A Three-Branch Time-Frequency Feature Fusion Method Based on Terahertz Signals for Identifying Delamination Defects...

Area of Science:

• Materials Science
• Non-Destructive Testing
• Signal Processing

Background:

• Delamination defects in composite materials pose significant risks to equipment safety, particularly in aerospace.
• Terahertz non-destructive testing (TNDT) is effective for internal structure analysis but current methods struggle to integrate time and frequency domain data.
• Existing TNDT approaches often rely on single domain features, limiting comprehensive defect detection.

Purpose of the Study:

• To develop an advanced TNDT method for precise identification of delamination defects in composite materials.
• To overcome the limitations of single-domain feature analysis in current terahertz detection techniques.
• To improve the accuracy and generalization ability of defect identification models for layered structures.

Main Methods:

• A novel Time-Frequency Feature-fusion Network (TFFN) with a three-branch architecture was proposed.
• The network extracts local time-frequency features, focuses on damage-sensitive frequency bands using attention mechanisms, and integrates features via Manifold Mixup.
• Adaptive fusion of extracted features is achieved through a cross-branch attention mechanism for final defect classification.

Main Results:

• The TFFN achieved high accuracies of 98.40% on glass fiber reinforced polymer (GFRP) and 98.63% on quartz fiber reinforced polymer (QFRP) datasets.
• The proposed method surpassed existing techniques by 2% and 1.25% on GFRP and QFRP datasets, respectively.
• Significant improvements in both defect identification accuracy and model generalization for layered structures were demonstrated.

Conclusions:

• The TFFN effectively integrates complementary time and frequency domain information for superior delamination defect detection in composite materials.
• This advanced TNDT approach enhances the safety and reliability of critical equipment, especially in the aerospace industry.
• The study highlights the potential of feature fusion techniques in advancing non-destructive testing methodologies.