Abstract
Optical measuring and inspection systems play an important role in automation as they allow a comprehensive and non-contact quality assessment of products and processes. In this field, too, systems are increasingly being used that apply artificial intelligence and machine learning, mostly by means of artificial neural networks. Results achieved with this approach are often very promising and require less development effort. However, the supplementation and replacement of classical image processing methods by machine learning methods is not unproblematic, especially in applications with high safety or quality requirements, since the latter have characteristics that differ considerably from classical image processing methods. In this paper, essential aspects and trends of machine learning and artificial intelligence for the application in optical measurement and inspection systems are presented and discussed.
Zusammenfassung
Optische Mess- und Prüfsysteme spielen in der Automatisierung eine wichtige Rolle zur Qualitätsbewertung von Produkten und Prozessen. Auch hier kommen zunehmend Systeme zum Einsatz, die künstliche Intelligenz und maschinelles Lernen, meist mittels künstlicher neuronaler Netze, anwenden. Die mit diesem Ansatz erzielten Ergebnisse sind oft sehr vielversprechend und erfordern weniger Entwicklungsaufwand. Die Ergänzung und Ablösung klassischer Bildverarbeitung durch Methoden des maschinellen Lernens ist jedoch insbesondere bei Anwendungen mit hohen Sicherheits- oder Qualitätsanforderungen nicht unproblematisch, da letztere Eigenschaften aufweisen, die sich von klassischen Bildverarbeitungsmethoden erheblich unterscheiden. In diesem Beitrag werden wesentliche Aspekte und Trends des maschinellen Lernens und der künstlichen Intelligenz für den Einsatz in optischen Mess- und Inspektionssystemen vorgestellt und diskutiert.
About the authors
Michael Heizmann is Professor of Mechatronic Measurement Systems at the Institute of Industrial Information Technology at the Karlsruhe Institute of Technology. His research areas include machine vision, image processing, image and information fusion, measurement technology, machine learning, artificial intelligence and their applications in industry.
Alexander Braun is Professor of Physics at the University of Applied Sciences in Düsseldorf. Formerly he was responsible for the optical quality of mass produced ADAS cameras at a Tier 1. His research areas focus on physical-realistic simulation of camera systems for ADAS/AD, numerical accuracy and fundamental limits of optical models, computer vision and machine learning, and optical metrology.
Markus Hüttel was Head of the Department Machine Vision and Signal Processing at the Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) from 2008 until his retirement in 2019. His area of expertise includes machine vision, image processing, image based metrology, machine learning and their application in industrial sectors.
Christina Klüver is a private lecturer of Soft Computing at the Institute of Computer Science and Business Administration at the University of Duisburg-Essen. Her research areas include methods of Artificial Intelligence and Artificial Life for the analysis of complex systems.
Erik Marquardt studied electrical engineering at RWTH Aachen University. There he did his doctorate on an optical measurement system. He worked in industry for 15 years, mainly in machine vision companies, before joining the Association of German Engineers (Verein Deutscher Ingenieure e.V., VDI) in 2012. Since then he has been working with experts in technical committees on the development of VDI standards for optical measurement systems and additive manufacturing.
Michael Overdick is responsible for the Technology Management at SICK AG, a supplier of sensors and sensor systems for the entire field of industrial automation. Until 2009 he was in charge of the research activities of Philips in the field of medical X-ray imaging, comprising all components of the imaging chain including the medical image processing.
Markus Ulrich is Professor of Machine Vision Metrology at the Institute of Photogrammetry and Remote Sensing at the Karlsruhe Institute of Technology. Until 2020, he was head of the research team at MVTec Software GmbH. His research areas include machine vision, close-range photogrammetry, image processing, machine learning and their applications in industry.
Acknowledgment
The authors would like to thank Matthias Günther, Günther Hasna and Ralf Lay for their contribution to the VDI status report “Machine Learning” [18], which is an essential basis for this contribution.
References
1. Series of Standards VDI/VDE 5575 X-ray optical systems. Beuth Verlag, Berlin.Search in Google Scholar
2. International Data Spaces. https://www.fraunhofer.de/de/forschung/fraunhofer-initiativen/industrial-data-space.html, 07.04.2020.Search in Google Scholar
3. VDMA: machine vision at record level. https://ibv.vdma.org/en/viewer/-/v2article/render/27036181, 07.04.2020.Search in Google Scholar
4. Image Database ImageNet. http://www.image-net.org/, 07.04.2020.Search in Google Scholar
5. Image Database Open Images. https://github.com/openimages/dataset, 07.04.2020.Search in Google Scholar
6. Intel Neural Compute Stick 2. https://software.intel.com/en-us/neural-compute-stick/, 07.04.2020.Search in Google Scholar
7. VDI/VDE/VDMA 2632 Part 2:2015-10 Machine vision – Guideline for the preparation of a requirement specification and a system specification. Beuth Verlag, Berlin, 2015.Search in Google Scholar
8. VDI/VDE/VDMA 2632 Part 3:2017-10 Machine vision/industrial image processing – Acceptance test of classifying machine vision systems. Beuth Verlag, Berlin, 2017.Search in Google Scholar
9. Evaluation of measurement data – Guide to the expression of uncertainty in measurement, 2018.Search in Google Scholar
10. VDI/VDE 5585 Part 1:2018-03 Technical temperature measurement; Temperature measurement with thermographic cameras; Metrological characterisation. Beuth Verlag, Berlin, 2018.Search in Google Scholar
11. VDI/VDE/VDMA 2632 Part 3.1:2019-08 (Draft) Machine vision/industrial image processing – Acceptance test of classifying machine vision systems – Test of classification performance. Beuth Verlag, Berlin, 2019.Search in Google Scholar
12. VDI status report “Terahertzsysteme und Anwendungsfelder” (in German). http://www.vdi.de/publikationen, 2020.Search in Google Scholar
13. F. Armknecht, C. Boyd, C. Carr, K. Gjøsteen, A. Jäschke, C. A. Reuter, and M. Strand. A guide to fully homomorphic encryption. Cryptology ePrint Archive, Report 2015/1192, https://eprint.iacr.org/2015/1192, 2015.Search in Google Scholar
14. MVTec Software GmbH. HALCON/HDevelop Reference Manual. HALCON 19.11, 2019.Search in Google Scholar
15. M. Sandler, A. Howard, M. Zhu, A. Zhmoginov and L.-C. Chen. MobileNetV2: Inverted Residuals and Linear Bottlenecks. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4510–4520, 2018.10.1109/CVPR.2018.00474Search in Google Scholar
16. C. Steger, M. Ulrich and C. Wiedemann. Machine Vision Algorithms and Applications. Wiley-VCH, Weinheim, 2 edition, 2018.Search in Google Scholar
17. F. van Veen. The Neural Network Zoo. http://www.asimovinstitute.org/neural-network-zoo, 07.04.2020.Search in Google Scholar
18. VDI Verein Deutscher Ingenieure e. V., Gesellschaft für Mess- und Automatisierungstechnik, Fachbereich Optische Technologien. Maschinelles Lernen – Neuronale Netze in optischen Mess- und Prüfsystemen, VDI-Statusreport (in German). https://www.vdi.de/ueber-uns/presse/publikationen/details/kuenstliche-intelligenz-mit-neuronalen-netzen-in-optischen-mess-und-pruefsystemen, 2019.Search in Google Scholar
© 2020 Walter de Gruyter GmbH, Berlin/Boston
