DMREF: High Throughput Design of Metallic Glasses with Physically Motivated Descriptors

University Of Wisconsin-Madison, Madison WI

Investigators

Abstract

Non-technical Description: Silica-based glasses are familiar to most of us from our experiences with everything from windows to wineglasses. However, when cooled quickly enough, some metal alloys can also form a glassy state. Metallic glasses have attractive properties such as high elastic modulus, excellent strength, good biocompatibility, and the ability to be processed like plastics. Applications include packaging, arterial stents, water purification, and Micro-Electro-Mechanical Systems gears and springs. The slowest cooling rate that still forms a glass is called the critical cooling rate. Although many metals can form a glass, only a rare set of alloys have slow enough critical cooling rate that they can form a significant bulk volume of glassy material, and these alloys are said to have good glass forming ability. Despite the importance of these materials and years of research, there are still no rigorous, consistent, and quantitative rules to predict the actual glass forming ability of a metallic alloy system. To solve this problem, this project will develop an extensible materials informatics framework for predicting the glass-forming ability of metal alloys, and then apply that framework to develop aluminum- and magnesium-based alloys with improved glass forming ability. Technical description: In order to discover new aluminum- and magnesium-based bulk metallic glasses with superior glass-forming ability, the team will execute a dual-loop iterative materials design approach. A rapid materials design loop will provide high-throughput materials discovery by integrating experimental and simulated data with machine learning methods. An unprecedented body of experimental data on glass forming ability and basic mechanical properties will be generated by combinatorial 3D printing synthesis, followed by rapid optical, microscopy, thermal, and nanomechanical characterization. A similarly unique database of liquid and glass thermodynamic, kinetic, and structural properties will be determined by automated, high-throughput ab initio molecular dynamics. Machine-learning methods, trained on the data and physically motivated descriptors from existing experiments and the ab initio molecular dynamics simulation, will search a space of up to hundreds of thousands of potential alloys for the most promising candidates, which will then be synthesized, characterized and used to refine the models. Slower descriptor design loop studies will study select alloys in detail with fluctuation electron microscopy and extensive simulations to develop improved descriptors, which will then be incorporated into the rapid materials design loop and further validated by their predictive ability. This work will produce the first set of large-scale databases with both true measures of glass forming ability and extensive thermophysical data from simulations, and integrate them to generate physical descriptor driven machine-learning models for iterative new metallic glass search and discovery. The PIs also plan to release the Materials Simulation Toolkit - Machine Learning (MASTML) as open source and build a user community around the language by ensuring that interested researchers are able to contribute to the MASTML codebase. This will allow a wider growth of the project. This aspect is of special interest to the software cluster in the Office of Advanced Cyberinfrastructure, which has provided co-funding for this award.