Wilk-Kołodziejczyk Dorota, Nowotny Aleksandra, Krzak Izabela, Tchórz Adam, Jaśkowiec Krzysztof, Małysza Marcin, Bitka Adam, Głowacki Mirosław, Książek Marzanna, Marcjan Łukasz
Department of Applied Computer Science and Modelling Department, Faculty of Metals Engineering and Industrial Computer Science, AGH University of Krakow, Czarnowiejska 66, 30-054, Krakow, Poland.
Łukasiewicz Research Network -Krakow Institute of technology, Zakopiańska 73, Krakow, Poland.
Sci Rep. 2025 Jan 13;15(1):1880. doi: 10.1038/s41598-025-86005-y.
CT images of castings made of ductile iron were analyzed in the paper. On these images, objects can be identified that can be considered as graphite precipitates or indicate the presence of a defect in the casting. Research conducted in this area is described, based on experimental data that allows to determine whether the indicated components present in the casting are graphite precipitation. Analyzing the results, a conclusion was drawn that the classification based solely on the input data used is insufficient. Such action allowed to obtain information that there are particles in the casting that can be both graphite separation and imperfections (in particular voids, porosities, discontinuities). These results are subjected to further analysis (pictures) to help decide whether the object is a separation or a discontinuity. The available (experimental) data make it possible to unequivocally identify belonging to one of these groups. The use of machine learning methods to recognize the relationships between the physical parameters of particles helps to improve the analysis process. An important aspect was the determination of three ranges in the scale of shades of gray, which were used to determine the labels for the input data. Lighter shades in the first range indicate slight differences in the density of the particle, and thus suggest the occurrence of cast fineness. The middle range corresponding to the darker shades of gray was assigned to particles that could be shrinkage porosities. The darkest shades corresponded to occurrences of gas porosities (voids). Shades of gray cannot be the only determinant of the type of microstructure component, because apart from imperfections, there are also graphite precipitations in the casting (shape and shade of gray resembling emptiness). It cannot be assumed that specific types of defects will occur in the tested object (e.g. only gas porosities), which requires additional analysis of the microstructure image.
本文分析了球墨铸铁铸件的CT图像。在这些图像上，可以识别出可被视为石墨析出物或表明铸件存在缺陷的物体。基于能够确定铸件中所示成分是否为石墨析出的实验数据，描述了在该领域进行的研究。通过对结果的分析得出结论，仅基于所使用的输入数据进行分类是不够的。这样的操作使得能够获得铸件中存在的颗粒既可能是石墨分离又可能是缺陷（特别是气孔、孔隙、不连续性）的信息。对这些结果进行进一步分析（图片），以帮助确定该物体是分离还是不连续性。现有的（实验）数据使得能够明确识别其属于这些组中的哪一组。使用机器学习方法来识别颗粒物理参数之间的关系有助于改进分析过程。一个重要方面是确定灰度等级的三个范围，用于确定输入数据的标签。第一个范围内较浅的灰度表示颗粒密度的细微差异，因此表明铸件的细化情况。对应于较深灰度的中间范围被分配给可能是缩孔的颗粒。最深的灰度对应于气孔（空洞）的出现。灰度等级不能是微观结构成分类型的唯一决定因素，因为除了缺陷之外，铸件中还存在石墨析出物（形状和灰度类似于空洞）。不能假设在测试对象中会出现特定类型的缺陷（例如仅气孔），这需要对微观结构图像进行额外分析。
