Spectral feature extraction and ensemble learning for multiclass aircraft damage identification.

Area of Science:

• Aerospace Engineering
• Computer Science
• Materials Science

Background:

• Conventional aircraft surface damage inspection methods are often limited in detecting subtle degradation.
• Various forms of surface damage, including corrosion, burn marks, and paint peeling, pose significant safety risks.
• Advanced imaging and machine learning offer potential for more effective detection.

Purpose of the Study:

• To develop and validate a robust methodology for identifying diverse aircraft surface damage types.
• To leverage hyperspectral imaging and ensemble machine learning for enhanced damage classification.
• To create a scalable, non-contact inspection solution for aerospace maintenance.

Main Methods:

• Collected hyperspectral intensity data from over 500 real and lab-induced aircraft surface samples.
• Engineered handcrafted features across spectral, statistical, and frequency domains.
• Developed a soft voting ensemble model combining Random Forest, XGBoost, and SVM classifiers.

Main Results:

• Achieved a peak classification accuracy of 92.6% for ten different aircraft surface damage types.
• Demonstrated high accuracy across various damage classes, outperforming conventional techniques.
• Validated the effectiveness of the ensemble model in classifying complex surface degradation patterns.

Conclusions:

• The hyperspectral imaging and ensemble machine learning pipeline provides an accurate and efficient method for aircraft surface damage identification.
• The developed system is suitable for real-time, non-contact, and scalable inspection workflows.
• This methodology shows strong potential for integration into automated drone-based or robotic inspection systems for aerospace maintenance.