Quantifying grain boundary structure – property relationships with atomistic simulations and data driven approaches

Cui, Yue

Permalink

https://hdl.handle.net/2142/115336 Copy

Description

Title

Quantifying grain boundary structure – property relationships with atomistic simulations and data driven approaches

Author(s)

Cui, Yue

Issue Date

2022-02-03

Director of Research (if dissertation) or Advisor (if thesis)

Chew, Huck Beng

Doctoral Committee Chair(s)

Chew, Huck Beng

Committee Member(s)

• Geubelle, Philippe
• Lambros, John
• Lopez-Pamies, Oscar

Department of Study

Aerospace Engineering

Discipline

Aerospace Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Multi-scale modelling
• Atomistic stresses
• Machine learning
• Grain boundary

Abstract

Grain boundaries are prevalent in almost all engineering materials and control many of their mechanical properties. In conventional polycrystalline metals, the grain boundaries obstruct the motion of dislocations because of the mismatch between grains. As the grain size decreases, dislocation starvation – coupled with the increase in density of grain boundaries – results in increased material strength, following the well-established Hall-Petch relation. At sufficiently small grain sizes of ~5 to 20 nm, however, the material behavior deviates from the Hall-Petch relationship since strengthening is now controlled by the interaction of individual dislocations with the grain boundaries, resulting in unit mechanisms such as dislocation emission, absorption, or transmission across the grain boundaries. It is now well-accepted that these unit mechanisms are controlled by the atomistic stress state at the boundaries, and the per-atom stresses along these grain boundaries are very important descriptors governing the grain boundary mechanisms of dislocation emission, transmission, and absorption. To-date, the atomistic stresses along a grain boundary can only be discerned from molecular dynamics (MD) simulations where there is an atomistic “constitutive law” in the form of interatomic potentials. In constrat, the notion of atomistic stress is not well-defined in quantum-mechanical calculations such as Density Functional Theory (DFT). Similarly, while the general configuration of atoms along a grain boundary can be discerned from high resolution transmission electron microscopy (TEM) imaging, the atomistic stresses cannot be easily quantified. The goal of my thesis research is to obtain an alternative and more general means of constructing the atomistic stresses applicable to atomistic structures beyond the MD domain, and use these atomistic stresses to construct quantitative grain boundary descriptors. A numerical technique, termed the sequential atom removal (SAR) approach, is developed to reconstruct the atomic stresses near a symmetrical-tilt Σ5(310)[001] Cu grain boundary. In the SAR approach, individual atoms near the boundary are sequentially removed to compute the pair (reaction) force between atoms, while correcting for changes to the local electron density caused by atom removal. This novel SAR approach accurately reproduces the spatially-varying virial stresses at a grain boundary governed by an embedded atom method potential. The SAR approach is subsequently used to extract the atomistic stresses of the grain boundary from DFT calculations. This has enabled the DFT-based construction of a continuum-equivalent traction distribution along the grain boundary, as a quantitative descriptor of the grain boundary atomic structure. Despite the success and high accuracy of the SAR approach, it is computationally expensive and only applicable to atomistic simulations. As an alternative and more general approach, artificial neural networks (ANN) for machine learning (ML), fed with a limited training dataset from MD simulations, is used to predict the local atomistic stresses from atomic position information across a series of equilibrium <110> symmetrical-tilt Cu grain boundary structures. Accuracy of the ML algorithm is found to depend on the type, sequence, and distortion of the grain boundary structural units. Accounting for these characteristics in the training dataset enables accurate predictions of the local atomistic stress distributions across the family of grain boundary structures. This ML-based constitutive modeling paves the way for direct interpretation of the equivalent stress state of atomistic structures beyond the MD domain, including those from high-resolution transmission electron microscopy (HRTEM) imaging and Density Functional Theory (DFT) modeling. Determining the per-atom atomistic stresses is fundamental to quantifying the grain boundary atomistic structures. Using a genetic algorithm, a dislocation-based representation of the grain boundary atomistic structure was obtained, by inversely reconstructing the location, type, and orientation of dislocations from the elastic stress fields of grain boundaries. This inverse approach captures the sparsely populated dislocations in low angle grain boundaries. However, several challenges remain when extending this genetic algorithm to higher angle grain boundaries, particularly in the presence of dislocation clusters with overlapping cores. The structural representation of higher angle grain boundaries is a subject of future work.

Graduation Semester

2022-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Yue Cui

Owning Collections