Study of the Freeze Casting Process by Artificial Neural Networks.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Materials Science

Background:

• Freeze casting is a versatile technique for creating porous materials, with porosity often linked to solid fraction.
• Existing linear models struggle to accurately predict porosity across diverse freeze casting applications.
• Developing advanced predictive models is crucial for optimizing freeze casting processes.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting porosity in freeze-cast materials.
• To analyze the influence of critical process parameters on freeze casting outcomes.
• To provide a generalized framework for diverse materials using experimental big data.

Main Methods:

• Training an ANN model using experimental data from freeze casting processes.
• Identifying and utilizing four key input parameters influencing final porosity.
• Validating the ANN model's predictive accuracy and convergence.
• Developing a user-friendly mini-program for practical porosity prediction.

Main Results:

• The ANN model accurately predicts porosity based on critical process parameters.
• The model effectively generalizes findings across various materials, surpassing linear relationships.
• Two model improvements were successfully implemented to infer additional information.
• A functional mini-program was created for real-time porosity prediction.

Conclusions:

• ANNs provide a powerful, data-driven approach to understand and optimize freeze casting.
• The developed model offers a generalized rule for diverse materials, enhancing process predictability.
• The user-friendly mini-program facilitates customized freeze casting experiment design.