International Journal of Fatigue,
Journal Year:
2024,
Volume and Issue:
188, P. 108487 - 108487
Published: July 14, 2024
The
impact
of
mechanical
fatigue
on
load-bearing
metallic
components
and
structures
is
highly
significant,
encompassing
economy,
environment
safety
aspects.
For
nearly
200
years,
engineers
scientists
have
been
relentlessly
trying
to
avoid
failures
understand
their
causes.
last
few
decades
seen
the
prominent
advent
a
wide
range
experimental
computational
techniques
that
allowed
us
make
once-unthinkable
advances
in
this
field.
Despite
progress,
significant
number
problems
remain
unsolved.
This
short
note
pinpoints
most
critical
aspect
failure:
conditions
initiate
or
allow
propagating
cracks
form.
Specifically,
fundamentals
are
examined,
while
acknowledging
crucial
role
multiphysics
aspects
often
present
real-life
engineering
applications.
first
part
frames
problem
by
introducing
essential
concepts
exploring
mechanistic
across
different
length
scales
loading
regimes.
Subsequently,
brief
but
wide-ranging
review
highlights
current
research
trends.
foundation
sets
stage
for
identifying
outstanding
challenges
potential
future
directions
conclusive
article.
Nature Communications,
Journal Year:
2024,
Volume and Issue:
15(1)
Published: May 21, 2024
Abstract
Machine-learned
computational
chemistry
has
led
to
a
paradoxical
situation
in
which
molecular
properties
can
be
accurately
predicted,
but
they
are
difficult
interpret.
Explainable
AI
(XAI)
tools
used
analyze
complex
models,
highly
dependent
on
the
technique
and
origin
of
reference
data.
Alternatively,
interpretable
real-space
employed
directly,
often
expensive
compute.
To
address
this
dilemma
between
explainability
accuracy,
we
developed
SchNet4AIM,
SchNet-based
architecture
capable
dealing
with
local
one-body
(atomic)
two-body
(interatomic)
descriptors.
The
performance
SchNet4AIM
is
tested
by
predicting
wide
collection
quantities
ranging
from
atomic
charges
delocalization
indices
pairwise
interaction
energies.
accuracy
speed
breaks
bottleneck
that
prevented
use
chemical
descriptors
systems.
We
show
group
indices,
arising
our
physically
rigorous
atomistic
predictions,
provide
reliable
indicators
supramolecular
binding
events,
thus
contributing
development
Chemical
Artificial
Intelligence
(XCAI)
models.
Computer Methods in Applied Mechanics and Engineering,
Journal Year:
2024,
Volume and Issue:
429, P. 117104 - 117104
Published: June 17, 2024
We
explore
the
potential
of
deep
Ritz
method
to
learn
complex
fracture
processes
such
as
quasistatic
crack
nucleation,
propagation,
kinking,
branching,
and
coalescence
within
unified
variational
framework
phase-field
modeling
brittle
fracture.We
elucidate
challenges
related
neural-network-based
approximation
energy
landscape,
ability
an
optimization
approach
reach
correct
minimum,
we
discuss
choices
in
construction
training
neural
network
which
prove
be
critical
accurately
efficiently
capture
all
relevant
phenomena.The
developed
is
applied
several
benchmark
problems
results
are
shown
qualitative
quantitative
agreement
with
finite
element
solution.The
robustness
tested
by
using
networks
different
initializations.
Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences,
Journal Year:
2025,
Volume and Issue:
383(2293)
Published: April 2, 2025
Scientific
machine
learning
is
a
new
class
of
approaches
that
integrate
physical
knowledge
and
mechanistic
models
with
data-driven
techniques
to
uncover
the
governing
equations
complex
processes.
Among
available
approaches,
universal
differential
(UDEs)
combine
prior
in
form
formulations
function
approximators,
such
as
neural
networks.
Integral
efficacy
UDEs
joint
estimation
parameters
for
both
approximators
using
empirical
data.
However,
robustness
applicability
these
resultant
hinge
upon
rigorous
quantification
uncertainties
associated
their
predictive
capabilities.
In
this
work,
we
provide
formalization
uncertainty
(UQ)
investigate
key
frequentist
Bayesian
methods.
By
analyzing
three
synthetic
examples
varying
complexity,
evaluate
validity
efficiency
ensembles,
variational
inference
Markov-chain
Monte
Carlo
sampling
epistemic
UQ
methods
UDEs.
This
article
part
theme
issue
‘Uncertainty
healthcare
biological
systems
(Part
2)’.
IEEE Transactions on Ultrasonics Ferroelectrics and Frequency Control,
Journal Year:
2024,
Volume and Issue:
unknown, P. 1 - 1
Published: Jan. 1, 2024
Intraventricular
vector
flow
mapping
(iVFM)
seeks
to
enhance
and
quantify
color
Doppler
in
cardiac
imaging.In
this
study,
we
propose
novel
alternatives
the
traditional
iVFM
optimization
scheme
by
utilizing
physicsinformed
neural
networks
(PINNs)
a
physics-guided
nnU-Net-based
supervised
approach.When
evaluated
on
simulated
images
derived
from
patientspecific
computational
fluid
dynamics
model
vivo
acquisitions,
both
approaches
demonstrate
comparable
reconstruction
performance
original
algorithm.The
efficiency
of
PINNs
is
boosted
through
dual-stage
pre-optimized
weights.On
other
hand,
nnU-Net
method
excels
generalizability
real-time
capabilities.Notably,
shows
superior
robustness
sparse
truncated
data
while
maintaining
independence
explicit
boundary
conditions.Overall,
our
results
highlight
effectiveness
these
methods
reconstructing
intraventricular
blood
flow.The
study
also
suggests
potential
applications
ultrafast
imaging
incorporation
equations
derive
biomarkers
for
cardiovascular
diseases
based
flow.
Energies,
Journal Year:
2024,
Volume and Issue:
17(5), P. 1180 - 1180
Published: March 1, 2024
Evaluating
uncertainty
in
CO2
injection
projections
often
requires
numerous
high-resolution
geological
realizations
(GRs)
which,
although
effective,
are
computationally
demanding.
This
study
proposes
the
use
of
representative
(RGRs)
as
an
efficient
approach
to
capture
range
full
set
while
reducing
computational
costs.
A
predetermined
number
RGRs
is
selected
using
integrated
unsupervised
machine
learning
(UML)
framework,
which
includes
Euclidean
distance
measurement,
multidimensional
scaling
(MDS),
and
a
deterministic
K-means
(DK-means)
clustering
algorithm.
In
context
intricate
3D
aquifer
storage
model,
PUNQ-S3,
these
algorithms
utilized.
The
UML
methodology
selects
five
from
pool
25
possibilities
(20%
total),
taking
into
account
reservoir
quality
index
(RQI)
static
parameter
reservoir.
To
determine
credibility
RGRs,
their
simulation
results
scrutinized
through
application
Kolmogorov–Smirnov
(KS)
test,
analyzes
distribution
output.
this
assessment,
40
wells
cover
entire
alongside
set.
end-point
indicate
that
structural,
residual,
solubility
trapping
within
follow
same
distribution.
Simulating
GRs
over
200
years,
involving
10
years
injection,
reveals
consistently
similar
patterns,
with
average
value
Dmax
0.21
remaining
lower
than
Dcritical
(0.66).
Using
methodology,
expenses
related
scenario
testing
development
planning
for
reservoirs
presence
uncertainties
can
be
substantially
reduced.
