EUVIMEDExplore EUVIMED
Crossref

Defect Classification in Wire Arc Additive Manufacturing via a Semi‑Supervised Teacher–Student Model

Maedeh Gourakani, Carmelo Mineo, Donatella Cerniglia, Andrea Bertazzo, Federica Tiso, Giovanni Avallone

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

DOI: 10.58286/33307 ↗

At a glance

Journal article in e-Journal of Nondestructive Testing, 2026-07. Written by Maedeh Gourakani and 5 co-authors.

AbstractBibliographic detailsCiteRelated studies

Abstract (original language)

Wire Arc Additive Manufacturing (WAAM) enables large-scale metal part production, yet process stability and part integrity remain challenging. Optical in-situ monitoring can reveal defects, but manual screening of image streams is impractical at production speed. This work addresses this challenge for critical defects like balling and bead-cut by validating a teacher-student semi-supervised learning framework. A ResNet18 teacher, trained on just 600 labeled images (7.7% of a 7,795-image dataset), generates pseudolabels to train a lightweight MobileNetV2 student for real-time deployment. The methodology incorporates a weighted Kullback-Leibler (KL) divergence loss to counteract class imbalance. The final student model achieved 98.33% classification accuracy with stable convergence. Model reliability was validated using Grad-CAM, which confirmed the model's focus on physically relevant defect features such as the bead's crosssectional profile rather than background noise. This study demonstrates a robust and data-efficient approach, enabling reliable, real-time quality control in WAAM with significantly reduced manual labeling. Future work will include architectural comparisons and further data scarcity analysis.

Cite

APA

Gourakani, M., Mineo, C., Cerniglia, D., Bertazzo, A., Tiso, F., & Avallone, G. (2026). Defect Classification in Wire Arc Additive Manufacturing via a Semi‑Supervised Teacher–Student Model. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33307

Vancouver

Gourakani M, Mineo C, Cerniglia D, Bertazzo A, Tiso F, Avallone G. Defect Classification in Wire Arc Additive Manufacturing via a Semi‑Supervised Teacher–Student Model. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33307

BibTeX
@article{Gourakani2026,
  author = {Gourakani, Maedeh and Mineo, Carmelo and Cerniglia, Donatella and Bertazzo, Andrea and Tiso, Federica and Avallone, Giovanni},
  title = {Defect Classification in Wire Arc Additive Manufacturing via a Semi‑Supervised Teacher–Student Model},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {7},
  doi = {10.58286/33307}
}
Start for free →