Dendritic microstructures form during the solidification of a variety of metal parts, from traditionally cast engine blocks to 3D-printed specialty tooling. These dendrites can evolve through growth, coarsening, fragmentation, and the formation of a Columnar-to-Equiaxed Transition (CET), which all can greatly affect material properties. However, the basic science behind these morphological and topological changes is still not fully understood. One reason is that under normal gravitational conditions, convection in the liquid during solidification causes fluid flow, complicating the underlying physics. Another reason is that the large 3D datasets obtained can be difficult to analyze computationally.The first roadblock is addressed in this work by analyzing data from three different microgravity experiments. The first data are from an Al-Si alloy system with dendrites nucleated and grown at a set temperature gradient and pulling velocity. Results from this set of experiments show that fragmentation of dendrite arms is more likely to occur in regions where dendrites with different orientations are in close proximity. The second dataset is for a Pb-Sn system of already nucleated and grown columnar dendrites with different volume fractions of the Sn phase. These samples were isothermally coarsened just above the eutectic temperature for different lengths of time. It is shown that the 40 vol% samples coarsen at a faster rate, while the 20 vol% samples have a higher rate of fragmentation and undergo a transition from dendrites to rods to spherical particles. The third group of experiments studies Pb-Sn samples with already formed dendrites that have undergone various temperature profiles in the mushy zone. The formulation of these is experiments is explained in detail. The sample measurements will show if a theorized critical cooling rate exists for the fragmentation process, and if so, allow this rate to be estimated. Additionally, the effects of volume fraction, initial microstructure, and other temperature profiles during solidification will be analyzed. The second roadblock is addressed by adapting machine learning procedures to better segment materials datasets. This is shown to be highly effective at binarizing dendritic samples. Best practices for improving the binarization of materials image datasets are provided.

Creator

• Thompson, Zachary

DOI

• https://doi.org/10.21985/n2-0y9r-qw72

Subject

• Materials Science

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:15913
• etdadmin_upload_876872

Date created

• 2022-01-01

Resource type

• Dissertation

Rights statement

• In Copyright