Embedded deep learning system for reliable operator support in TFM-based weld inspection
e-Journal of Nondestructive Testing · 2026-07 · 31 (7)
DOI: 10.58286/33336 ↗Kurz gefasst
Fachartikel in der Zeitschrift e-Journal of Nondestructive Testing, 2026-07. Verfasst von Robin Guyon und 5 weiteren Autor:innen.
Abstract (Originalsprache)
In the field of pipeline weld inspection, ultrasonic techniques, particularly Total Focusing Method (TFM) imaging, are widely used for their ability to provide high-resolution images. However, in industrial contexts, the analysis of these images remains a major challenge due to the large volume of mostly pristine data that experts must review according to a procedure, making inspections time-consuming and costly. Artificial Intelligence offers a promising approach to reduce this workflow by filtering pristine data and providing additional indications regarding the presence and nature of defects. Several challenges nevertheless arise: first, the detection process must tolerate no false negatives, as an image containing a defect must never be classified as pristine. Second, as is common in non-destructive testing (NDT), access to sufficiently large annotated defect datasets is limited. Finally, a third challenge is achieving real-time processing, ensuring that the analysis is delivered instantly at the speed of manual inspection. In this study, we present a deep learning model embedded within an in-house developed acquisition system. Trained on augmented simulated data, it significantly reduces the number of pristine images requiring review while meeting the reactivity constraints of manual inspection.
Zitieren
APA
Guyon, R., Newson, M., Roué, D., Gervais, A., Miorelli, R., & Fisher, C. (2026). Embedded deep learning system for reliable operator support in TFM-based weld inspection. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33336
Vancouver
Guyon R, Newson M, Roué D, Gervais A, Miorelli R, Fisher C. Embedded deep learning system for reliable operator support in TFM-based weld inspection. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33336
BibTeX
@article{Guyon2026,
author = {Guyon, Robin and Newson, Matthew and Roué, David and Gervais, Auguste and Miorelli, Roberto and Fisher, Clément},
title = {Embedded deep learning system for reliable operator support in TFM-based weld inspection},
journal = {e-Journal of Nondestructive Testing},
year = {2026},
volume = {31},
number = {7},
doi = {10.58286/33336}
}
EUVIMED