Loading...
Procedural analyses of image-based and multiscale investigations of materials
Becker, C. Gus
Becker, C. Gus
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2023
Date Submitted
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Methods of material investigation often produce data in the form of images. Visual
assessments can convey qualitative analyses of these images, but expressing these
initial analyses in a quantitative manner is nontrivial. This work studies the role of image
processing procedures in extracting information for material investigations.
The first objective of this work was to explore the feasibility of in-situ, microfocus
x-radiography to monitor the composition of a binary alloy during solidification. Features
in an image are represented by localized intensities which must be connected to physical
quantities before quantitative information can be interpreted in an image. In this study, an
image processing procedure was developed to reduce noise across a large number of
radiographs captured during directional solidification. The final radiograph in the series,
depicting the as-solidified sample, was compared with composition data captured via
energy dispersive spectroscopy (EDS) to show radiographic intensity trends matched the
compositional trends in the EDS data.
The second objective of this work was to determine the success of image processing
procedures for identifying and tracking solid-liquid (S-L) interfaces during in-situ, melt
pool analysis of solidification experiments. These measurements allowed for the
calculation of solidification velocity. When paired with information about the thermal
gradients in these experiments (obtainable by comparison with simulations), this
information allows for a characterization of the structure and therefore properties of the
as-solidified metal. The detection procedures were developed for synchrotron
x-radiography of simulated additive manufacturing (AM) and dynamic transmission
electron microscopy (DTEM) of thin film rapid solidification and compared with manual
measurements. This work sought to determine whether these procedures would be able
to reduce inconsistencies due to subjective judgment calls made during manual
annotation. In the case of the AM simulator, the results showed large deviations due to
noise, but a higher amount of the detected measurements were within the average
manual distribution. For the rapid solidification, detected results matched the manual
results closely.
The third objective of this work was to improve the segmentation of multi-sized,
irregularly-shaped, and tightly-clustered particles in a 2D image. There are many
algorithms to automatically segment features within images, but these algorithms work
best for features with uniform sizes and shapes that have well-defined boundaries. In this
iii
study, a new segmentation procedure is proposed which consists of three steps:
preprocessing to over-identify particles, application of a watershed algorithm to
intentionally over-segment these particles into regions, and a custom algorithm to
selectively merge these over-segmented regions based on edge intensity between
regions. The resulting merged-region results and the results from a typical watershed
segmentation are both compared with a manual segmentation of the same image. The fit
between the merged-region results and the manual segmentation is calculated to be
higher than the fit between the typical watershed and manual segmentation.
The final objective of this work was to develop a flexible workflow for generating 3D
geometries of granular materials from microfocus x-ray computed tomography (XCT)
data. These geometries are to be used as initial conditions in image-based physics
simulations. A software package called Segmentflow was developed to contain the
workflow functions. Segmentflow is controlled by an input file which specifies the input
data to be segmented, the segmentation parameters, and the output format of the
results. The features of Segmentflow are exhibited by creating simulation-ready
geometries from an XCT scan of a mock high explosives system consisting of F50 silica
sand and a Kel-F polymer binder. The geometries are verified by analyzing the
segmented particles and comparing the results to a typical size distribution of F50 sand.
A variety of mesh postprocessing is also performed to show how Segmentflow can be
used to control the complexity of a simulation.
Associated Publications
Rights
Copyright of the original work is retained by the author.