Deep learning powered optical microscopy for steel research

0568817 - ÚPT 2023 CZ eng A - Abstract
Mikmeková, Šárka - Čermák, Jan - Ambrož, Ondřej - Jozefovič, Patrik - Zouhar, Martin - Materna Mikmeková, Eliška
Deep learning powered optical microscopy for steel research.
16th Multinational Congress on Microscopy, 16MCM, 04-09 September 2022, Brno, Czech Republic. Book of abstracts. Brno: Czechoslovak Microscopy Society, 2022 - (Krzyžánek, V.; Hrubanová, K.; Hozák, P.; Müllerová, I.; Šlouf, M.). s. 347-348. ISBN 978-80-11-02253-2.
[Multinational Congress on Microscopy /16./. 04.09.2022-09.09.2022, Brno]
Grant - others:AV ČR(CZ) LQ100652201
Program: Prémie Lumina quaeruntur
Institutional support: RVO:68081731
https://www.16mcm.cz/wp-content/uploads/2022/09/16MCM-abstract-book.pdf

Steel is by far the world´s most important, multi-functional, and most adaptable material. The excellent mechanical properties of steels are determined by their microstructure. The
microstructure of advanced steels, such as advanced high strength steels (AHSS), is a combination of different phases or constituents with complex substructures and its classification is extremely challenging. Traditionally used imaging techniques, such as light optical microscopy (LOM) or confocal microscopy (CM), turn out to be insufficient for precise structural characterization of the AHSS, and high-resolution imaging of their surface by state-of-the-art scanning electron microscopes (SEM) is required. However, common metallographic laboratories are equipped only with the LOM or the CM at best. As a result, they are not able to sufficiently characterize the structure of modern steels. In this work, we employ artificial intelligence techniques - namely deep learning - to enhance the OM micrographs of the AHSS in order to raise phase separation. Each training data-point consists of a trinity of corresponding micrographs obtained by the OM, the CM, and the ultra-high resolution SEM. We will demonstrate that the deep learning methods have the potential to enhance the contrast among
fine features in the OM micrographs and enable accurate secondary phase identification in the AHSS using LOM.
Permanent Link: https://hdl.handle.net/11104/0340087