EUVIMEDSkopri EUVIMED
Crossref

Generating Synthetic Ultrasonic Testing Data with Deep Learning: Comparing Simulated and Deep Learning-Based Flaw Responses

Caleb Watson

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

DOI: 10.58286/33360 ↗

At a glance

Journal article in e-Journal of Nondestructive Testing, 2026-07. Written by Caleb Watson.

AstrattDettalji biblijografiċiCiteRelated studies

Abstract (original language)

A study was performed on deep learning techniques capable of synthesizing a subset of NDE data; ultrasonic testing (UT). Synthesizing UT datasets can assist in the development of NDE training activities, NDE Digital Twins, assisted analysis tools, and other advanced techniques across many industries. Using 215,536 ultrasonic amplitude waveforms (A-scans) collected from a 304 stainless-steel component containing 10 Electrical Discharge Machined (EDM) notches of differing heights, a Generative Adversarial Network (GAN) based synthesizing technique was studied. This technique yielded both quantitative and qualitative results of deep-learning-based generative models that were compared against UT data produced by an NDE dedicated simulation software (CIVA) and experimental data. The results show that synthetic notches can be generated with a root-mean-squared error of 3.7dB in amplitude difference when compared to experimental notches. Additionally, the synthesized notch responses displayed higher statistical accuracy in comparison to the CIVA simulation responses in terms of absolute amplitude deviation from data collected with physical UT instrumentation in a lab setting, while falling short in time of flight deviation. Improvements and stabilization can likely be implemented through improved synthesizing techniques and further research.

Cite

APA

Watson, C. (2026). Generating Synthetic Ultrasonic Testing Data with Deep Learning: Comparing Simulated and Deep Learning-Based Flaw Responses. e-Journal of Nondestructive Testing, 31(7). https://doi.org/10.58286/33360

Vancouver

Watson C. Generating Synthetic Ultrasonic Testing Data with Deep Learning: Comparing Simulated and Deep Learning-Based Flaw Responses. e-Journal of Nondestructive Testing. 2026;31(7). doi:10.58286/33360

BibTeX
@article{Watson2026,
  author = {Watson, Caleb},
  title = {Generating Synthetic Ultrasonic Testing Data with Deep Learning: Comparing Simulated and Deep Learning-Based Flaw Responses},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {7},
  doi = {10.58286/33360}
}
Start for free →