Application of machine learning in real-space structural characterization of nanomaterials

The first step in real-space structural characterization of nanomaterials is essentially equivalent to the creation of a digital copy of the materials sample, where the characterization is far from finished. Unlike the traditional ensemble averaging characterization such as diffraction techniques, real-space characterization directly provides image-like data. Although the image-like data look more intuitive to human beings, they are considered “rawer” for the quantitative information extraction for scientific research. Meanwhile, the heterogeneous nature of nanomaterials such as crystal grains, defects, and polydispersity further exaggerates the difficulty in their data analysis. For a long time, real-space imaging only played the supporting roles in nanomaterials characterization. Examples include solely being displayed for demonstration or providing for the manual measurement of some local features. The recent development of real-space characterization techniques such as high-throughput imaging, time series imaging, and tomography further render the situation more serious by increasing the data volume and dimensionality. To ensure the efficiency in communications of scientific research, quantitative analysis has to be performed to transfer those large-volume real-space characterization data to digestible and concise conclusions. The recent advancements in machine learning and especially those image-based algorithms such as image classification and image segmentation undoubtedly open opportunities for real-space characterization data analysis. This dissertation intends to apply cutting-edge machine learning algorithms to tackle the challenges in real-space nanomaterials characterization. Specifically, supervised neural networks are employed to achieve the accurate nanoparticle tracking in liquid-phase transmission electron microscopy videos under high noise, and then an unsupervised neural network training workflow is developed to solve the missing-wedge artifact in electron tomography, as examples of data processing in high-dimensional real-space characterization. Next, the applications of unsupervised machine learning on data interpretation are demonstrated, where dimension-reduction and clustering algorithms visualize and summarize the typical features presenting in the large-volume characterization data. Lastly, an automation of electron tomography is achieved by the real-time data processing and the feedback control of the equipment, where the fast electron tomography at continuous time points is realized, further increasing dimensionality of the real-space nanomaterials characterization.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Materials Science & Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Yao, Lehan

Contributors dc:contributor

• Chen, Qian
• Schroeder, Charles M.
• Statt, Antonia
• Murphy, Catherine Jones

• Electron Microscopy
• Machine Learning
• Nanomaterials

Statement dc:rights

• Copyright 2024 Lehan Yao

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125532

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125532