Advanced fault diagnosis in milling cutting tools using vision transformers with semi-supervised learning and...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing

Background:

• Machine cutting tools (MCT) face challenges in fault diagnosis due to limited labeled data in intelligent manufacturing.
• Existing methods struggle with generalization and accuracy under data constraints.

Purpose of the Study:

• To develop a semi-supervised fault diagnosis framework for MCT using Vision Transformers (ViTs).
• To enhance diagnostic accuracy and generalization, especially with scarce labeled data.
• To provide a scalable and data-efficient solution for predictive maintenance in Industry 4.0.

Main Methods:

• Utilizing time-frequency scalograms from Continuous Wavelet Transform (CWT) of Acoustic Emission (AE) signals as input.
• Employing a Vision Transformer (ViT) architecture for extracting local and global features.
• Integrating pseudo-label generation, uncertainty quantification, and dynamic teacher-student knowledge distillation.
• Implementing an adaptive model refinement loop for continuous improvement.

Main Results:

• Achieved a diagnostic accuracy of 99.68% on real-world milling machine AE data.
• Demonstrated high reliability in identifying subtle fault variations across diverse datasets.
• The proposed framework effectively handles limited labeled data scenarios.

Conclusions:

• The semi-supervised ViT framework offers a robust and accurate solution for MCT fault diagnosis.
• The method is data-efficient, scalable, and interpretable, suitable for Industry 4.0 applications.
• This approach significantly advances predictive maintenance capabilities.