A Neural-Assisted Triangulation Framework for Broadband Optical Monitoring of Multilayer Coatings
Coatings · 26.10.2026 · 16 (11) · 1276
DOI: 10.3390/coatings16111276 ↗Kurz gefasst
Fachartikel in der Zeitschrift Coatings, online erschienen am 26.10.2026. Verfasst von Yiming Guo und 5 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: China Postdoctoral Science Foundation, Natural Science Basic Research Program in Shaanxi Province.
Abstract (Originalsprache)
Broadband optical monitoring provides rich spectral information for the fabrication of multilayer optical coatings, but conventional online characterization generally relies on iterative thickness re-estimation, whose computational burden increases when information from multiple deposition stages is incorporated. To address this issue, this work proposes a neural-assisted triangulation framework that organizes multistage residual spectra within a fixed-length historical observation window, thereby maintaining a fixed-dimensional estimator input. The framework further reformulates conventional iterative online thickness re-estimation as direct neural prediction of incremental thickness-state corrections. A shared one-dimensional convolutional encoder extracts spectral features from the residual history, which are then fused with the current pre-update thickness state for thickness-state updating. The updated thickness estimates are subsequently used to reconstruct the multilayer optical model and calculate the reference spectrum for the next deposition stage. The estimator is trained offline using synthetic thickness-state perturbations and a physics-guided objective that combines direct state supervision with spectral-reconstruction and spectral-gradient constraints, while only neural inference is required during online deposition. The proposed framework was implemented on an ion-beam sputtering platform and evaluated through three consecutive fabrication runs of an 82-layer long-wave-pass filter. All three fabricated filters satisfied the specified spectral requirements, with average stopband transmittance ranging from 0.090% to 0.155% and average passband transmittance ranging from 95.008% to 95.105%. The mean cutoff wavelength was 712.461 nm, with a standard deviation of 0.278 nm and a maximum inter-run difference of 0.555 nm. These results demonstrate the practical feasibility of the proposed framework for broadband monitoring of high-layer-count optical coatings.
Förderung
- China Postdoctoral Science Foundation
- Natural Science Basic Research Program in Shaanxi Province
Zitieren
APA
Guo, Y., Ji, X., Ji, M., Pei, Y., Wang, B., & He, Z. (2026). A Neural-Assisted Triangulation Framework for Broadband Optical Monitoring of Multilayer Coatings. Coatings, 16(11), 1276. https://doi.org/10.3390/coatings16111276
Vancouver
Guo Y, Ji X, Ji M, Pei Y, Wang B, He Z. A Neural-Assisted Triangulation Framework for Broadband Optical Monitoring of Multilayer Coatings. Coatings. 2026;16(11):1276. doi:10.3390/coatings16111276
BibTeX
@article{Guo2026,
author = {Guo, Yiming and Ji, Xiaohan and Ji, Ming and Pei, Yuhui and Wang, Bo and He, Zhuangzhuang},
title = {A Neural-Assisted Triangulation Framework for Broadband Optical Monitoring of Multilayer Coatings},
journal = {Coatings},
year = {2026},
volume = {16},
number = {11},
pages = {1276},
doi = {10.3390/coatings16111276}
}
EUVIMED