Multi-Objective Bayesian Optimization of Complex, Concentrated Alloys

High-entropy alloys (HEA) and the broader class of multi-principal element alloys (MPEA) and baseless alloys, collectively called complex, concentrated alloys (CCA), sit at the forefront of the state-of-the-art in high-strength metals research. The near equiatomic concentrations of components in these alloys promote a multitude of desirable effects, ranging from entropic phase stabilization to atomic size mismatch and local magnetic strengthening mechanisms. Often, the combined effects of these properties lead to CCAs with high temperature mechanical strengths that far surpass that of our best high-strength steels and Ni superalloys. While the origins of these properties are typically rooted in the compositional disorder of CCAs, their complex compositional domain is also what ultimately makes the task of engineering or tailoring specific properties a nearly insurmountable hurdle when approached with traditional experiment-based optimization techniques. In this work, we demonstrate how Bayesian inference led multiscale modeling of relevant atomic and microscopic properties can instead be used to efficiently navigate the vast compositional space of these CCAs to optimize multiple properties at the same time. We utilize a diversity guided multi-objective Bayesian optimization (MOBO) scheme, involving objective modeling via Gaussian processes (GP), to optimize sets of mechanical and thermodynamic properties in several refractory body-centered cubic (BCC) CCA systems, including MoNbTaTi, MoNbTaWVCrMn, and MoNbTaWTiV. In doing so, we also attempt to base descriptions of important macroscopic properties on easy-to-obtain atomic level properties from multiscale ab-initio and molecular dynamics calculations to best expedite the process of data generation and objective optimization. By identifying the compositional neighborhoods that yield the best combinations of desired alloy properties, a large portion of the legwork can be eliminated before any physical experiment is performed, greatly reducing the associated cost and time to develop new state-of-the-art CCAs.<br/><br/><br/>SAND2022-15070 AX<br/><br/>Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525. -or- if you have a character limit, you may use: SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525