|About this Abstract
||2022 TMS Annual Meeting & Exhibition
||Late News Poster Session
||Data-driven Approach to Design of Multicomponent Metallic Glasses
||Anurag Bajpai, Krishanu Biswas
|On-Site Speaker (Planned)
The current study incorporates Mendeleev Number (MN) into the definition of atomic topology to design MMG alloy systems more accurately. The incorporation of MN emphasizes the importance of electronic interactions between atoms and bond orbitals in the amorphous phase formation in multicomponent alloy systems. This understanding of atomic topology, together with other key physical, thermodynamical, and kinetic features, was used to develop a near-foolproof design strategy for multicomponent metallic glasses (MMGs) prediction using machine learning (ML). Feature engineering was used to optimize the descriptor space, and various ML algorithms were used to classify the amorphous multicomponent alloys over crystalline counterparts. Based on various evaluation metrics, the Support Vector Machine (SVM) algorithm emerged as the best classifier, with a testing set accuracy of 92.7 percent. Several novel high entropy alloys (HEAs) were synthesized into ribbons via melt spinning and characterized based on the ML model's predictions to validate its outcomes.
||Computational Materials Science & Engineering, High-Entropy Alloys, Machine Learning