IEEE Transactions on Transportation Electrification,
Год журнала:
2023,
Номер
10(2), С. 4421 - 4431
Опубликована: Сен. 11, 2023
Convolutional
neural
networks
(CNN)
have
developed
rapidly
in
recent
years,
which
has
greatly
promoted
the
advancement
of
intelligent
fault
diagnosis.
Most
currently
available
CNN-based
diagnostic
models
are
under
presumption
that
acquired
mechanical
signals
invulnerable
to
noise.
However,
transmission
systems
usually
operate
fluctuating
conditions
(e.g.,
variable
speed
and
strong
noise
scenarios),
making
fault-related
pulse
information
signal
easily
swamped
by
Therefore,
it
is
challenging
for
these
existing
approaches
achieve
satisfactory
results
industrial
scenarios.
To
deal
with
this
problem,
an
online
knowledge
distillation-based
multiscale
threshold
denoising
network
(OKD-MTDN)
research
work.
The
main
innovations
contributions
work
include:
1)
introducing
a
novel
convolutional
module,
called
Multiscale
Module
(MCM),
alongside
Global
Attention
(GAM),
extracting
range
discriminative
features
generated
from
signals;
2)
designing
multi-dilated
module
(MTDM)
expand
receptive
field
filter
out
interference
features;
3)
establishing
distillation
(OKD)
algorithm
improve
generalization
capability
OKD-MTDN.
hF-MS
planetary
gearbox
dataset
real-running
high-speed
rail
utilized
verify
effectiveness
proposed
method.
Experimental
show
OKD-MTDN
can
various
nonstationary
Measurement Science and Technology,
Год журнала:
2024,
Номер
35(9), С. 092001 - 092001
Опубликована: Май 22, 2024
Abstract
Rolling
bearings
are
critical
components
that
prone
to
faults
in
the
operation
of
rotating
equipment.
Therefore,
it
is
utmost
importance
accurately
diagnose
state
rolling
bearings.
This
review
comprehensively
discusses
classical
algorithms
for
fault
diagnosis
based
on
vibration
signal,
focusing
three
key
aspects:
data
preprocessing,
feature
extraction,
and
identification.
The
main
principles,
features,
application
difficulties,
suitable
occasions
various
thoroughly
examined.
Additionally,
different
methods
reviewed
compared
using
Case
Western
Reserve
University
bearing
dataset.
Based
current
research
status
diagnosis,
future
development
directions
also
anticipated.
It
expected
this
will
serve
as
a
valuable
reference
researchers
aiming
enhance
their
understanding
improve
technology
diagnosis.
IEEE Transactions on Instrumentation and Measurement,
Год журнала:
2024,
Номер
73, С. 1 - 11
Опубликована: Янв. 1, 2024
Convolutional
neural
network
(CNN)-based
intelligent
fault
diagnosis
approaches
have
showcased
remarkable
performance
in
the
assessment
of
machine
safety.
The
data
monitored
from
mechanical
systems
industries
is
primarily
characterized
by
class
imbalance.
Nevertheless,
most
current
CNN-based
models
are
designed
under
assumption
balanced
sample
distributions,
which
do
not
align
with
prevalent
conditions
observed
real
industrial
scenarios.
To
tackle
this
challenge,
a
state-of-the-art
multiattention-based
feature
aggregation
convolutional
(MFACN)
developed
study.
key
contributions
study
outlined
as
follows:
1)
designs
an
attention-based
multiscale
module
(AMM)
and
(MFAM)
to
facilitate
comprehensive
learning
across
multiple
levels;
2)
robust
CNN
model
based
on
AMM
MFAM
established.
constructed
can
explore
abundant
information
signals;
3)
dual
focal
loss
(DFL)
function
introduced
enhance
diagnostic
results
assess
applicability
proposed
MFACN
health
state
identification,
two
experiments
were
conducted
using
bearing
dataset
planetary
gearbox
dataset.
experimental
unequivocally
show
that
surpasses
seven
other
approaches,
especially
when
dealing
imbalanced
datasets.
