A methodology for the top-down design of ceramic materials composed of two or more phases has been developed. It is demonstrated on the example of the well-known alumina/zirconia (AZ) material system. Core of the method is an automated simulation chain based on representative volume elements for generating a database of microstructure-property relations. The database emanating from this simulation chain was used to train machine learning models for enabling fast predictions of material microstructure according to preset material properties. A gradient boosting algorithm provided reliable and fast calculations for the exemplary chosen thermal and mechanical properties of the AZ material system. This enables reverse identification of selected microstructural parameters needed to obtain a specific value of a material property of interest, or briefly: top-down ceramic material design. • An automatized microstructure-property simulation chain for ceramics was developed. • The simulation concept can handle complex phenomena like internal thermal stresses. • A machine learning model (gradient boosting) was trained on calculated properties. • The ML model enables fast reverse identification of microstructural parameters.
Read full abstract