Hybrid Machine Learning/Physics-Based Approach for Predicting Oxide Glass-Forming Ability

A novel methodology of predicting specific compositions for glass forming alloys based on elemental cluster selection, liquidus lines, atomic packing efficiency and ab initio calculations is presented and discussed. The proposed composition selection model has lead to the discovery of a number of novel, soon to be reported Mg, Cu, Zn and Ag-based bulk metallic glasses. The proposed model may also be used to explain high glass forming ability and physical properties of known BMG compositions and to pin-point new or superior BMG compositions in existing glass forming systems. Further, the aforementioned model shows strong correlations between proposed elemental clusters, glass forming ability and BMG ductility. This model has also shown applicable adaptation to known ceramic oxide glass forming systems.

Download Full-text

Machine learning prediction of glass-forming ability in bulk metallic glasses

Computational Materials Science ◽

10.1016/j.commatsci.2021.110362 ◽

2021 ◽

Vol 192 ◽

pp. 110362

Author(s):

Jie Xiong ◽

San-Qiang Shi ◽

Tong-Yi Zhang

Keyword(s):

Machine Learning ◽

Metallic Glasses ◽

Bulk Metallic Glasses ◽

Glass Forming Ability ◽

Glass Forming

Download Full-text

Prediction of Glass Forming Ability of Bulk Metallic Glasses-Machine-Learning

10.21203/rs.3.rs-145304/v1 ◽

2021 ◽

Author(s):

Jaideep Reddy Gedi ◽

Tanay Saboo ◽

KAMESWARI PRASADA Rao AYYAGARI

Keyword(s):

Neural Network ◽

Machine Learning ◽

Artificial Neural Network ◽

Random Forest ◽

Critical Thickness ◽

Glassy Alloy ◽

Glass Forming Ability ◽

Machine Learning Techniques ◽

Glass Forming ◽

Artificial Neural

Abstract Bulk-Metallic-Glass has been a fascinating class of metallic systems with remarkable corrosion resistance, elastic modulus and wear resistance, while evaluating the glass forming ability has been a very interesting aspect for decades. Machine learning techniques viz., artificial neural networks and random-forest based models have been developed in this work to predict the glass forming ability, given the composition of the bulk metallic glassy alloy. A new criterion of classification of atoms present in a bulk metallic glassy alloy is proposed. Feature importance analysis confirmed that the accuracy of the prediction depends mainly on change in enthalpy of mixing and change in entropy of mixing. However, among the artificial neural network random forest models developed, the former showed a promising accuracy in prediction of the glass formation ability (critical thickness). It has been successfully demonstrated and validated with experimental critical thickness that the glass forming ability can be predicted using an artificial neural network given the elemental composition alone. A computational algorithm was also developed to classify the atoms as big/ small in a given alloy. The outcome of this algorithm was used by models developed by training with experimental data.

Download Full-text