• Britz, D.
• Müller, M.
• Detemple, E.
• Laub, M.
• Bachmann, B.-I.
• Motz, Christian
• Scherff, F.
• Mücklich, F.
• Staudt, T.

Abstract

<jats:title>Abstract</jats:title><jats:p>The prior austenite grain size (PAGS) represents one of the most significant microstructural parameters for steel research and process development. Since the PAGS directly correlates with recrystallisation during rolling in the manufacturing process of steel plates, it has a huge influence on its mechanical properties. Methods to determine the PAGS reliably and reproducibly are in high demand. There are several different approaches, based on different working principles, aiming to measure the PAGS. In this paper, the focus will be held on chemical etching methods because they allow, other than indirect techniques, space-resolved images as output, coupled with a fast application with good statistics and do not necessarily require a pretreatment of the specimen that can alter properties of interest. A parameter study has been conducted to identify unknown influencing variables as well as to tune well known parameters for their application to low-carbon steels. In the scope of this work, a novel and objective way of determining the PAGS is being presented. A reproducible approach has been developed that is able to automatically reconstruct the prior austenite grain boundaries (PAGB) from low-carbon steels and thereby determining the PAGS. Based on an improved etching recipe, a routine could be elaborated using modern methods of machine learning in the field of computer vision that is able to quantitatively analyze optical micrographs. Semantic segmentation is used to detect the PAGB based on correlative EBSD data and expert’s annotations; thus, reconstructing the prior morphological microstructure. Therefore, besides the determination of the average grain size, the distribution of the PAGS and their morphological parameters can be quantified.</jats:p>

Topics

• impedance spectroscopy
• Carbon
• grain
• grain size
• steel
• etching
• electron backscatter diffraction
• machine learning