Unsupervised Segmentation of Subsurface Delaminations in Large‐Tow Carbon Fiber‐Reinforced Polymer Composites Using Long‐Pulse Thermography

ABSTRACT Large‐tow carbon fiber‐reinforced polymer (CFRP) composites offer cost and manufacturing advantages for large‐scale structural applications; however, tow‐induced structural texture and anisotropic heat diffusion generate strong directional and grid‐like thermal backgrounds that obscure weak delamination responses in long‐pulse thermography. This study proposes an unsupervised framework for segmenting subsurface delaminations in large‐tow CFRP laminates. Statistical background subtraction is first used to attenuate initial thermal‐field nonuniformity. A component‐difference prior derived from principal component analysis and independent component analysis (PCA–ICA) is then constructed to suppress redundant thermal background and enhance sparse, non‐Gaussian defect responses. Row‐ and column‐wise median modeling removes tow‐induced stripe components, while multiscale saliency separates localized dark‐spot responses from slowly varying thermal trends. Finally, dual‐quantile geodesic reconstruction expands high‐confidence seeds within a relaxed candidate domain, thereby recovering diffuse defect boundaries while limiting background leakage. Experiments on large‐tow CFRP laminates containing embedded polytetrafluoroethylene inserts show that the proposed method produces more continuous defect masks and fewer background false positives than Otsu thresholding, adaptive thresholding, conventional region growing, and direct principal‐component segmentation. The proposed framework achieves an intersection over union of 0.8717, a Dice coefficient of 0.9315, a precision of 0.9450, and an area error of −2.83% without requiring pixel‐level training annotations. These results demonstrate the potential of the framework as an interpretable, label‐free approach for thermographic inspection of large‐tow CFRP composites.

Authors

Institutions

Publication Details

Journal
Polymer Composites
Published
2026-11-17
DOI
https://doi.org/10.1002/pc.71846
Primary Topic
Thermography and Photoacoustic Techniques
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Unsupervised Segmentation of Subsurface Delaminations in Large‐Tow Carbon Fiber‐Reinforced Polymer Composites Using Long‐Pulse Thermography

Shuang Yan Yang, Jun Wang, Kaiwen Chen, Lihang Guo et al.
2 links
Polymer Composites
Thermography and Photoacoustic Techniques
article

Unsupervised Segmentation of Subsurface Delaminations in Large‐Tow Carbon Fiber‐Reinforced Polymer Composites Using Long‐Pulse Thermography

Shuang Yan Yang, Jun Wang, Kaiwen Chen, Lihang Guo, Yujie Wu, Xi Chen, Jinfeng Hu
article en

Abstract

ABSTRACT Large‐tow carbon fiber‐reinforced polymer (CFRP) composites offer cost and manufacturing advantages for large‐scale structural applications; however, tow‐induced structural texture and anisotropic heat diffusion generate strong directional and grid‐like thermal backgrounds that obscure weak delamination responses in long‐pulse thermography. This study proposes an unsupervised framework for segmenting subsurface delaminations in large‐tow CFRP laminates. Statistical background subtraction is first used to attenuate initial thermal‐field nonuniformity. A component‐difference prior derived from principal component analysis and independent component analysis (PCA–ICA) is then constructed to suppress redundant thermal background and enhance sparse, non‐Gaussian defect responses. Row‐ and column‐wise median modeling removes tow‐induced stripe components, while multiscale saliency separates localized dark‐spot responses from slowly varying thermal trends. Finally, dual‐quantile geodesic reconstruction expands high‐confidence seeds within a relaxed candidate domain, thereby recovering diffuse defect boundaries while limiting background leakage. Experiments on large‐tow CFRP laminates containing embedded polytetrafluoroethylene inserts show that the proposed method produces more continuous defect masks and fewer background false positives than Otsu thresholding, adaptive thresholding, conventional region growing, and direct principal‐component segmentation. The proposed framework achieves an intersection over union of 0.8717, a Dice coefficient of 0.9315, a precision of 0.9450, and an area error of −2.83% without requiring pixel‐level training annotations. These results demonstrate the potential of the framework as an interpretable, label‐free approach for thermographic inspection of large‐tow CFRP composites.

Polymer Composites
Wuhan University of Technology (CN)
Justis- og beredskapsdepartementet
Industry, innovation and infrastructure
Thermography and Photoacoustic Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon