet) for deep learning.
• Application of transfer learning to adapt models from data-rich to data-scarce domains.

Main Results:

• The proposed RPM-DAM-IResNet framework achieved over 99.5% accuracy on bearing and gear datasets.
• Demonstrated 100% accuracy in key small-sample transfer learning tasks.
• Outperformed existing state-of-the-art methods in accuracy and robustness.

Conclusions:

• The unified RPM-DAM-IResNet framework with a targeted small-sample transfer strategy offers superior accuracy and robustness for fault diagnosis.
• The method effectively addresses challenges of limited fault data and cross-condition generalization.