Enhanced image segmentation of refractories using synchrotron X-ray computed tomography and machine learning techniques
- The microstructure of refractory materials is complex, featuring a variety of mineral phases, agglomerates, defects, and controlled porosity. The behavior of refractories at high temperatures adds another layer of complexity, as phase transitions and particle rearrangements can strongly affect their properties. To analyze such intricate microstructure, advanced imaging techniques such as Synchrotron X-ray Computed Tomography (SXCT) allow detailed 3D visualization and quantification of features up to 1 μm. However, the intricacy of these microstructures makes phase identification (known as image segmentation) in digital images a challenging process. X-ray images often contain noise and image artifacts, making the analysis more difficult. Therefore, this work describes image segmentation and artifact reduction methods to characterize refractories using X- ray imaging. We studied refractory ceramics used in the aerospace industry, primarily composed of fused silica. For imageThe microstructure of refractory materials is complex, featuring a variety of mineral phases, agglomerates, defects, and controlled porosity. The behavior of refractories at high temperatures adds another layer of complexity, as phase transitions and particle rearrangements can strongly affect their properties. To analyze such intricate microstructure, advanced imaging techniques such as Synchrotron X-ray Computed Tomography (SXCT) allow detailed 3D visualization and quantification of features up to 1 μm. However, the intricacy of these microstructures makes phase identification (known as image segmentation) in digital images a challenging process. X-ray images often contain noise and image artifacts, making the analysis more difficult. Therefore, this work describes image segmentation and artifact reduction methods to characterize refractories using X- ray imaging. We studied refractory ceramics used in the aerospace industry, primarily composed of fused silica. For image segmentation, the traditional approach of greyscale thresholding was compared with machine learning. Greyscale thresholding relies on predefined algorithms to assign phases based on intensity values. In contrast, machine learning extracts patterns from large datasets, enabling more adaptive and accurate segmentation. By combining high-resolution SXCT and machine learning analysis algorithms, we successfully segmented previously uncharacterized 3D microstructural key features of refractories, including agglomerates, grain boundaries, pore size distribution and interconnectivity. Compared to traditional methods, the machine learning-enhanced segmentation presented a more accurate quantification of porosity and defects. The integration of advanced imaging techniques with machine learning segmentation significantly improves the characterization of refractory materials, providing a more precise understanding of the relationship between microstructure and material performance, supporting the development of innovative industrial solutions.…


| Autor*innen: | Paula Campos de Oliveira |
|---|---|
| Koautor*innen: | Henning MarkötterORCiD, Wen Zhang, Mustapha EddahORCiD, Martinus Putra WidjajaORCiD, Clément Remacha, Giovanni BrunoORCiD |
| Dokumenttyp: | Vortrag |
| Veröffentlichungsform: | Präsentation |
| Sprache: | Englisch |
| Jahr der Erstveröffentlichung: | 2025 |
| Organisationseinheit der BAM: | 8 Zerstörungsfreie Prüfung |
| 8 Zerstörungsfreie Prüfung / 8.5 Röntgenbildgebung | |
| DDC-Klassifikation: | Technik, Medizin, angewandte Wissenschaften / Ingenieurwissenschaften / Ingenieurwissenschaften und zugeordnete Tätigkeiten |
| Freie Schlagwörter: | Aerospace; Ceramics; Machine learning; Refractories; Synchrotron X-ray Tomography |
| Themenfelder/Aktivitätsfelder der BAM: | Material |
| Material / Materialdesign | |
| Veranstaltung: | 19th Biennial Unified International Technical Conference on Refractories (UNITECR 2025) |
| Veranstaltungsort: | Cancún, Mexico |
| Beginndatum der Veranstaltung: | 27.10.2025 |
| Enddatum der Veranstaltung: | 30.10.2025 |
| Verfügbarkeit des Dokuments: | Datei im Netzwerk der BAM verfügbar ("Closed Access") |
| Datum der Freischaltung: | 07.11.2025 |
| Referierte Publikation: | Nein |
| Eingeladener Vortrag (wissenschaftliche Konferenzen): | Nein |

