High-throughput Computing and Machine Learning Accelerated Alloy Design

Pengfei Guan*

Beijing Computational Science Research Center

ABSTRACT: With the development of computing capability and methods, computational simulation has gradually played an indispensable role in alloy design and r&d, providing a necessary foundation for the implementation of the “Material Genome Project” - a new paradigm in material science research. We will introduce our progresses and prospect related to the alloy composition design from two aspect: 1) Aiming to the precipitation phase strengthening in Al alloy, we demonstrate a strategy to stabilize nanoprecipitates in Al–Cu alloys, based on computational design that identifies synergistic solutes (Sc and Fe) that simultaneously segregate to the θ -Al2Cu/Al interface and strongly bond to one another. Furthermore, Sc and Fe are predicted to each segregate into a separate atomic plane, forming a sandwiched structure reinforcing the interface. This interfacial architecture was realized through a simple heat treatment in a Sc–Fe–Si triple-microalloyed Al–Cu model alloy. Such a back-to-back layered interface, thermodynamically stable and kinetically robust, is found to suffocate nanoprecipitate coarsening at 300°C, enabling a dramatic reduction in creep rate;Based on machine learning and neural network framework, we developed an automated code for constructing the interaction between multiple elements, and realized the large scale molecular dynamics simulation with first principle precision. Our progress bridged the quantum mechanics calculations and the classical mechanics simulation based on the conception of the potential energy surface, and provided the possibility for investigating the structure evolution in complex alloy by large-scale simulations with high precision.

Keywords: high throughput calculations;alloy design; machine learning; multi-scale simulation