January_February_2022_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J A N U A R Y / F E B R U A R Y 2 0 2 2 7 MACHINE LEARNING | AI QUANTUM MODELING FOR HIGH-ENTROPY ALLOYS Materials researchers at the DOE’s Ames Laboratory, Iowa, recently identi- fied the source of strength and ductili- ty in high-entropy alloys (HEAs), along with ways to tune them. This discov- ery may help the power generation and aviation industries develop advanced turbine engines, which work more ef- ficiently at higher temperatures. Be- cause the elements that make up HEAs can vary, as well as their relative pro- portions, experimentally testing every possible combination and its properties would be difficult. The scientists used a quantum- mechanical modeling method to com- putationally discover and predict the atomic structure of an especially prom- ising HEA system, Fe x Mn 80−x Co 10 Cr 10 , and how transformations and defects in that structure would result in a stronger and more ductile material. “When we can pinpoint these transformations and the effect they have on a material’s properties, we can predict the strength of it, and we can deliberately design strength and duc- tility into these very complex alloys,” says scientist Duane Johnson. These predictions were confirmed experimen- tally by studying single crystal samples with selective area and electron back- scatter diffraction. According to Johnson, theory- guided computational design holds great promise for optimizing the perfor- mance of these materials, making them stronger, more ductile, and in many cases less expensive. “Using this predic- tive method, we’ve been able to speed up our alloy development timeline by more than 50%, and demonstrate 10- 20% higher operational temperatures,” he says. In the case of aviation, he be- lieves this could translate into hundreds of millions of dollars in cost savings and a significant reduction in greenhouse gas emissions. ameslab.gov . QUANTUM MECHANICS PLUS MACHINE LEARNING An engineering team from Co- lumbia University, New York, de- veloped a new computation tech- nique that combines quantumme- chanics and machine learning to accurately predict the reduction temperature of metal oxides to their base metals. They say their approach is as computationally ef- ficient as conventional calculations at zero temperature and more accurate than computationally demanding sim- ulations of temperature effects using quantum chemistry methods. “This new study is, to our knowl- edge, the first time that a hybrid ap- proach, combining computational cal- culations with AI, has been attempted for this application. And it’s the first de- monstration that quantum-mechanics- based calculations can be used for the design of high-temperature processes,” says Alexander Urban, assistant profes- sor of chemical engineering. The researchers know that at very low temperatures, calculations based on quantum mechanics can accurate- ly predict the energy that chemical re- actions require or release. They com- bined this theory with a machine learn- ing model that learned the tempera- ture dependence from widely available high-temperature measurements. They designed their approach, which fo- cuses on extracting metal at high tem- peratures, to also predict the change of the free energy with the temperature, whether it was high or low. The team is now working on extending this ap- proach to other temperature-depen- dent materials properties, such as sol- ubility, conductivity, and melting, that are needed to design electrolytic met- al extraction processes that are car- bon-free and powered by clean electric energy. engineering.columbia.edu. Electron backscatter diffraction phase maps of (a) 40 at% Fe, and (b) 45 at% Fe alloys show single-phase (fcc) and dual-phase (fcc+hcp) microstructures, respectively. Courtesy of DOI: 10.1103/PhysRevLett.127.115704. Schematic of bridging the quantumworld and high-temperature metal extraction with machine learning. Courtesy of R. Ortiz de la Morena and J.A. Garrido Torres/Columbia Engineering.