EUVIMEDEUVIMED entdecken
Crossref

Optimising Synthetic Data for Machine Learning Applications on Surface Breaking Thermal Fatigue Cracks

James Gaffney, Thomas Beckingham, Daniel Colquitt, Will Daniels, Stewart G. Haslinger

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

DOI: 10.58286/33339 ↗

Kurz gefasst

Fachartikel in der Zeitschrift e-Journal of Nondestructive Testing, 2026-07. Verfasst von James Gaffney und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Non-destructive evaluation (NDE) is used throughout industry to ensure the integrity of safety critical components. NDE practitioners use ultrasonic testing (UT) to find, size and characterise defects, such as thermal fatigue (TF) surface-breaking cracks in the nuclear sector. Pulse-echo is often utilised because it is relatively quick and simple to implement. There is an ever-increasing quantity of data to be interpreted. As such, machine learning (ML) is being explored as a tool to aid interpretation. ML is well suited to analysing this type of data, but critical to its accuracy is a high-quality dataset to train from. Unfortunately, there is a paucity of labelled surface-breaking crack data available for training. To overcome this hurdle, a high fidelity yet entirely synthetic dataset is presented here. The accuracy of the A-scan simulations is informed by insight and measurements from practical inspection. This dataset leads to accurate height predictions (with a mean average error of 0.19mm) when tested on measured data of crack heights between 0.96mm and 4mm. The specificity of the CNN will be discussed in the context of the training data; can this CNN perform well on measurement setups (e.g. transformer characteristics, alternative materials) not included within the training data.

Zitieren

APA

Gaffney, J., Beckingham, T., Colquitt, D., Daniels, W., & Haslinger, S. G. (2026). Optimising Synthetic Data for Machine Learning Applications on Surface Breaking Thermal Fatigue Cracks. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33339

Vancouver

Gaffney J, Beckingham T, Colquitt D, Daniels W, Haslinger SG. Optimising Synthetic Data for Machine Learning Applications on Surface Breaking Thermal Fatigue Cracks. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33339

BibTeX
@article{Gaffney2026,
  author = {Gaffney, James and Beckingham, Thomas and Colquitt, Daniel and Daniels, Will and Haslinger, Stewart G.},
  title = {Optimising Synthetic Data for Machine Learning Applications on Surface Breaking Thermal Fatigue Cracks},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {7},
  doi = {10.58286/33339}
}
Kostenlos starten →