Determining Temperature-Dependent Vickers Hardness with Machine Learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning

Background:

• Assessing structural material hardness at elevated temperatures is critical but challenging.
• Existing methods require complex experimental setups or computationally intensive simulations.

Purpose of the Study:

• To develop a machine-learning (ML) model for predicting temperature-dependent material hardness.
• To utilize chemical composition and crystal structure as input features for the ML model.

Main Methods:

• Compiled 593 Vickers hardness data points from literature across various temperatures.
• Trained an extreme gradient boosting (XGBoost) ML model using composition and smooth overlap of atomic positions (SOAP) structural descriptors.
• Employed bootstrap aggregating (bagging) to assess model variance and validate predictions.

Main Results:

• Achieved high prediction accuracy with R-squared of 0.91 and Mean Absolute Error (MAE) of 2.52 GPa.
• Demonstrated strong agreement between ML predictions and experimental hardness data.
• Successfully validated the model's ability to differentiate polymorphs and predict high-temperature hardness.

Conclusions:

• The developed ML model accurately predicts temperature-dependent hardness from material composition and structure.
• This computational approach offers a powerful tool for accelerating materials discovery and design.
• The model's performance highlights the potential of ML in addressing complex materials science challenges.