Inverse design of anisotropic spinodoid materials with prescribed diffusivity.

Area of Science:

• Materials Science
• Computational Science
• Chemical Engineering

Background:

• Material microstructure dictates effective properties like mass transport in porous materials.
• Tuning microstructure allows for control over material properties.
• Spinodoid structures offer tunable anisotropy and are efficient models for porous materials.

Purpose of the Study:

• To develop a computational method for designing porous materials with specific, anisotropic mass transport properties.
• To leverage machine learning for predicting effective diffusivity in three dimensions.
• To enable efficient inverse design of microstructures based on desired properties.

Main Methods:

• Utilized Gaussian random fields to model spinodoid-like porous structures with tunable anisotropy.
• Employed a convolutional neural network (CNN) to predict effective diffusivity tensor components.
• Integrated CNN predictions into an approximate Bayesian computation (ABC) framework for inverse design.

Main Results:

• The CNN accurately predicted effective diffusivity in all three principal directions for the modeled microstructures.
• The inverse design approach, guided by the CNN, successfully generated microstructures with prescribed target diffusivities.
• The combined approach proved computationally efficient for designing materials with specific transport characteristics.

Conclusions:

• Convolutional neural networks can effectively predict anisotropic transport properties in porous materials.
• Approximate Bayesian computation coupled with machine learning enables efficient inverse design of functional material microstructures.
• This methodology offers a powerful tool for tailoring porous materials for specific applications requiring controlled mass transport.