A Data Fusion Method for Non-Destructive Testing by Means of Artificial Neural Networks

Area of Science:

• Aeronautics and Materials Science
• Non-Destructive Testing (NDT)
• Machine Learning Applications

Background:

• Aircraft parts require rigorous inspection during manufacturing, assembly, and service to detect diverse defects.
• Traditional Non-Destructive Testing (NDT) methods like Ultrasonic (US) waves and Eddy Current (EC) have limitations: US cannot detect near-surface flaws (blind zone), while EC is limited by the skin effect.

Purpose of the Study:

• To develop a data fusion method combining US and EC NDT techniques to overcome their individual limitations in defect detection.
• To utilize Machine Learning, specifically Artificial Neural Networks (ANNs), for data fusion and defect characterization.

Main Methods:

• Development of a data fusion method using Artificial Neural Networks (ANNs).
• Creation of a simulated training database with US and EC signals for aluminum blocks containing Side Drill Holes (SDHs).
• Experimental validation using an aluminum block with real SDHs, followed by ANN analysis and error estimation.

Main Results:

• Trained ANNs successfully characterized real Side Drill Holes (SDHs) in an aluminum block.
• Estimation errors for flaw depths and radii were, on average, within 4%, with a peak error of 11% for one SDH.
• The combined NDT approach demonstrated effectiveness in defect characterization, validating the data fusion method.

Conclusions:

• The integration of US and EC NDT methods via ANNs offers a complementary approach to enhance defect detection and characterization in aircraft components.
• This data fusion technique effectively addresses the limitations of individual NDT methods, improving inspection accuracy for near-surface and subsurface flaws.
• The study provides experimental validation and quantifies estimation errors, highlighting the practical applicability of the developed method in aerospace NDT.