Accelerating Elastic Property Prediction in Fe-C Alloys through Coupling of Molecular Dynamics and Machine Learning

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning

Background:

• High-quality experimental data for material properties is scarce and expensive to obtain, limiting the development of accurate predictive models.
• Machine learning (ML) models require substantial datasets for reliable property prediction, a challenge exacerbated by data scarcity in materials science.

Purpose of the Study:

• To address the challenge of data scarcity by generating a large dataset of elastic properties for Fe-C alloys.
• To develop and evaluate machine learning models for predicting the bulk and shear moduli of Fe-C alloys.
• To compare the efficacy of individual ML models versus ensemble methods for property prediction.

Main Methods:

• Generated a dataset of thousands of elastic property data points for Fe-C alloys using molecular dynamics (MD) simulations.
• Employed a reference-free Modified embedded atom method (RF-MEAM) interatomic potential, fitted using ab-initio calculations.
• Trained and evaluated various ML algorithms, including super learner (SL) ensemble techniques, using alloy composition, structure, and temperature as inputs.

Main Results:

• Successfully generated an extensive dataset for Fe-C alloy elastic properties.
• Developed ML models capable of accurately predicting bulk and shear moduli.
• Demonstrated that ensemble ML techniques, like super learner, can further refine prediction accuracy.

Conclusions:

• Molecular dynamics simulations coupled with machine learning provide a powerful approach to overcome data scarcity in materials science.
• This study accelerates the prediction of elastic properties in Fe-C alloys, paving the way for efficient materials design.
• The generated dataset and developed models serve as a valuable resource for future research in alloy property prediction.