EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

CNN-Assisted Reconstruction of Non-Uniform Temperature Profiles Using Reflected Spectra of a Single Fiber Bragg Grating

Sohyeon Joo, Kyoung Joon Kim

Applied Sciences · 06.11.2026 · 16 (21) · 11010

DOI: 10.3390/app162111010 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 06.11.2026. Verfasst von Sohyeon Joo und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministry of Educatio, Pukyong National University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Under non-uniform temperature conditions, the reflected spectrum of a fiber Bragg grating (FBG) can broaden and distort, making a single Bragg wavelength insufficient to describe the axial temperature distribution. This study investigates a one-dimensional convolutional neural network (1D-CNN) for reconstructing non-uniform temperature profiles from the full reflected spectra of a single uniform FBG. A numerical dataset of 2000 paired spectra and temperature profiles was generated using the transfer matrix method (TMM), which relates forward- and backward-propagating optical field amplitudes across grating segments. Four representative profiles—linear, exponential, hyperbolic, and Gaussian—were considered to represent typical temperature distributions associated with heat conduction and localized heating. The CNN was trained to map the 1001-point normalized reflectivity spectrum to a 200-point axial temperature profile. Across five independent training runs, the CNN achieved a mean absolute percentage error (MAPE) of 11.02 ± 0.60%, a root mean square error (RMSE) of 3.06 ± 0.38 K, and a coefficient of determination (R2) of 0.971 ± 0.005 on the independent test set. Partial least squares (PLS) regression yielded a higher RMSE of 4.08 K. TMM-based iterative optimization reproduced the target profile with errors close to numerical precision but required 28.50 s per spectrum, compared with 6.85 ms for the CNN. Under simulated Gaussian intensity noise, the RMSE of a fixed CNN increased to 4.89 K at 30 dB and 10.73 K at 20 dB, whereas wavelength jitter with a standard deviation of 5 pm yielded an RMSE of 2.76 K. These results suggest that the proposed CNN can reduce online reconstruction time for the investigated temperature profiles, although experimental validation remains necessary.

Förderung

  • Ministry of Educatio
  • Pukyong National University

Zitieren

APA

Joo, S., & Kim, K. J. (2026). CNN-Assisted Reconstruction of Non-Uniform Temperature Profiles Using Reflected Spectra of a Single Fiber Bragg Grating. Applied Sciences, 16(21), 11010. https://doi.org/10.3390/app162111010

Vancouver

Joo S, Kim KJ. CNN-Assisted Reconstruction of Non-Uniform Temperature Profiles Using Reflected Spectra of a Single Fiber Bragg Grating. Applied Sciences. 2026;16(21):11010. doi:10.3390/app162111010

BibTeX
@article{Joo2026,
  author = {Joo, Sohyeon and Kim, Kyoung Joon},
  title = {CNN-Assisted Reconstruction of Non-Uniform Temperature Profiles Using Reflected Spectra of a Single Fiber Bragg Grating},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {21},
  pages = {11010},
  doi = {10.3390/app162111010}
}
Kostenlos starten →