Machine learning analysis based on deep learning for fatigue diagnostics in carbon fiber reinforced polymers

Area of Science:

• Materials Science
• Mechanical Engineering
• Artificial Intelligence

Background:

• Fatigue degradation in Carbon Fiber Reinforced Polymer (CFRP) structures presents a significant challenge for long-term structural health monitoring (SHM).
• Reliable methods are needed to detect and classify fatigue states in CFRP composites for ensuring structural integrity.

Purpose of the Study:

• To propose a hybrid deep learning framework for accurate fatigue state classification in CFRP composites.
• To leverage sensor-based monitoring data for advanced SHM applications.

Main Methods:

• A hybrid deep learning framework combining a 1D-CNN for spatial pattern extraction and an xLSTM network for temporal dependency capture.
• Mutual Information-based feature selection and a Bagging-based ensemble classifier for robust fatigue discrimination.

Main Results:

• The proposed framework achieved an average classification accuracy of 99% on the NASA-CFRP dataset.
• Demonstrated effective spatiotemporal feature extraction and fusion for fatigue diagnosis.

Conclusions:

• The developed hybrid deep learning approach offers a reliable solution for fatigue diagnosis in CFRP structures.
• The framework supports maintenance decision-making, with potential applications in membrane-based gas separation systems.