The Open Civil Engineering Journal,
Journal Year:
2024,
Volume and Issue:
18(1)
Published: Oct. 4, 2024
Aims
This
research
seeks
to
improve
the
reliability
and
sustainability
of
tunnel
construction
by
employing
automated
AI
techniques
manage
geotechnical
aleatoric
uncertainties.
It
utilizes
machine
learning
models,
including
Gradient
Boosting
Machines
(GBM),
AdaBoost,
Hidden
Markov
Models
(HMM),
Deep
Q-Networks
for
Reinforcement
Learning,
predict
reduce
environmental
impacts.
The
effectiveness
these
algorithms
is
assessed
using
various
performance
metrics
demonstrate
their
impact
on
enhancing
processes.
Background
While
vital
modern
infrastructure
development,
it
poses
significant
challenges.
Traditional
methods
assessing
impacts
often
rely
manual
overly
simplistic
models
that
fail
consider
complex
interactions
inherent
uncertainties
factors.
aims
overcome
limitations
applying
techniques,
particularly
algorithms,
more
accurately
mitigate
Objective
goal
this
study
increase
AI-based
address
both
focuses
deploying
such
as
GBM,
HMM,
Learning
forecast
negative
algorithms'
measured
against
criteria
in
optimizing
outcomes.
Methods
applies
Q-Networks,
enhance
construction's
sustainability.
These
are
designed
while
accounting
models'
evaluated
like
accuracy,
precision,
recall,
F1
score,
log
loss,
mean
squared
error
(MSE),
log-likelihood,
cumulative
reward,
convergence
rate,
policy
stability,
indicating
substantial
improvements
practices.
Results
shows
significantly
enhances
GBM
achieved
a
high
accuracy
0.92
an
score
0.90.
Additionally,
effectively
identified
optimal
strategies,
resulting
reward
950.
outcomes
highlight
capability
uncertainties,
leading
safer,
resilient
development.
Conclusion
findings
suggest
integrating
substantially
improves
projects.
approaches
with
providing
predictive
scores
strategies.
Adopting
technologies
could
result
sustainable,
infrastructure,
underscoring
potential
transforming
Journal of Rock Mechanics and Geotechnical Engineering,
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 1, 2024
The
warming
and
thawing
of
permafrost
are
the
primary
factors
that
impact
stability
embankments
in
cold
regions.
However,
due
to
uncertainties
thermal
boundaries
soil
properties,
stochastic
modeling
regimes
is
challenging
computationally
expensive.
To
address
this,
we
propose
a
knowledge-integrated
deep
learning
method
for
predicting
regime
Geotechnical
knowledge
embedded
training
data
through
numerical
modeling,
while
neural
network
learns
mapping
from
boundary
property
fields
temperature
field.
effectiveness
our
verified
comparison
with
monitoring
analysis
results.
Experimental
results
show
proposed
achieves
good
accuracy
small
coefficient
variation.
It
still
provides
satisfactory
as
variation
increases.
an
efficient
approach
predict
heterogeneous
embankments.
can
also
be
used
other
engineering
investigations
require
modeling.
Physics of Fluids,
Journal Year:
2025,
Volume and Issue:
37(3)
Published: March 1, 2025
Real-time
wave
prediction
is
crucial
for
optimizing
offshore
renewable
energy
capture
and
ensuring
the
safety
of
floating
structures.
However,
stochastic
nonlinear
nature
waves
presents
significant
challenges
accurate
robust
predictions.
This
study
proposes
a
model
based
on
Deep
Operator
Network
(DON-WP),
which
learns
operator
to
map
historical
heights
future
heights.
By
leveraging
this
operator-learning
framework,
demonstrates
strong
generalization
across
function
space,
enabling
it
adapt
previously
unseen
conditions.
Specifically,
branch
net
encodes
data
into
functional
representations,
while
trunk
captures
targets
as
evaluation
points
output
function.
These
outputs
are
then
combined
through
element-wise
operations
generate
precise
The
model's
ability
robustness
validated
using
tank
experimental
multiple
sea
states,
its
performance
compared
with
Long
Short-Term
Memory
network-based
probabilistic
(Deep-WP).
Results
show
that
DON-WP,
trained
single
state,
achieves
over
30%
higher
accuracy
most
horizons
up
60%
improvement
shorter
steps
Deep-WP,
requires
retraining
each
state.
highlights
DON-WP
an
effective
approach
dynamics
modeling,
potential
advance
systems
enhance
The Journal of Chemical Physics,
Journal Year:
2025,
Volume and Issue:
162(11)
Published: March 17, 2025
The
piezoionic
effect
holds
significant
promise
for
revolutionizing
biomedical
electronics
and
ionic
skins.
However,
modeling
this
multiphysics
phenomenon
remains
challenging
due
to
its
high
complexity
computational
limitations.
To
address
problem,
study
pioneers
the
application
of
deep
operator
networks
effectively
model
time-dependent
effect.
By
leveraging
a
data-driven
approach,
our
significantly
reduces
time
compared
traditional
finite
element
analysis
(FEA).
In
particular,
we
trained
DeepONet
using
comprehensive
dataset
generated
through
FEA
calibrated
experimental
data.
Through
rigorous
testing
with
step
responses,
slow-changing
forces,
dynamic-changing
show
that
captures
intricate
temporal
dynamics
in
both
horizontal
vertical
planes.
This
capability
offers
powerful
tool
real-time
phenomena,
contributing
simplifying
design
tactile
interfaces
potentially
complementing
existing
imaging
technologies.
Frontiers in Earth Science,
Journal Year:
2025,
Volume and Issue:
13
Published: April 25, 2025
Accurate
prediction
of
water
inrush
volumes
is
essential
for
safeguarding
tunnel
construction
operations.
This
study
proposes
a
method
predicting
volumes,
leveraging
the
eXtreme
Gradient
Boosting
(XGBoost)
model
optimized
with
Bayesian
techniques.
To
maximize
utility
available
data,
654
datasets
missing
values
were
imputed
and
augmented,
forming
robust
dataset
training
validation
XGBoost
(BO-XGBoost)
model.
Furthermore,
SHapley
Additive
explanations
(SHAP)
was
employed
to
elucidate
contribution
each
input
feature
predictive
outcomes.
The
results
indicate
that:
(1)
constructed
BO-XGBoost
exhibited
exceptionally
high
accuracy
on
test
set,
root
mean
square
error
(RMSE)
7.5603,
absolute
(MAE)
3.2940,
percentage
(MAPE)
4.51%,
coefficient
determination
(R
2
)
0.9755;
(2)
Compared
performance
support
vector
mechine
(SVR),
decision
tree
(DT),
random
forest
(RF)
models,
demonstrates
highest
R
smallest
error;
(3)
importance
yielded
by
SHAP
groundwater
level
(
h
>
water-producing
characteristics
W
burial
depth
H
rock
mass
quality
index
RQD
).
proposed
volume
dataset,
thereby
aiding
managers
in
making
informed
decisions
mitigate
risks
ensuring
safe
efficient
advancement
projects.