Recent technological advancements across all fields have led to a rapid demand for the development and deployment of new materials with distinct properties suitable for the desired applications. This, in turn, has necessitated efficient, robust material characterization systems for the investigation and analysis of the relationships between material microstructures and their mechanical properties. With the development of more effective characterization and data acquisition tools for high throughput data generation, autonomous approaches including machine learning and computer vision have attracted significant attention from researchers due to their potential application in the analysis of the interactions between the microstructure and properties of different materials.
In a recent research paper published in the journal Material Science and Engineering A, University of California at Santa Barbara researchers Dr. Zhe Chen and Professor Samantha Daly identified deformation twins in magnesium through clustering and computer vision. They tracked the evolution of twinning under applied uniaxial compression and its relationship with the material microstructure.
Briefly, the authors commenced their data analysis by segmenting a microscale deformation map that spanned a mm-scale FOV by k-means clustering. Consequently, multiple approaches were utilized to segment clusters as either twinned or non-twinned. These approaches included tracking the shape and size evolution of clusters, use of a global Schmid factor to determine the likelihood of twin activation, and use of similarities in the centroid and surface strains in the clusters. The experiment utilized a combination of scanning electron microscopy and digital image correlation (SEM-DIC), where a chemically-functionalized nanoparticle assembly was applied to create the speckle pattern necessary for DIC. In addition, in-SEM test automation and improved imaging were realized through external codes. The developed experimental and analytical approaches were validated using a ground truth data set, from which possible improvements were identified.
The study of deformation twinning is important, as twinning accommodates plastic strains and significantly affects material properties due to new boundaries that impede slip dislocation as a result of lattice reorientation. Additionally, it is necessary to examine a large number of data points existing within a single test specimen and similar test conditions for specific material properties. The authors commented that the proposed approach is applicable for numerous test conditions and materials, thus giving insights into the relationships between material parameters, resulting microstructures, test conditions, and globally-observed responses.
This experimental approach, as with other emergent approaches, enables rich data sets that need proper and effective mechanisms to manage the resultant information. This work introduces the application of clustering and computer vision to displacement data for the segmentation and identification of deformation mechanisms, enabling high-throughput statistics of microstructure-mechanism interactions. The fusion of large data analytics with the rich data sets that are increasingly resultant from the rapid development of experimental capabilities holds tremendous potential for materials advancement.
Chen, Z., & Daly, S. (2018). Deformation twin identification in magnesium through clustering and computer vision. Materials Science and Engineering: A, 736, 61-75.Go To Materials Science and Engineering: A