EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

WeldVGG: A VGG-Inspired Deep Learning Model for Weld Defect Classification from Radiographic Images with Visual Interpretability

Gabriel López, Pablo Duque Ramírez iD, Emanuel Vega iD, Felix Pizarro, Joaquin Toro, Carlos Parra iD

Sensors · 06.10.2025 · 25 (19) · 6183

DOI: 10.3390/s25196183 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 06.10.2025. Verfasst von Gabriel López und 5 weiteren Autor:innen. Beteiligte Einrichtung: Escuela de Ingeniería Informática, Pontificia Universidad Católica de Valparaíso (PUCV), Av. Brasil 2241, Valparaíso 2362807, Chile. Die Arbeit wurde bisher 16-mal zitiert (Crossref) und nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: UNIVERSIDAD TÉCNICA FEDERICO SANTA MARÍA. PROYECTO DE INNOVACIÓN-03-2025. Departamento de Mecánica, UNIVERSIDAD TÉCNICA FEDERICO SANTA MARÍA.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Visual inspection remains a cornerstone of quality control in welded structures, yet manual evaluations are inherently constrained by subjectivity, inconsistency, and limited scalability. This study presents WeldVGG, a deep learning-based visual inspection model designed to automate weld defect classification using radiographic imagery. The proposed model is trained on the RIAWELC dataset, a publicly available collection of X-ray weld images acquired in real manufacturing environments and annotated across four defect conditions: cracking, porosity, lack of penetration, and no defect. RIAWELC offers high-resolution imagery and standardized class labels, making it a valuable benchmark for defect classification under realistic conditions. To improve trust and explainability, Grad-CAM++ is employed to generate class-discriminative saliency maps, enabling visual validation of predictions. The model is rigorously evaluated through stratified cross-validation and benchmarked against traditional machine learning baselines, including SVC, Random Forest, and a state-of-the-art architecture, MobileNetV3. The proposed model achieves high classification accuracy and interpretability, offering a practical and scalable solution for intelligent weld inspection. Furthermore, to prove the model’s ability to generalize, a test on the GDXray was performed, yielding positive results. Additionally, a Wilcoxon signed-rank test was conducted separately to assess statistical significance between model performances.

Schlagwörter

Stichwörter der Autor:innen

Grad-CAM++convolutional neural networkindustrial AIinterpretabilityradiographic inspectionvisual testingweld defect classificationweld inspection

Autor:innen und Institutionen

  • Escuela de Ingeniería Informática, Pontificia Universidad Católica de Valparaíso (PUCV), Av. Brasil 2241, Valparaíso 2362807, Chile.
  • Departamento de Mecánica, Universidad Técnica Federico Santa María, Av. Federico Sta. María 6090, Viña del Mar 2520000, Chile.

ORCID-Profile: Pablo Duque Ramírez, Emanuel Vega, Carlos Parra

Förderung

  • UNIVERSIDAD TÉCNICA FEDERICO SANTA MARÍA. PROYECTO DE INNOVACIÓN-03-2025. Departamento de Mecánica
  • UNIVERSIDAD TÉCNICA FEDERICO SANTA MARÍA

Zitieren

APA

López, G., Ramírez, P. D., Vega, E., Pizarro, F., Toro, J., & Parra, C. (2025). WeldVGG: A VGG-Inspired Deep Learning Model for Weld Defect Classification from Radiographic Images with Visual Interpretability. Sensors, 25(19), 6183. https://doi.org/10.3390/s25196183

Vancouver

López G, Ramírez PD, Vega E, Pizarro F, Toro J, Parra C. WeldVGG: A VGG-Inspired Deep Learning Model for Weld Defect Classification from Radiographic Images with Visual Interpretability. Sensors. 2025;25(19):6183. doi:10.3390/s25196183

BibTeX
@article{Lpez2025,
  author = {López, Gabriel and Ramírez, Pablo Duque and Vega, Emanuel and Pizarro, Felix and Toro, Joaquin and Parra, Carlos},
  title = {WeldVGG: A VGG-Inspired Deep Learning Model for Weld Defect Classification from Radiographic Images with Visual Interpretability},
  journal = {Sensors},
  year = {2025},
  volume = {25},
  number = {19},
  pages = {6183},
  doi = {10.3390/s25196183}
}
Kostenlos starten →