EUVIMEDEUVIMED entdecken
Crossref

Quantifying corrosion progression through text–image embeddings with CLIP

Ramon Helwing, Frank Walther

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

DOI: 10.58286/32998 ↗

Kurz gefasst

Fachartikel in der Zeitschrift e-Journal of Nondestructive Testing, 2026-07. Verfasst von Ramon Helwing und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Monitoring corrosion in industrial components is essential for ensuring structural integrity, operational safety, and economic efficiency. However, the nonlinear nature of corrosion leads to irregular visual and physical surface changes that conventional image-processing methods struggle to capture. Supervised deep-learning approaches require extensive annotated datasets, often unavailable for diverse material conditions. This study uses the pre-trained CLIP (Contrastive Language–Image Pre-training) model for semantic analysis of nonlinear corrosion progression. By embedding images and descriptive text into a shared latent space, the method detects subtle optical variations, identifies characteristic corrosion features, and derives a directional metric representing progression from initiation to advanced states. Optical analyses of corrosion test plates provide a controlled dataset for evaluating surface evolution and validating the semantic similarity metrics. The results indicate that latent space analysis offers a practical, data-driven approach to automated corrosion monitoring.

Zitieren

APA

Helwing, R., & Walther, F. (2026). Quantifying corrosion progression through text–image embeddings with CLIP. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/32998

Vancouver

Helwing R, Walther F. Quantifying corrosion progression through text–image embeddings with CLIP. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/32998

BibTeX
@article{Helwing2026,
  author = {Helwing, Ramon and Walther, Frank},
  title = {Quantifying corrosion progression through text–image embeddings with CLIP},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {7},
  doi = {10.58286/32998}
}
Kostenlos starten →