Intelligent recognition of composite material damage based on deep learning and infrared testing

Area of Science:

• Materials Science
• Aerospace Engineering
• Artificial Intelligence

Background:

• Composite materials are critical for aircraft performance and safety.
• Infrared nondestructive testing (NDT) is vital for detecting damage in these materials.
• Current manual inspection methods are inefficient, and distinguishing diverse damage types from infrared images alone is challenging.

Purpose of the Study:

• To develop an intelligent method for improved damage detection in aircraft composite materials.
• To leverage infrared signals (temporal data) for more precise damage type identification.
• To enhance the efficiency and accuracy of damage detection in composite structures.

Main Methods:

• Development of a novel 1D-YOLOv4 network, incorporating a modified neck and 1D-CNN.
• Application of the algorithm to identify both infrared images and temporal infrared signals from composite materials.
• Comparative analysis of detection performance using original, fitted, first derivative, and second derivative data.

Main Results:

• The proposed 1D-YOLOv4 network achieved a recognition accuracy of 98.3%, an Average Precision (AP) of 91.9%, and a kappa score of 0.997.
• The algorithm demonstrated superior effectiveness compared to YOLOv3, YOLOv4, and YOLOv4+Neck networks.
• First derivative data yielded the best detection outcomes among the analyzed data types.

Conclusions:

• The developed 1D-YOLOv4 network significantly improves the accuracy and efficiency of damage detection in aircraft composite materials.
• Integrating temporal infrared signal analysis enhances the ability to distinguish between different damage types.
• The findings suggest a promising intelligent approach for NDT in aerospace applications.