Enhanced phase prediction of high-entropy alloys through machine learning and data augmentation

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• High-entropy alloys (HEAs) possess unique properties determined by their phase structure, crucial for targeted applications.
• Predicting HEA phase structures is challenging due to the vast elemental combinations and complexity.
• Machine learning (ML) offers a promising avenue for efficient phase structure prediction in HEAs.

Purpose of the Study:

• To develop and validate an ML model for accurate HEA phase structure prediction.
• To address data scarcity in HEA datasets through data augmentation techniques.
• To identify key compositional features influencing HEA phase formation.

Main Methods:

• Utilized a dataset of 544 high-entropy alloy configurations for training.
• Employed a Generative Adversarial Network (GAN) to augment the limited dataset.
• Implemented ML models for predicting intermetallic, solid solution, and amorphous phases.
• Validated model performance on an independent dataset and analyzed feature importance.

Main Results:

• Data augmentation with GANs significantly improved ML model performance, achieving 94.77% average accuracy.
• The validated model demonstrated 100% prediction accuracy on an independent dataset.
• Accurate prediction of FCC and BCC phases for solid solution HEAs was achieved (peak accuracy 98%).
• Feature importance analysis revealed correlations between composition and phase formation consistent with experimental data.

Conclusions:

• The proposed ML strategy effectively enhances the accuracy and generalizability of HEA phase prediction.
• GAN-based data augmentation is a viable solution for overcoming data limitations in HEA research.
• This work accelerates the rational design and application of high-entropy alloys through reliable phase prediction.