Prediction of nanocomposite properties and process optimization using persistent homology and machine learning.

Area of Science:

• Materials Informatics
• Catalysis
• Nanomaterials

Background:

• Physical property prediction and synthesis optimization are crucial in materials informatics.
• Understanding oxygen permeability in fuel cell electrodes is vital for performance.

Purpose of the Study:

• To develop a machine learning model for predicting oxygen permeability and optimizing synthesis temperature.
• To analyze the relationship between material structure and oxygen permeability using persistence diagrams.

Main Methods:

• Utilized ridge regression with persistence diagrams derived from transmission electron microscopy (TEM) tomographic images.
• Applied machine learning to Pt/CeO2 nanocomposite structures, comparing L2 and L1 regularization.
• Analyzed the correlation between ridge regression coefficients and persistence diagrams.

Main Results:

• L2 regularization was found to be more suitable than L1 for analyzing complex nanocomposite structures.
• The model successfully predicted the activation energy of oxygen permeability.
• Identified the formation process of CeO2 structures and their impact on pre-exponential factor and activation energies.

Conclusions:

• The developed machine learning approach accurately predicts physical properties and optimizes synthesis processes.
• The study provides insights into structure-property relationships in nanomaterials for fuel cell applications.
• Determined optimal annealing temperatures for enhanced material performance.