Accelerated crystal structure prediction of multi-elements random alloy using expandable features.

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning Applications

Background:

• Material properties are dictated by crystal structures, crucial for high entropy alloys (HEAs) mechanical performance.
• Predicting crystal structures is essential for discovering new functional materials.
• Machine learning (ML) excels at phase prediction but requires extensive, costly datasets for multi-element alloys.

Purpose of the Study:

• To develop an efficient ML approach for predicting multi-element alloy structural phases.
• To overcome the challenge of large dataset requirements in ML for alloy development.
• To enable accurate crystal structure prediction without training on multi-element alloy data.

Main Methods:

• Designed a transformation module to convert raw features into an expandable form.
• Trained a ML model using a dataset of binary alloys.
• Applied the trained model to predict the structural phases of multi-element alloys and HEAs.

Main Results:

• Achieved 80.56% accuracy in predicting multi-element alloy phases and 84.20% for HEA phases.
• Demonstrated successful application of a model trained on binary alloys to multi-element systems.
• Reduced computational cost for HEA prediction by over three orders of magnitude.

Conclusions:

• The developed method efficiently predicts multi-element alloy crystal structures without large training datasets.
• This approach offers a computationally inexpensive alternative to traditional ML methods for alloy discovery.
• The technique holds potential for predicting diverse structural properties of novel multi-element alloys.