AI model overcomes spatial overfitting for accurate defect depth measurement

Researchers have developed a novel spatio-temporal decoupling architecture to improve the accuracy of measuring subsurface delamination depth in carbon-fiber-reinforced polymers using optical pulsed thermography. This method separates the localization of defects from the measurement of their depth, addressing the challenge of spatial dataset bias where models might memorize calibration defect geometry instead of learning the physical relationship between thermal diffusion and depth. By using regularized XGBoost with L1/L2 penalties, the system achieved a mean absolute error of 0.056 mm and root mean square error of 0.085 mm, enabling the generation of three-dimensional defect models. AI

IMPACT Improves accuracy in material defect detection, potentially enhancing structural integrity assessments.

RANK_REASON Academic paper detailing a new methodology and model performance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Advanced Multi-Layer Perceptron
• carbon-fiber-reinforced polymer
• Delaunay
• Gradient boosting machines
• L2
• Optical pulsed thermography
• random forest
• XGBoost

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →