In this contribution, we propose a general framework to inversely designing microstructures using microstructure-property linkages. In a data-based approach statistical descriptors and Bayesian optimisation are employed. For deriving profound correlations, large databases are necessary. Experiments alone are prohibitively expensive. Therefore, computational augmentation is employed to allow for data-driven approaches.
The general framework consists of four steps: (1) characterisation of microstructure images by translation-invariant descriptors by MCRpy [1], (2) reconstruction of three-dimensional microstructures from the descriptors by MCRpy [1], (3) numerical simulations to compute effective mechanical properties and (4) correlation of descriptors and properties and prediction/identification of descriptors of further microstructures for improved quality of the correlation. Steps 2 through 4 are repeated until a microstructure with desired properties is found.
This framework is applied and presented at the example of microstructure with hard precipitates of variable position and morphology in a softer matrix. For this purpose, a synthetic initial database of microstructures and corresponding properties is created using DREAM.3D [2]. Here, mechanical properties, such as yield strength, Young’s modulus and fatigue indicator parameters are considered. Augmenting this small dataset by in silico reconstructed microstructures and their simulated effective properties allows for deriving improved structure-property linkages and, thus, finding potentially optimal microstructures or predicting properties.

References
[1] P. Seibert, A. Raßloff et al., Integrating Materials and Manufacturing Innovation, 2022, 11, 450-466.
[2] M.A. Groeber, M.A. Jackson, Integrating Materials, 2014, 3, 56-72