Journal of Computational Design and Engineering,
Год журнала:
2024,
Номер
11(3), С. 56 - 71
Опубликована: Апрель 16, 2024
Abstract
Metasurfaces
can
effectively
attenuate
Rayleigh
waves
propagating
in
soil,
safeguarding
structures
from
ambient
vibrations
or
earthquakes.
However,
there
remains
a
lack
of
efficient
approaches
for
designing
metasurfaces
that
isolate
within
desired
frequency
ranges
under
different
site
conditions.
This
study
presents
deep
learning
(DL)-based
topology
optimization
method
isolating
target
range,
which
has
potential
applications
surface
wave
control.
The
proposed
DL
model
employs
variational
autoencoder
to
transform
high-dimensional
and
discrete
topologies
into
low-dimensional
continuous
latent
vectors,
reducing
the
design
difficulty.
On
this
basis,
conditional
tandem
neural
network
is
constructed
optimize
vectors
soil
conditions,
improving
efficiency
verifying
universality
method.
reliability
validated
through
100
tests
with
determination
coefficients
more
than
0.99.
In
addition,
generations
same
are
explored,
providing
designers
choices.
insulation
capabilities
designed
against
Metro-induced
demonstrated
time-
frequency-domain
responses.
presented
DL-aided
provides
novel
insight
customization
manipulating
waves.
ACS Applied Materials & Interfaces,
Год журнала:
2024,
Номер
16(23), С. 29547 - 29569
Опубликована: Май 29, 2024
The
use
of
metamaterials
in
various
devices
has
revolutionized
applications
optics,
healthcare,
acoustics,
and
power
systems.
Advancements
these
fields
demand
novel
or
superior
that
can
demonstrate
targeted
control
electromagnetic,
mechanical,
thermal
properties
matter.
Traditional
design
systems
methods
often
require
manual
manipulations
which
is
time-consuming
resource
intensive.
integration
artificial
intelligence
(AI)
optimizing
metamaterial
be
employed
to
explore
variant
disciplines
address
bottlenecks
design.
AI-based
also
enable
the
development
by
parameters
cannot
achieved
using
traditional
methods.
application
AI
leveraged
accelerate
analysis
vast
data
sets
as
well
better
utilize
limited
via
generative
models.
This
review
covers
transformative
impact
for
current
challenges,
emerging
fields,
future
directions,
within
each
domain
are
discussed.
Materials Horizons,
Год журнала:
2024,
Номер
11(11), С. 2615 - 2627
Опубликована: Янв. 1, 2024
We
introduce
a
novel
deep
learning-based
inverse
design
framework
with
data
augmentation
for
chiral
mechanical
metamaterials
Bézier
curve-shaped
bi-material
rib
realizing
wide
range
of
negative
thermal
expansion
and
Poisson's
ratio.
Applied Energy,
Год журнала:
2023,
Номер
355, С. 122216 - 122216
Опубликована: Ноя. 18, 2023
Segmented
thermoelectric
generators
(STEGs)
provide
an
excellent
platform
for
thermal
energy
harvesting
devices
because
they
improve
power
generation
performance
across
a
broad
range
of
operating
temperatures.
Despite
the
benefit
direct
energy-to-electricity
conversion,
conventional
STEG
optimization
approaches
are
unable
to
systematic
method
selecting
optimal
multiple
stacks
p-
and-n-type
materials
(TEs)
legs
from
set
numerous
TE
materials.
In
this
study,
we
propose
based
on
machine
learning
find
maximization.
A
deep
neural
network
(DNN)
is
trained
using
initial
dataset
generated
via
Finite
Element
Method
(FEM),
with
inputs
including
temperature-dependent
properties
and
n-type
materials,
lengths
each
segment,
external
loads,
as
well
corresponding
outputs.
The
DNN
captures
inherent
nonlinear
relationship
between
these
combination
genetic
algorithm
(GA)
efficiently
navigates
vast
design
space
88
p-type
70
along
device
factors.
It
formulates
four
stacked
segment
pairs
in
n-leg
TEGs,
targeting
new
superior
designs
enhanced
power,
efficiency,
or
both.
iteratively
refined
active
(AL)
by
incorporating
enhance
prediction
accuracy.
optimized
STEGs
exhibit
efficiency
that
1.91
1.5
times
higher,
respectively,
than
top
training
composed
157.916
STEGs.
Furthermore,
compared
TEG
without
segmentation,
our
discovered
high-performing
designs.
