Back to search

Georgia Institute of Technology

Stochastic Microstructure Functions: Statistically Conditioned Microstructure Generation, Big Datasets, and Quantification

Abstract

dc:description.abstract

Over the last decade, the accessibility of expressive and large datasets has unlocked the transformative potential of data science and machine learning toolsets in a wide variety of fields. However, adoption of these toolsets in the sciences and engineering has been limited because of the challenge of curating such datasets. Materials Informatics – the study of materials behavior and processing using data science – is no exception. In this dissertation, I present a practical and theoretical analysis of the Stochastic Microstructure Function (SMF) formalism – a leading statistical formalism for the study of materials – that culminates in an overarching framework for the curation of large heterogeneous microstructure datasets. I touch on three topics relating to this goal: Microstructure Quantification, Statistically Conditioned Generation, and Dataset Curation. From the practical perspective, I will explore how each of these tasks can be efficiently and systematically addressed using elements of the SMF concept. In particular, I will present an exhaustive study of microstructure generation within this formalism. I will propose generative algorithms that are two orders of magnitude faster, more stable (i.e., able to extrapolate), and can be constructed with less training data (i.e., require just a single training image) than previous state of art models. Additionally, these generative models allow the user to systematically incorporate much more complex hierarchies of material features than were previously possible. In addition to these practical developments, this dissertation grapples with two prominent theoretical questions. First, it analyzes the compatibility between the Stochastic Microstructure Function idea and modern deep learning techniques. I demonstrate the natural harmony that exists between the two, using deep learning models to estimate intractable elements of the SMF as well as using the SMF formalism to understand and direct the performance of deep learning models. Finally, I present a direct analysis of the SMF as a mathematical object, documenting the information content of its prominent statistical fingerprints, the n-point statistics.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Mechanical Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Robertson, Andreas Euan
Advisor dc:contributor.advisor
  • Kalidindi, Surya R.
Committee members dc:contributor.committeemember
  • McDowell, David L.
  • Joseph, Roshan V.
  • Schaefer, Florian T.
  • Stebner, Aaron

Subjects

dc:subject × 10

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/77649

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Robertson, Andreas Euan. Stochastic Microstructure Functions: Statistically Conditioned Microstructure Generation, Big Datasets, and Quantification. Doctoral thesis, Georgia Institute of Technology, 2023. https://hdl.handle.net/1853/77649