| Abstract Scope |
Recent advances in hardware technology as well as sophisticated methods for post-processing of Electron Backscatter Diffraction (EBSD) and Energy Dispersive X-Ray Spectroscopy (EDS) data have opened up new possibilities for detailed quantitative microstructure characterization of polycrystalline Ni-based superalloys. However, combining EBSD and EDS scans to reconstruct the true morphology of primary ã’ particles remains challenging, as some important microstructural features exist at a scale below the EDS method's lateral resolution limit, which leads to undesired artifacts at ã/ã’ interfaces. We present an automated computer vision architecture capable of resolving the meso-scale features of polycrystalline ã/ã’ microstructures with a level of detail that has not previously been demonstrated. Our methodology involves the following steps: 1. The combination of multiple elemental EDS maps. 2. Edge-preserving filtering of EDS maps using a non-local-means algorithm. 3. Unsupervised machine learning phase segmentation based on k-means clustering and 4. An automated artifact correction for the combination of EDS and EBSD information based on morphological conditions. In this manner, digital micrographs are reconstructed in a way that allows for quantitative determination of meaningful numeric metrics by utilizing methods from the field of algorithmic geometry. Various microstructural entities, such as discrete primary ã’ particles, mixed ã/ã’ grains, or ã grains can be characterized separately, including properties of related boundaries. Geometric characteristics can be quantified in terms of the local arrangement and cluster behavior of particle groups, as well as their spacings. The present work contributes to the development of digital workflows for precise and automatic microstructure characterization. |