
Acta Mechanica, Год журнала: 2024, Номер 235(8), С. 5257 - 5272
Опубликована: Июнь 11, 2024
Abstract
The
deep
operator
network
(DeepONet)
structure
has
shown
great
potential
in
approximating
complex
solution
operators
with
low
generalization
errors.
Recently,
a
sequential
DeepONet
(S-DeepONet)
was
proposed
to
use
learning
models
the
branch
of
predict
final
solutions
given
time-dependent
inputs.
In
current
work,
S-DeepONet
architecture
is
extended
by
modifying
information
combination
mechanism
between
and
trunk
networks
simultaneously
vector
multiple
components
at
time
steps
evolution
history,
which
first
literature
using
DeepONets.
Two
example
problems,
one
on
transient
fluid
flow
other
path-dependent
plastic
loading,
were
demonstrate
capabilities
model
handle
different
physics
problems.
trained
inverse
parameter
identification
via
genetic
algorithm
application
model.
almost
all
cases,
achieved
an
$$R^2$$
Язык: Английский