| Résumé | High-integrity die castings require controlled strength and ductility for structural applications. These properties are the product of the local microstructure of the material after die filling and solidification. In this paper, a workflow of metallographic imaging, image analysis, and machine learning is investigated to estimate the mechanical properties at specific locations in a casting based on the local microstructure. Approximately 180 tensile specimens were first extracted from high pressure vacuum die cast Aural™-2/F plates at 1.8, 3.0 and 4.7 mm thickness and tested. Cross-sectional optical micrographs were then taken close to fracture locations at different magnifications to observe the microstructure. Image analysis routines were developed and applied to systematically quantify the key microstructural characteristics that are expected to affect strength and ductility. Challenges related to sampling of multi-scale and heterogeneous material, imaging resolution, high-volume analysis automation, and statistical descriptions were addressed to seek out compromises between characterization effort and accuracy. Finally, the predictive capability of different families of machine learning algorithms was tested with the dataset of the extracted microstructural characteristics for yield strength, elongation at break, and area reduction at fracture. Feature importance was also evaluated to determine key microstructural characteristics used in correlations. This work therefore assesses the potential for local, destructive estimation of expected in-service mechanical behaviour, for instance in regions where tensile coupons cannot be extracted. Validated relationships between microstructure and properties could also eventually complement simulation-based microstructure predictions from process parameters in an integrated computational materials engineering framework for designing new, lightweight die-cast structural components. |
|---|
