Predicting Porosity in Freeze Casting with Explainable Machine Learning.

Area of Science:

• Materials Science
• Manufacturing Engineering
• Computational Materials Science

Background:

• Freeze casting is a key method for creating porous materials with tunable properties.
• Predicting porosity from process parameters is complex and vital for material design.
• Existing methods lack the precision needed for optimizing freeze cast materials.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting porosity in freeze casting.
• To identify key process parameters influencing porosity using explainable AI.
• To enhance the design and optimization of freeze cast materials.

Main Methods:

• Utilized experimental data from 252 research papers on ceramics, polymers, and composites.
• Applied machine learning algorithms: CatBoost, Random Forest, and XGBoost.
• Employed Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• CatBoost model achieved the highest predictive accuracy with an R² of 0.81 on the test set.
• SHAP analysis identified solid loading as the most influential parameter.
• Lower solid loading was correlated with higher predicted porosity, aligning with theoretical expectations.

Conclusions:

• Machine learning, particularly CatBoost, offers a powerful tool for predicting porosity in freeze casting.
• Explainable AI (SHAP) provides crucial insights into material behavior and parameter influence.
• This approach can guide experimental design and optimize material properties for specific applications.