3D Data for Fatigue in Superalloys
13
5 Future Needs
The most obvious limitation of the TriBeam approach, like most other 3D techniques, is the cost and time required for acquisition and reconstruction of the
dataset. Crystallographic orientation mapping in SEM is usually performed via
EBSD, which has seen recent speed improvements with the replacement of CCDs
with CMOS cameras into standard phosphor-optics-type setups. It is likely that rates
will continue somewhat with direct electron (DE) detectors, which may also have
the advantage of enhanced electron sensitivity. We expect these CMOS and DE
systems combined with emerging indexing algorithms [63–66, 68, 77] will decrease
collection times and increase data mapping quality. The other significant time
restriction is (if necessary) the glancing angle surface cleaning of the femtosecond
laser-ablated surfaces. Currently the TriBeam uses a Ga + FIB with 65 nA beam
current. Xenon plasma FIBs are available and have been integrated into a new
prototype TriBeam [60] to produce currents up to 20 times higher than a Ga + FIB
[85], which can scale to a similar 20× surface cleanup rate increase, depending on
the material.
Data sharing, provenance, and portability have become a key issue for the
large-scale and collaborative efforts required to tackle scientific problems with 3D
data. A new software and data infrastructure, BisQue [86–88], has been useful for
addressing the data challenges and providing a platform on which data versions can
be synchronized between collaborative institutions and parallelized, parameterized
processing of data workflows is possible.
Data merging from various modalities including HR-DIC, synchrotron X-ray
DCT, and TriBeam tomography is challenging due to the complex distortions
associated with each experimental method. For instance, SEMs can have spatial
distortions and drift distortions from the electron optics and sample charging effects
[71, 89]. New algorithms are being developed to perform and address data merging
including those used for combining synchrotron diffraction contrast tomography
and TriBeam tomography [50] and a generalized multimodal data merging approach
using an evolutionary optimization machine learning algorithm [76].
Furthermore, developments in digital image correlation (DIC) via highresolution DIC and Heaviside-DIC [90] and coupling with EBSD data are being
used to predict strain localization, slip transmission across boundaries, and how
strain can create “microvolumes” [91], where non-Schmid-type loading conditions
are imposed on adjacent grains across a grain boundary. Opportunities for the
targeted investigation of the influence of the subsurface 3D grain structure on strain
localization and transmission phenomenon are also emerging.
At the precipitate scale, the glide of dislocations that locally shear precipitates
results in strain localization along twin boundaries [15, 16, 71, 92, 93]. These are
ultimately sites for crack initiation, and their intersection with grain boundaries
dominates the early stages of crack growth [6, 12, 94–98]. The new 3D characterization capabilities described here, in combination with multiscale plasticity models,
ultimately enable much higher fidelity prediction of properties such as yield strength
13
5 Future Needs
The most obvious limitation of the TriBeam approach, like most other 3D techniques, is the cost and time required for acquisition and reconstruction of the
dataset. Crystallographic orientation mapping in SEM is usually performed via
EBSD, which has seen recent speed improvements with the replacement of CCDs
with CMOS cameras into standard phosphor-optics-type setups. It is likely that rates
will continue somewhat with direct electron (DE) detectors, which may also have
the advantage of enhanced electron sensitivity. We expect these CMOS and DE
systems combined with emerging indexing algorithms [63–66, 68, 77] will decrease
collection times and increase data mapping quality. The other significant time
restriction is (if necessary) the glancing angle surface cleaning of the femtosecond
laser-ablated surfaces. Currently the TriBeam uses a Ga + FIB with 65 nA beam
current. Xenon plasma FIBs are available and have been integrated into a new
prototype TriBeam [60] to produce currents up to 20 times higher than a Ga + FIB
[85], which can scale to a similar 20× surface cleanup rate increase, depending on
the material.
Data sharing, provenance, and portability have become a key issue for the
large-scale and collaborative efforts required to tackle scientific problems with 3D
data. A new software and data infrastructure, BisQue [86–88], has been useful for
addressing the data challenges and providing a platform on which data versions can
be synchronized between collaborative institutions and parallelized, parameterized
processing of data workflows is possible.
Data merging from various modalities including HR-DIC, synchrotron X-ray
DCT, and TriBeam tomography is challenging due to the complex distortions
associated with each experimental method. For instance, SEMs can have spatial
distortions and drift distortions from the electron optics and sample charging effects
[71, 89]. New algorithms are being developed to perform and address data merging
including those used for combining synchrotron diffraction contrast tomography
and TriBeam tomography [50] and a generalized multimodal data merging approach
using an evolutionary optimization machine learning algorithm [76].
Furthermore, developments in digital image correlation (DIC) via highresolution DIC and Heaviside-DIC [90] and coupling with EBSD data are being
used to predict strain localization, slip transmission across boundaries, and how
strain can create “microvolumes” [91], where non-Schmid-type loading conditions
are imposed on adjacent grains across a grain boundary. Opportunities for the
targeted investigation of the influence of the subsurface 3D grain structure on strain
localization and transmission phenomenon are also emerging.
At the precipitate scale, the glide of dislocations that locally shear precipitates
results in strain localization along twin boundaries [15, 16, 71, 92, 93]. These are
ultimately sites for crack initiation, and their intersection with grain boundaries
dominates the early stages of crack growth [6, 12, 94–98]. The new 3D characterization capabilities described here, in combination with multiscale plasticity models,
ultimately enable much higher fidelity prediction of properties such as yield strength