Applied Sciences,
Год журнала:
2024,
Номер
14(17), С. 7855 - 7855
Опубликована: Сен. 4, 2024
Given
the
critical
role
of
true
triaxial
strength
assessment
in
underground
rock
and
soil
engineering
design
construction,
this
study
explores
sandstone
using
data-driven
machine
learning
approaches.
Fourteen
distinct
test
datasets
were
collected
from
existing
literature
randomly
divided
into
training
(70%)
testing
(30%)
sets.
A
Multilayer
Perceptron
(MLP)
model
was
developed
with
uniaxial
compressive
(UCS,
σc),
intermediate
principal
stress
(σ2),
minimum
(σ3)
as
inputs
maximum
(σ1)
at
failure
output.
The
optimized
Harris
hawks
optimization
(HHO)
algorithm
to
fine-tune
hyperparameters.
By
adjusting
structure
activation
function
characteristics,
final
made
continuously
differentiable,
enhancing
its
potential
for
numerical
analysis
applications.
Four
HHO-MLP
models
different
functions
trained
validated
on
set.
Based
comparison
prediction
accuracy
meridian
plane
analysis,
an
high
predictive
meridional
behavior
consistent
theoretical
trends
selected.
Compared
five
traditional
criteria
(Drucker–Prager,
Hoek–Brown,
Mogi–Coulomb,
modified
Lade,
Weibols–Cook),
demonstrated
superior
performance
both
datasets.
It
successfully
captured
complete
variation
space,
showing
smooth
continuous
envelopes
deviatoric
planes.
These
results
underscore
model’s
ability
generalize
across
conditions,
highlighting
a
powerful
tool
predicting
geotechnical
Journal of Rock Mechanics and Geotechnical Engineering,
Год журнала:
2023,
Номер
16(6), С. 2310 - 2325
Опубликована: Сен. 5, 2023
The
prediction
of
liquefaction-induced
lateral
spreading/displacement
(Dh)
is
a
challenging
task
for
civil/geotechnical
engineers.
In
this
study,
new
approach
proposed
to
predict
Dh
using
gene
expression
programming
(GEP).
Based
on
statistical
reasoning,
individual
models
were
developed
two
topographies:
free-face
and
gently
sloping
ground.
Along
with
comparison
conventional
approaches
predicting
the
Dh,
four
additional
regression-based
soft
computing
models,
i.e.
Gaussian
process
regression
(GPR),
relevance
vector
machine
(RVM),
sequential
minimal
optimization
(SMOR),
M5-tree,
compared
GEP
model.
results
indicate
that
less
bias,
as
evidenced
by
root
mean
square
error
(RMSE)
absolute
(MAE)
training
(i.e.
1.092
0.815;
0.643
0.526)
testing
0.89
0.705;
0.773
0.573)
in
ground
topographies,
respectively.
overall
performance
topology
was
ranked
follows:
>
RVM
M5-tree
GPR
SMOR,
total
score
40,
32,
24,
15,
10,
For
condition,
SMOR
21,
19,
8,
Finally,
sensitivity
analysis
showed
both
ground,
liquefiable
layer
thickness
(T15)
major
parameter
percentage
deterioration
(%D)
value
99.15
90.72,
Journal of Rock Mechanics and Geotechnical Engineering,
Год журнала:
2024,
Номер
16(11), С. 4532 - 4553
Опубликована: Май 17, 2024
To
optimize
the
excavation
of
rock
using
underground
blasting
techniques,
a
reliable
and
simplified
approach
for
modeling
fragmentation
is
desired.
This
paper
presents
multistep
experimental-numerical
methodology
simplifying
three-dimensional
(3D)
to
two-dimensional
(2D)
quasi-plane-strain
problem
reducing
computational
costs
by
more
than
100-fold.
First,
in
situ
tests
were
conducted
involving
single-hole
free-face
dolomite
mass
1050-m-deep
mine.
The
results
validated
laser
scanning.
craters
then
compared
with
four
analytical
models
calculate
radius
crushing
zone.
Next,
full
3D
model
was
prepared
simulating
crack
length
Based
on
stable
propagation
zones
observed
experiments,
2D
prepared.
properties
high
explosive
(HE)
slightly
reduced
match
shape
number
radial
cracks
zone
between
models.
final
used
reproduce
various
cut-hole
scenarios
observe
effects
residual
further
fragmentation.
presence
preexisting
found
be
crucial
fragmentation,
particularly
when
borehole
situated
near
free
face.
Finally,
an
optimization
study
performed
determine
possibility
losing
continuity
at
different
positions
within
well
relation
Journal of Rock Mechanics and Geotechnical Engineering,
Год журнала:
2024,
Номер
unknown
Опубликована: Июль 1, 2024
In
underground
mining,
especially
in
entry-type
excavations,
the
instability
of
surrounding
rock
structures
can
lead
to
incalculable
losses.
As
a
crucial
tool
for
stability
analysis
critical
span
graph
must
be
updated
meet
more
stringent
engineering
requirements.
Given
this,
this
study
introduces
support
vector
machine
(SVM),
along
with
multiple
ensemble
(bagging,
adaptive
boosting,
and
stacking)
optimization
(Harris
hawks
(HHO),
cuckoo
search
(CS))
techniques,
overcome
limitations
traditional
methods.
The
indicates
that
hybrid
model
combining
SVM,
bagging,
CS
strategies
has
good
prediction
performance,
its
test
accuracy
reaches
0.86.
Furthermore,
partition
scheme
is
adjusted
based
on
CS-BSVM
399
cases.
Compared
previous
empirical
or
semi-empirical
methods,
new
overcomes
interference
subjective
factors
possesses
higher
interpretability.
Since
relying
solely
one
technology
cannot
ensure
credibility,
further
genetic
programming
(GP)
kriging
interpolation
techniques.
explicit
expressions
derived
through
GP
offer
probability
value,
technique
provide
interpolated
definitions
two
subclasses.
Finally,
platform
developed
above
three
approaches,
which
rapidly
feedback.
