Speaker
Description
Detection and characterization of non-metallic inclusions in casting (including continuous casting and ingot casting) is essential for ensuring steel quality. In recent years, image analysis processing of metallic materials based on machine learning and deep learning has developed rapidly, which a new technical tool for detecting defects in engineering materials, e.g. steels. There are several kinds of inclusions in continuous casting slabs. Different kinds of inclusions have the obviously different features in in SEM, which lays the foundation for the use of machine learning to distinguish between the types of inclusions. This work firstly selected MnS and Al2O3, two typical types of inclusions in casting slabs of duplex stainless steels (DSSs) as experimental objects and established a semantic segmentation machine learning (ML) model to identified MnS and Al2O3 particles using SEM images. The U-Net architecture is selected in this work due to its excellent performance in image segmentation tasks with limited data. Different types of inclusions were characterized by EDS for the model prediction. Furthermore, the established model for inclusions classification is not limited to distinguishing between MnS and Al2O3, other types of inclusions such as MgO, SiO2, and MgAl2O4 etc. as well as casting defects (pore, crack, etc.) in the continuous casting slabs can be detected by applying the current semantic segmentation model. A reasonable prediction accuracy of the established methodology demonstrates the potential for employing AI-based method machine learning in the continuous casting process for high quality steel production.
| Speaker Company/University | Lulea University of Technology |
|---|