EUVIMEDScopri EUVIMED
Crossref

Defect segmentation using deep learning to reduce the complexity of pulsed thermography data

Julian Grimm, Günther Mayr, Gerald Zauner, Gernot Mayr

e-Journal of Nondestructive Testing · 2026-07 · 31 (7)

DOI: 10.58286/33418 ↗

At a glance

Journal article in e-Journal of Nondestructive Testing, 2026-07. Written by Julian Grimm and 3 co-authors.

AbstractDati bibliograficiCiteRelated studies

Abstract (original language)

High-rate production of safety-critical composite parts requires fast and reliable methods for non-destructive testing (NDT). Under current regulations, fully autonomous decisions are not permitted, so manual inspections are still very common. As a result, human inspectors are faced with the challenge of inspecting large volumes of data, which is time consuming and prone to human errors. In this work, a data-driven framework for AI-assisted defect localization in composite parts is presented. The parts are inspected using pulsed thermography and relevant features are extracted from the acquired data using resampling and mathematical transformations. Then the result is analyzed by a semantic segmentation model that produces a single pixel-wise defect map. The model is trained on a large set of synthetically generated measurements using a physics-based simulation model. The generalization capability of the model is then evaluated by directly applying it to real measurement data of a demonstration part with embedded FODs (foreign object debris), i.e. performing zero-shot sim-to-real transfer. The model achieved high evaluation metrics (recall, F2-Score, etc.), indicating that simulation-based data generation is a promising strategy for obtaining accurate defect segmentation models for active thermography.

Cite

APA

Grimm, J., Mayr, G., Zauner, G., & Mayr, G. (2026). Defect segmentation using deep learning to reduce the complexity of pulsed thermography data. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33418

Vancouver

Grimm J, Mayr G, Zauner G, Mayr G. Defect segmentation using deep learning to reduce the complexity of pulsed thermography data. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33418

BibTeX
@article{Grimm2026,
  author = {Grimm, Julian and Mayr, Günther and Zauner, Gerald and Mayr, Gernot},
  title = {Defect segmentation using deep learning to reduce the complexity of pulsed thermography data},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {7},
  doi = {10.58286/33418}
}
Start for free →