Prediction of Solute Segregation at Metal/Oxide Interfaces Using Machine Learning Approaches.

Area of Science:

• Materials Science
• Computational Materials Science
• Surface Science

Background:

• Atomic structure and chemistry at metal/oxide interfaces govern material properties.
• Studying semi-coherent interfaces with misfit dislocations using DFT is computationally intensive due to large atom counts.

Purpose of the Study:

• To explore solute segregation behavior at the Fe/Y2O3 interface, a model for nuclear reactor cladding.
• To develop and validate a machine learning (ML) approach combined with DFT calculations for predicting solute segregation.
• To identify key factors influencing solute segregation and reduce computational costs.

Main Methods:

• Employed density functional theory (DFT) to calculate segregation energies (ESeg).
• Developed and trained machine learning (ML) models using DFT-derived ESeg data.
• Investigated chemical and geometric factors affecting solute segregation at metal/oxide interfaces.

Main Results:

• Identified key chemical and geometric factors influencing solute segregation and their competitive effects on ESeg.
• Achieved high accuracy in predicting solute segregation at a specific Fe/Y2O3 interface using ML models.
• Demonstrated ML models' ability to predict segregation at a different Fe/Y2O3 interface orientation with over 45x reduced computational cost compared to DFT.

Conclusions:

• The combined DFT-ML approach accurately predicts solute segregation at metal/oxide interfaces.
• ML models significantly reduce the computational expense of studying complex interfaces.
• This methodology offers a powerful tool for understanding and designing materials for applications like nuclear cladding.