Journal of Marine Science and Engineering,
Год журнала:
2023,
Номер
11(10), С. 1898 - 1898
Опубликована: Сен. 29, 2023
In
this
study,
deep
neural
network
(DNN)
and
transfer
learning
(TL)
techniques
were
employed
to
predict
the
viscous
resistance
wake
distribution
based
on
positions
of
flow
control
fins
(FCFs)
applied
containerships
various
sizes.
Both
methods
utilized
data
collected
through
computational
fluid
dynamics
(CFD)
analysis.
The
position
fin
(FCF)
hull
form
information
as
input
data,
output
included
coefficients
components
propeller
axial
velocity.
base
DNN
model
was
trained
validated
using
a
source
dataset
from
1000
TEU
containership.
grid
search
cross-validation
technique
optimize
hyperparameters
model.
Then,
for
varying
To
enhance
accuracy
feature
prediction
with
limited
amount
rate
optimization
conducted.
Transfer
involves
retraining
reconfiguring
model,
verified
fine-tuning
method
results
study
can
provide
designers
performance
evaluation
by
predicting
distribution,
without
relying
CFD
Journal of Marine Science and Engineering,
Год журнала:
2023,
Номер
11(4), С. 835 - 835
Опубликована: Апрель 15, 2023
This
paper
presents
a
review
of
the
different
methods
and
techniques
used
to
optimize
ship
hulls
over
last
six
years
(2017–2022).
shows
percentages
reduction
in
resistance,
thus
fuel
consumption,
improve
ships’
energy
efficiency,
towards
achieving
goal
maritime
decarbonization.
Operational
research
machine
learning
are
common
decision
support
find
optimal
solution.
covers
four
areas
hulls,
including
hull
form,
structure,
cleaning
lubrication.
In
each
area
research,
several
computer
programs
used,
depending
on
study’s
complexity
objective.
It
has
been
found
that
no
specific
method
is
considered
optimum,
while
combination
can
achieve
more
accurate
results.
Most
work
focused
concept
stage
design,
operational
conditions
recently
taken
place,
an
improvement
efficiency.
The
finding
this
study
contributes
mapping
scientific
knowledge
technology
identifying
relevant
topic
areas,
recognizing
gaps
opportunities.
also
helps
present
holistic
approaches
future
supporting
realistic
solutions
sustainability.
Reduced-order
models
(ROMs)
have
been
extensively
employed
to
understand
complex
systems
efficiently
and
adequately.
In
this
study,
a
novel
parametric
ROM
framework
is
developed
produce
Eulerian–Lagrangian
simulations.
This
study
employs
two
typical
strategies
reproduce
the
physical
phenomena
of
gas–solid
flow
by
predicting
adequate
dynamics
modal
coefficients
in
ROM:
(i)
based
on
radial-basis
function
(RBF)
interpolation,
termed
ROM-RBF
(ii)
long–short
term
memory
(LSTM)
neural
network,
ROM-LSTM.
ROM,
an
advanced
technique,
namely,
Lanczos-based
proper
orthogonal
decomposition
(LPOD),
transform
numerical
snapshots
into
coefficients.
Validation
tests
are
conducted
system
such
as
spouted
bed.
The
coherent
structures
flows
shown
be
captured
LPOD
technique.
Besides,
comparison
with
high-fidelity
simulations,
our
proposed
ROMs
simulate
significantly
reducing
calculation
time
several
orders
magnitude
faithfully
macroscopic
properties.
particular,
compared
ROM-RBF,
ROM-LSTM
can
capture
fields
more
accurately
within
flows.
International Journal of Naval Architecture and Ocean Engineering,
Год журнала:
2024,
Номер
16, С. 100596 - 100596
Опубликована: Янв. 1, 2024
Designing
a
hull
form
typically
involves
beginning
with
reference
based
on
ship
owner
requirements,
editing
the
to
satisfy
and
determining
most
efficient
form.
Numerical
analyses
using
computational
fluid
dynamics
(CFD)
were
employed
assess
performance
of
However,
these
require
extensive
resources,
making
it
challenging
perform
thorough
within
design
timeframe.
To
address
this
issue,
paper
proposes
an
approach
that
defining
range
forms
characteristic
curves,
predicting
their
deep
neural
networks
(DNNs),
subsequently
optimal
predictions.
Initially,
small
was
defined
four
curves
parameterized
29
variables.
Fairness
optimization
performed
define
surface.
By
varying
parameters,
896
different
generated,
CFD
analysis
conducted
for
each
variant.
These
data
then
used
build
DNN
model
capable
parameters.
The
accuracy
evaluated,
resulting
in
mean
absolute
error
(MAE)
2.835%.
Subsequently,
is
combined
genetic
algorithm
identify
set
parameters
form,
This
process
revealed
reduced
total
hydrodynamic
resistance
by
approximately
7%
compared
initial
design.
Consequently,
study
demonstrates
effectiveness
proposed
method
deriving
ships.
As
an
important
technology
in
ocean
engineering
and
aerospace
fields,
the
development
of
flow
field
super-resolution
reconstruction
stems
from
urgent
need
for
high-fidelity
analysis.
In
order
to
avoid
randomness
difficulty
parameter
adjustment
caused
by
machine-learning-based
methods
reconstruction,
this
paper
uses
idea
dynamic
mode
decomposition
(DMD),
introduces
numerical
method
Schur–Padé
real
power
matrix,
proposes
a
temporal
prediction
DMD-α,
which
only
matrix
manipulation
realize
periodic
at
any
time.
Taking
wave
formed
movement
trimaran
regular
waves
as
example,
selection
strategy
based
on
DMD-α
is
proposed
take
accuracy
efficiency
into
account.
Furthermore,
proper
orthogonal
Kriging
surrogate
models
are
combined
with
arbitrary
side-hull
layout
validate
robustness
method.
The
results
show
that
stable,
efficient,
can
obtain
prediction,
has
great
potential
complex
fields
optimization
design
fluid
performances
various
structures.
