AI-Based Signal Analysis and Flaw Detection for Eddy Current Testing of Heat Exchanger Tubes
e-Journal of Nondestructive Testing · 2026-07 · 31 (7)
DOI: 10.58286/32900 ↗Kurz gefasst
Fachartikel in der Zeitschrift e-Journal of Nondestructive Testing, 2026-07. Verfasst von Ryohei Fujita und 2 weiteren Autor:innen.
Abstract (Originalsprache)
In Eddy Current Testing (ECT) for heat exchanger tubes, manual analysis is complex and time-consuming due to the need to distinguish flaw signals from various pseudo-signals, such as those from baffle plates or scales. In this study, an AI model for automated analysis was developed to improve inspection efficiency and reliability. We proposed a 1D-ResNet architecture that processes multiple multi-frequency ECT signals (differential and absolute) simultaneously. A key part of our approach was constructing a high-quality dataset by employing hard-negative mining, intentionally sampling various pseudo-signals as "normal" data. Finally, class weighting was applied to address the significant dataset imbalance. The trained model demonstrated excellent performance on an unseen test dataset of field data, achieving 97% Recall and 88% Precision for flaw detection. This study demonstrates that our 1D-ResNet model, combined with specialized dataset engineering, is a highly effective approach for automated ECT signal analysis.
Zitieren
APA
Fujita, R., Sanokawa, M., & Yamabe, S. (2026). AI-Based Signal Analysis and Flaw Detection for Eddy Current Testing of Heat Exchanger Tubes. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/32900
Vancouver
Fujita R, Sanokawa M, Yamabe S. AI-Based Signal Analysis and Flaw Detection for Eddy Current Testing of Heat Exchanger Tubes. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/32900
BibTeX
@article{Fujita2026,
author = {Fujita, Ryohei and Sanokawa, Misaki and Yamabe, Shota},
title = {AI-Based Signal Analysis and Flaw Detection for Eddy Current Testing of Heat Exchanger Tubes},
journal = {e-Journal of Nondestructive Testing},
year = {2026},
volume = {31},
number = {7},
doi = {10.58286/32900}
}
EUVIMED