ABSTRACT This review presents AI‐based approaches for predicting key properties of silicon nitride (Si 3 N 4 ) ceramics from both microstructural images and process parameters. Convolutional neural networks (CNNs) successfully predicted bending strength, fracture toughness, and dielectric breakdown strength directly from microstructural images, with higher accuracy for bending strength due to its strong correlation with visible grain features. Fracture toughness showed slightly lower accuracy, reflecting the influence of grain boundary phases that are only partially captured in images. Analysis of CNN representations showed clustering by attributes like grain size and porosity, indicating that structural features can be distinguished from additional factors affecting properties. Machine learning regression models successfully predicted thermal conductivity from processing conditions, and the prediction accuracy was further improved by incorporating predicted relative density as an additional input data. Model interpretability was assessed using SHAP (SHapley Additive Explanations), which confirmed the critical role of relative density in capturing the combined effects of sintering additives and conditions. These results demonstrate that AI can integrate structural and non‐structural information for accurate and interpretable property prediction. This approach provides a foundation for data‐driven materials design and virtual experimentation in advanced ceramics.
Furushima et al. (Wed,) studied this question.