Scientific machine learning: Prediction of freeform microlens shapes
Journal of the European Optical Society-Rapid Publications · 2026 · 22 (2) · 73
DOI: 10.1051/jeos/2026067 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of the European Optical Society-Rapid Publications, online erschienen am 03.11.2026. Verfasst von Annika Dehm und 3 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deutsche Forschungsgemeinschaft, Carl-Zeiss-Stiftung.
Abstract (Originalsprache)
The electrical deformation of liquid polymer droplets offers a promising approach for the fabrication of customized freeform microlenses. This work investigates Scientific Machine Learning models for predicting the resulting lens shapes before UV curing. Three different neural network architectures are compared: a pointwise Multilayer Perceptron, a sequence-wise 1D convolutional U-Net and a self-attention-based Transformer Encoder for sequence regression. Prior knowledge of the geometric microlens boundaries is embedded either as a soft penalty term in the loss function or as a hard boundary constraint through a trial function layer. The sequence-wise architectures show improved prediction accuracy compared with pointwise regression. The best RMSE is obtained with the hard-constrained 1D U-Net (3.1 μm), followed closely by the hard-constrained Transformer (3.3 μm). The results show that the imposed boundary constraints improve physical consistency locally, while future work will include global constraints to further enhance the overall robustness of the predicted shapes.
Förderung
- Deutsche Forschungsgemeinschaft
- Carl-Zeiss-Stiftung
Zitieren
APA
Dehm, A., Fischer, L., Dohmen, M., & Heinrich, A. (2026). Scientific machine learning: Prediction of freeform microlens shapes. Journal of the European Optical Society-Rapid Publications, 22(2), 73. https://doi.org/10.1051/jeos/2026067
Vancouver
Dehm A, Fischer L, Dohmen M, Heinrich A. Scientific machine learning: Prediction of freeform microlens shapes. Journal of the European Optical Society-Rapid Publications. 2026;22(2):73. doi:10.1051/jeos/2026067
BibTeX
@article{Dehm2026,
author = {Dehm, Annika and Fischer, Lucas and Dohmen, Mike and Heinrich, Andreas},
title = {Scientific machine learning: Prediction of freeform microlens shapes},
journal = {Journal of the European Optical Society-Rapid Publications},
year = {2026},
volume = {22},
number = {2},
pages = {73},
doi = {10.1051/jeos/2026067}
}
EUVIMED