Speaker
Description
Understanding the microstructures of materials has become increasingly crucial for studying their structure-property relationships, i.e., the influence of structural descriptors on macroscopic material properties. In materials with polycrystalline microstructures, the 3D grain architecture significantly influences mechanical properties. Imaging techniques like diffraction contrast tomography and 3D electron backscattered diffraction (EBSD), which can provide informative 3D image data (i.e., grain maps) of polycrystalline materials, are often not readily available. However, 3D image data is highly valuable not only for computing structural descriptors that quantify the material microstructures but also for serving as input for spatially resolved simulations, enabling a deeper understanding of material properties. Conversely, 2D image data acquired via 2D EBSD is more accessible, but it has limitations—such as its inability to serve as input for spatially resolved 3D simulations of material behavior—and quantifying 3D structures from 2D data is often non-trivial. This talk introduces a computational method to address these challenges by generating digital twins of the 3D morphology of material microstructures using stochastic 3D modeling calibrated with 2D image data [1]. The method employs parametric models from stochastic geometry, particularly parametric random tessellations, to generate random virtual 3D grain architectures, including those with curved grain boundaries. Calibration of model parameters to 2D data is achieved through methods of generative artificial intelligence (AI). Specifically, a neural network (discriminator) is trained to guide selection of model parameter such that 2D cross-sections of generated structures statistically match the experimentally measured 2D image data. The method is demonstrated using 2D EBSD data of recycled AA6082 aluminum alloys, where the reconstructed 3D grain architectures enable a quantitative characterization of the 3D structure. Then, the reconstructed 3D grain maps can serve as input for spatially resolved 3D simulations of crystal plasticity [2]. The parametric nature of the stochastic 3D models allows for systematic parameter variation, enabling the generation of additional structural scenarios beyond those calibrated to experimental data. In this way, the database of 3D microstructures can be expanded to derive data-driven process-structure-property relationships and to virtually design materials with tailored mechanical properties.
References
[1] L. Fuchs, O. Furat, D.P. Finegan, J. Allen, F.L.E. Usseglio-Viretta, B. Ozdogru, P.J. Weddle, K. Smith and V. Schmidt. “Generating multi-scale Li-ion battery cathode particles with radial grain architectures using stereological generative adversarial networks.” Communications Materials (in print).
[2] B. Klusemann, B. Svendsen and H Vehoff. International Journal of Plasticity 50 (2013) 109–126.
