EUVIMEDEUVIMED entdecken
Crossref

High-Resolution Imaging for Additive Manufactured Metals Using Laser Ultrasonics With Hybrid GNN-OCSVM Algorithm

Dan Chen, Xunfa Kang, Tao Gong, Anmin Yin, Shifeng Guo, Zhaofeng Liang, Hui Zhou, Min Yang, Wenxing Zhang

IEEE Transactions on Instrumentation and Measurement · 2025 · 74 · 1-11

DOI: 10.1109/tim.2025.3627338 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2025. Verfasst von Dan Chen und 8 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 40 Literaturangaben. Gefördert von: National Key Research and Development Program of China, National Natural Science Foundation of China, Shenzhen Science and Technology Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The coupling of multimodal ultrasonic waves generated by pulsed lasers, combined with wave scattering caused by surface roughness and randomly distributed pores, presents significant challenges for achieving high-resolution defect imaging in metal additive manufacturing (MAM) components using conventional ultrasonic methods. This article proposes a probabilistic imaging strategy based on a hybrid deep learning algorithm, namely a graph neural network integrated with a one-class support vector machine (GNN-OCSVM), for subsurface defect characterization using laser ultrasonics. The graph neural network (GNN)-based denoising module enhances the signal-to-noise ratio (SNR) by aggregating correlated features through topological relationships among ultrasonic signals. Subsequently, the one-class support vector machine (OCSVM) algorithm estimates the probability of defect presence from the denoised signals, enabling probabilistic reconstruction of C-scan images of subsurface defects. Laser ultrasonic experiments were conducted on AlSi10Mg and TC4 MAM specimens containing subsurface defects of varying sizes and geometries. The experimental results demonstrate that the GNN-OCSVM algorithm: 1) achieves superior defect boundary definition compared to scanning acoustic microscopy (SAM); 2) maintains a relative dimensional error of merely 2%–6% when benchmarked against X-ray imaging; and 3) outperforms other anomaly detection algorithms (ADAs), including Deep Isolation Forest, K-Means and TimesNet, in both defect shape reconstruction accuracy and size measurement precision. These findings demonstrate that the proposed AI algorithm, combined with laser ultrasonics, provides enhanced imaging accuracy for subsurface defect detection under challenging multimodal wave coupling conditions, offering a viable solution for nondestructive evaluation of additive-manufactured components.

Förderung

  • National Key Research and Development Program of China
  • National Natural Science Foundation of China
  • Shenzhen Science and Technology Program
  • Special Projects in Key Areas of Guangdong Provincial Department of Education
  • Natural Science Foundation of Inner Mongolia Autonomous Region of China
  • Department of Education of Guangdong Province
  • Research Fund of Shenzhen Polytechnic University

Zitieren

APA

Chen, D., Kang, X., Gong, T., Yin, A., Guo, S., Liang, Z., Zhou, H., Yang, M., & Zhang, W. (2025). High-Resolution Imaging for Additive Manufactured Metals Using Laser Ultrasonics With Hybrid GNN-OCSVM Algorithm. IEEE Transactions on Instrumentation and Measurement, 74, 1-11. https://doi.org/10.1109/tim.2025.3627338

Vancouver

Chen D, Kang X, Gong T, Yin A, Guo S, Liang Z, et al. High-Resolution Imaging for Additive Manufactured Metals Using Laser Ultrasonics With Hybrid GNN-OCSVM Algorithm. IEEE Transactions on Instrumentation and Measurement. 2025;74:1-11. doi:10.1109/tim.2025.3627338

BibTeX
@article{Chen2025,
  author = {Chen, Dan and Kang, Xunfa and Gong, Tao and Yin, Anmin and Guo, Shifeng and Liang, Zhaofeng and Zhou, Hui and Yang, Min and Zhang, Wenxing},
  title = {High-Resolution Imaging for Additive Manufactured Metals Using Laser Ultrasonics With Hybrid GNN-OCSVM Algorithm},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2025},
  volume = {74},
  pages = {1--11},
  doi = {10.1109/tim.2025.3627338}
}
Kostenlos starten →