Speaker
Description
D. Sörensen1*, A. Morillo1, C. Sinz1, D. Staudenecker1, T. Bernthaler1, A. K. Choudhary1, F. Trier1, M. Weller2, C. Metzmacher2
1 Matworks GmbH, Aalen, Germany
2 Carl Zeiss SMT GmbH, Oberkochen, Germany
*David.Soerensen@matworks.de
High-alloy corrosion resistant steels such as 1.4112 (X90CrMoV18) are widely used in demanding applications due to their high hardness and wear resistance. These properties arise from a martensitic matrix containing a high-volume fraction of chromium-rich carbides. However, pronounced carbide clustering or carbide banding may negatively influence mechanical performance, particularly tensile strength and fracture behavior.
This work presents the development of an automated image analysis workflow for the quantification of carbide banding in metallographic sections of 1.4112 steel. First, a metallographic preparation route was established to enable artifact-free polishing of samples containing large carbide clusters and contrast etching via final polishing. High resolution optical light microscopy was then used for semi-automated image acquisition to ensure consistent image quality suitable for quantitative evaluation.
Carbide and pore detection was performed using a supervised deep learning approach based on semantic segmentation to separate both feature from the martensitic matrix properly. Using learning technique, deep learning model was trained on manually annotated micrographs to automatically identify carbides and pores. The segmentation results enable the quantitative evaluation of carbide distributions and carbide area fractions within defined regions of interest (1 × 1 mm²). The analysis procedure is conceptually aligned with ASTM E1268-19, while extending the evaluation toward automated determination of carbide surface fractions.
The objective of this work is to establish quantitative microstructural descriptors that allow correlation of carbide banding with mechanical properties. Initial results indicate that carbide area fraction can be quantified on a microscopic large scale. With this, locally strong inhomogeneity appearance can be visualized and correlated with the fracture behavior of tensile test samples.