Surrogate-based optimization of microstructural features of structural materials

This paper proposes a methodology for surrogate-based microstructural optimization of structural metals that integrates a limited number of 3D image-based numerical simulations with microstructural quantification and coarsening and optimisation processes. The support vector machine that was used had an infill sampling criterion to reduce the number of numerical trials, and the proposed methodology was found to be effective for wrought 2024 aluminium alloy with irregularly shaped particles. Appropriate objective functions were defined to assess particle damage. The number of design parameters, which quantitatively express the size, shape, and spatial distribution of particles, was initially 41, but they were reduced to four during a two-step coarsening process. The surrogate model provided highly accurate predictions, and the size, shape, and spatial distribution values of the optimal and weakest particles were successfully identified. It was shown that the optimal particle was small, spherical, sparsely dispersed, and perpendicular to the loading direction. However, it was also found that the smallest and most independent particle with a spherical shape was not necessarily strong, which implies the effects of particle clustering. It was also concluded that the dependency of in-situ particle strength on size was of crucial importance for weaker particles. The shape and spatial distribution of stronger particles were, however, more crucial for suppressing their internal stress than was their size. The results show that the proposed methodology offers a cost-efficient solution for microstructural designs involving 3D high-fidelity simulations that cannot be obtained with the existing approaches for developing materials.