In precision photonics, even a tiny defect can decide whether a component works as intended. A small scratch, a pinhole in a coating, or a trace of contamination may reduce image quality or shorten the lifetime of an optical system. AI-driven inspection helps find such problems automatically. Instead of relying only on fixed rules, the software learns what a good component normally looks like and then highlights anything unusual. This is especially useful when defective samples are rare and every component is expensive. Transparent or reflective parts, changing light, and extremely small defects are still among the hardest cases [1, 2].
Modern inspection systems are becoming both faster and more flexible. Some use a teacher-student approach: one model learns normal features and teaches a smaller model to reproduce them. A defect appears where the student can no longer match the teacher. EfficientAD has shown that this comparison can be performed in only milliseconds, making it suitable for inspection directly on a production line [3]. Other methods use large vision-language models to recognize defects in products they have not seen before [4]. Researchers also test components from several viewing angles, because a defect may be hidden from a single camera [5].
At Photonics International R&D center, we combine AI with established measurement methods. Our research may involve GaAs wafers, photocathode layers, microchannel plates, coated optics, and precision assemblies. Cameras and microscopes reveal visible defects, while interferometry and spectrophotometry measure changes in shape or optical performance. AI brings these signals together and creates an anomaly map that shows where a possible problem is located. The aim is not simply to label a component as good or bad. We also want to understand the size and importance of the defect, link it to a production step, and explain why the system raised an alert.
The longer-term goal is a closed feedback loop. Inspection results would not only reject faulty parts; they would also help engineers adjust a process before the same defect appears again. Before this can happen, the system must remain reliable across different cameras, lighting conditions, operators, and production batches. This approach supports the European goal of more scalable, resilient, and AI-enabled photonics manufacturing [6]. By developing AI inspection alongside our component programmes, the Center can improve manufacturing readiness without losing the measurement traceability required for scientific, industrial, and defence use.
REFERENCES
[1] Heckler-Kram, L., Neudeck, J.-H., Scheler, U., König, R., & Steger, C. (2026). The MVTec AD 2 dataset: Advanced scenarios for unsupervised anomaly detection. International Journal of Computer Vision, 134, Article 175. https://doi.org/10.1007/s11263-026-02743-0 [2] International Organization for Standardization. (2017). Optics and photonics—Preparation of drawings for optical elements and systems—Part 7: Surface imperfections (ISO Standard No. 10110-7:2017). https://www.iso.org/standard/65444.html
[3] Batzner, K., Heckler, L., & König, R. (2024). EfficientAD: Accurate visual anomaly detection at millisecond-level latencies. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 128–138). https://openaccess.thecvf.com/content/WACV2024/html/Batzner_EfficientAD_Accurate_Visual_Anomaly_Detection_at_Millisecond-Level_Latencies_WACV_2024_paper.html
[4] Zhou, Q., Pang, G., Tian, Y., He, S., & Chen, J. (2024). AnomalyCLIP: Object-agnostic prompt learning for zero-shot anomaly detection. In The Twelfth International Conference on Learning Representations. https://proceedings.iclr.cc/paper_files/paper/2024/file/d7b50b8ac2c781a12f26155f48310d8d-Paper-Conference.pdf
[5] Wang, C., Zhu, W., Gao, B.-B., Gan, Z., Zhang, J., Gu, Z., Qian, S., Chen, M., & Ma, L. (2024). Real-IAD: A real-world multi-view dataset for benchmarking versatile industrial anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 22883–22892). https://doi.org/10.1109/CVPR52733.2024.02159
[6] Photonics21. (2026). Light driving the future: Photonics strategic research and innovation agenda 2028–2034. https://www.photonics21.org/download/ppp-services/photonics-downloads/Photonics_Strategic_Research_and_Innovation_Agenda_2028-2034_C1.pdf