A process-structure-property model via physics-based/data-driven hybrid methods for freeze-cast porous ceramics in Si3N4-Si2N2O case system

Abstract

For the engineering applications of freeze-cast porous ceramics, the demand targets are often multiple and competing, which is a challenging problem to seek a Nash equilibrium in the high-dimensional design space. An accurate and robust quantification of process-structure-property correlations would provide an effective path to find the set of Pareto optimal materials for one specific need. In this work, using porous Si3N4-Si2N2O ceramics as the model materials, a hybrid model for the quantitative design of the microstructure and mechanical properties is developed from four physics-based process-microstructure models with sintering, solidification, phase transformation and grain growth kinetic theories, and the subsequent data-driven structure-property model utilizing a machine learning method, artificial neural network (ANN). The SHapely Additive exPlanations (SHAP) analysis is further introduced to interpret the ANN model and mathematically identify the contribution of each microstructure feature descriptor toward target mechanical property outputs. These results present a systematic understanding of the process-structure-property relationships through the hybrid model, guiding the optimal design of the freeze-cast porous ceramics with required microstructures and mechanical properties.