Abstract
Electron backscatter diffraction (EBSD) is the principal technique used to obtain crystallographic information
in the scanning electron microscope (SEM). EBSD provides microstructural and crystallographic
information at high speeds with a large angular range and a high spatial resolution, whilst having a
high level of automation. As a result, EBSD is a statistically reliable technique, capable of producing
microstructural data over large areas. Whilst the technique is becoming increasingly widespread,
new technological developments are continuously emerging that offer advantages for aspects such as;
accessibility, experimental workflows, spatial resolution and applications development.
This work aims to evaluate and characterise some of the latest developments for a complete EBSD
detection system, to further the understanding of the capabilities of the technique. The evaluation
considers: i) the capture of electron backscattering patterns (EBSPs) using a novel Timepix-based
detector in a reflection Kikuchi diffraction (RKD) geometry, ii) crystal orientation determination via
a novel Hough-based indexing algorithm and iii) the segmentation of orientation data utilising two
contrasting grain reconstruction algorithms. These developments represent three of the main stages of
an EBSD workflow; pattern capture, crystal indexing and post-processing. Particular attention is given
to the standardisation of evaluation procedures for each stage of the technique using both simulated
and experimental data.
The collection of EBSPs at normal incidence to the electron beam was revisited via the use of a fully
operational Timepix EBSD detection system in a novel geometry. The detection system consisted of
all necessary hardware, firmware and processing software. The Timepix-based EBSD detection system,
with a small form factor and zero readout noise, demonstrated the capability to collect high quality
EBSPs in the RKD geometry despite a lower backscatter electron yield. The use of an appropriate
curve fit, and corresponding Gaussian FWHM measurement permitted a first direct physical spatial
resolution comparison between pattern quality (PQ) and normalised cross correlation (NCC) based
metrics. The optimum physical spatial resolution for a molybdenum bicrystal was 22 nm and 31 nm
for the PQ and NCC metrics respectively, when using an accelerating voltage of 12 kV. The automation
of the collection of EBSD data sets, inclusive of a flatfield image, with varying accelerating voltage,
working distance, exposure time and energy threshold was also possible. The optimum parameters for
the RKD geometry for an iron specimen were found to be: an accelerating voltage range of 8 – 12 kV,
working distance range of 10 – 14 mm and energy threshold ∼ 5 kV below the incident energy.
An indexing accuracy better than 0.2◦ is consistently achieved for a novel Hough-based indexing
algorithm that utilises a 2-D spherical mapping and generalised Voronoi diagram for band-reflector pair
correspondence searching. A mean accuracy for 2500 EBSPs approached 0.1◦ under optimum experimental
conditions. The simultaneous fitting of Kikuchi band normals produced a low sensitivity with
respect to the pattern centre calibration in the Z-direction (PCz), with an increased sensitivity in PCx
and PCy. The impact of EBSP noise level, image size and detector-specimen distance on the accuracy
of the indexer was also evaluated. Mis-indexing, as caused by pseudo-symmetry in quartz, is consistently
avoided when 13 or more bands are used for indexing. The accuracy of two contrasting EBSD
grain reconstruction algorithms; referred to as the flood-fill and clustering algorithms, were found to
be 0.0 – 0.7 % and 0.0 – 113.7 % respectively, for a range of synthetically generated microstructures.
The clustering algorithm was observed to over-segment grains under certain conditions, however, the
observation of low-angle grain boundaries for sub-grain analyses is possible using this approach.