Characterization of defects using ultrasonic arrays: a dynamic classifier approach

Area of Science:

• Nondestructive Evaluation (NDE)
• Ultrasonic Testing
• Materials Science

Background:

• Accurate defect characterization is crucial for assessing material integrity in NDE.
• Distinguishing between crack-like and volumetric defects is a key challenge.
• Ultrasonic array measurements offer rich data for defect analysis.

Purpose of the Study:

• To develop and validate a novel method for accurate defect characterization using ultrasonic scattering matrices.
• To classify defects into distinct categories, differentiating crack-like flaws from volumetric voids.
• To dynamically select optimal pattern classifiers for improved defect assessment.

Main Methods:

• Extraction of ultrasonic scattering matrices from array measurements.
• Feature extraction using Principal Component Analysis (PCA).
• Classification using Quadratic Discriminant Analysis (QDA) and Support Vector Machine (SVM) with dynamic classifier selection based on local accuracy estimation.

Main Results:

• Simulations showed accurate depth measurements for 4441/4636 samples with <10% error.
• Arbitrarily shaped volumetric defects were effectively approximated as ellipses.
• Experimental characterization of six subwavelength scatterers achieved sizing accuracy within 0.14λ.

Conclusions:

• The proposed method effectively characterizes and classifies material defects using ultrasonic scattering matrices.
• Dynamic classifier selection enhances the robustness and accuracy of defect assessment.
• The approach shows significant promise for real-world NDE applications.