Physics-Based Data Generation with SVR Surrogate Modeling for PDMS Thin Film Emissivity Prediction
Academic Journal of Applied Sciences · 07.07.2026 · 2 (2) · 28-34
DOI: 10.54097/2as25961 ↗Kurz gefasst
Fachartikel in der Zeitschrift Academic Journal of Applied Sciences, online erschienen am 07.07.2026. Verfasst von Yifan Wu. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
This study combines Transfer Matrix Method (TMM) simulation with Support Vector Regression (SVR) to predict the infrared emissivity of PDMS/SiO₂ thin films across λ=2.0-14.0μm and thicknesses of 100–1000 nm. Physical consistency is enforced during data generation: every TMM spectrum is validated against ε=1-R-T, and points violating |R+T+ε-1|>10-3 are excluded from training. The SVR surrogate, trained on 82,938 validated spectra with an RBF kernel and Halton hyperparameter search, achieves hold-out R2 = 0.959 (RMSE = 0.047, MAE = 0.026) and runs 275× faster than direct TMM evaluation. An overlap-aware multi-region partitioning scheme reduces boundary prediction errors below 1%. The atmospheric-window band (8–13 μm) reaches R2 = 0.985. The complete pipeline, including data, model, and configuration, is publicly archived.
Zitieren
APA
Wu, Y. (2026). Physics-Based Data Generation with SVR Surrogate Modeling for PDMS Thin Film Emissivity Prediction. Academic Journal of Applied Sciences, 2(2), 28-34. https://doi.org/10.54097/2as25961
Vancouver
Wu Y. Physics-Based Data Generation with SVR Surrogate Modeling for PDMS Thin Film Emissivity Prediction. Academic Journal of Applied Sciences. 2026;2(2):28-34. doi:10.54097/2as25961
BibTeX
@article{Wu2026,
author = {Wu, Yifan},
title = {Physics-Based Data Generation with SVR Surrogate Modeling for PDMS Thin Film Emissivity Prediction},
journal = {Academic Journal of Applied Sciences},
year = {2026},
volume = {2},
number = {2},
pages = {28--34},
doi = {10.54097/2as25961}
}
EUVIMED