Data-Augmented Manifold Learning Thermography for Defect Detection and Evaluation of Polymer Composites

Area of Science:

• Materials Science
• Non-Destructive Testing
• Composite Materials

Background:

• Infrared thermography is used for non-destructive testing of composite materials.
• Current methods face challenges with limited data and noisy images, hindering defect detection.

Purpose of the Study:

• To introduce a novel Generative Manifold Learning Thermography (GMLT) method for improved defect detection.
• To address limitations in image data and feature extraction for composite material evaluation.

Main Methods:

• Utilized spectral normalized generative adversarial networks for image augmentation and dataset enrichment.
• Applied manifold learning for unsupervised dimensionality reduction of thermal images.
• Employed partial least squares regression for defect visualization and mapping.

Main Results:

• GMLT successfully generated virtual thermal images to enhance the dataset.
• The method achieved effective unsupervised dimensionality reduction and defect visualization.
• Experimental results on carbon fiber-reinforced polymers showed superior performance compared to existing techniques.

Conclusions:

• GMLT offers a significant advancement in detecting subsurface defects in composite materials.
• The proposed technique overcomes limitations of traditional infrared thermography.
• This method provides a robust framework for defect analysis and evaluation.