A Hybrid Deep Learning Framework for Fault Diagnosis in Milling Machines

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Industrial monitoring faces challenges with noisy vibration signals, limited fault data, and operational variability.
• Accurate fault diagnosis in milling tools is crucial for preventing failures and optimizing production.

Purpose of the Study:

• To develop a robust hybrid fault-diagnosis framework for milling cutting tools.
• To address challenges of noise, limited data, and operational variability in industrial settings.

Main Methods:

• Signal preprocessing to remove baseline drift and enhance signal-to-noise ratio.
• Construction of logarithmic continuous wavelet scalograms for time-frequency analysis.
• Application of Canny edge operator for feature enhancement and artifact suppression.
• Dual-branch encoder for discriminative feature representation capturing local and long-range dependencies.
• Ensemble decision mechanism with validation-guided weights for reliable fault identification.

Main Results:

• Achieved 99.78% accuracy in experimental validation on real-world cutting tool data.
• Demonstrated strong resilience to environmental noise and consistent performance across variable operating conditions.
• Framework proved lightweight, scalable, and deployable for practical industrial applications.

Conclusions:

• The proposed hybrid framework offers a practical and effective solution for high-precision tool fault diagnosis.
• The methodology successfully overcomes limitations of noisy data and limited fault labels in industrial environments.
• The system ensures reliable, robust, and stable fault identification under diverse and challenging conditions.