International journal of mechanical system dynamics,
Год журнала:
2024,
Номер
unknown
Опубликована: Ноя. 5, 2024
Abstract
To
address
the
difficulty
in
extracting
early
fault
feature
signals
of
rolling
bearings,
this
paper
proposes
a
novel
weak
diagnosis
method
for
bearings.
This
combines
Improved
Complementary
Ensemble
Empirical
Mode
Decomposition
with
Adaptive
Noise
(ICEEMDAN)
and
Maximum
Correlated
Kurtosis
Deconvolution
(IMCKD).
Utilizing
kurtosis
criterion,
intrinsic
mode
functions
obtained
through
ICEEMDAN
are
reconstructed
denoised
using
IMCKD,
which
significantly
reduces
noise
measured
signal.
approach
maximizes
energy
amplitude
at
characteristic
frequency,
facilitating
identification.
Experimental
studies
on
two
test
benches
demonstrate
that
effectively
interference
highlights
frequency
components.
Compared
traditional
methods,
it
improves
signal‐to‐noise
ratio
more
accurately
identifies
features,
meeting
requirements
discriminating
bearing
faults.
The
proposed
study
was
applied
to
vibration
gearbox
bearings
new
high‐speed
wire
department
Long
Products
Mill.
It
successfully
extracted
information
faults,
achieving
expected
diagnostic
results.
further
validates
effectiveness
ICEEMDAN–IMCKD
practical
engineering
applications,
demonstrating
significant
value
detecting
impact
characteristics
Symmetry,
Год журнала:
2025,
Номер
17(3), С. 427 - 427
Опубликована: Март 12, 2025
For
mechanical
equipment
to
operate
normally,
rolling
bearings—which
are
crucial
parts
of
rotating
machinery—need
have
their
faults
diagnosed.
This
work
introduces
a
bearing
defect
diagnosis
technique
that
incorporates
three-channel
feature
fusion
and
is
based
on
enhanced
Residual
Networks
Bidirectional
long-
short-term
memory
networks
(ResNet-BiLSTM)
model.
The
can
effectively
establish
spatial-temporal
relationships
better
capture
complex
features
in
data
by
combining
the
powerful
spatial
extraction
capability
ResNet
bidirectional
temporal
modeling
BiLSTM.
Specifically,
one-dimensional
vibration
signals
first
transformed
into
two-dimensional
images
using
Continuous
Wavelet
Transform
(CWT)
Markov
Transition
Field
(MTF).
upgraded
ResNet-BiLSTM
network
then
used
extract
combine
original
signal
along
with
from
two
types
images.
Finally,
experimental
validation
performed
datasets.
results
show
compared
other
state-of-the-art
models,
computing
cost
greatly
reduced,
params
flops
at
15.4
MB
715.24
MB,
respectively,
running
time
single
batch
becomes
5.19
s.
fault
accuracy
reaches
99.53%
99.28%
for
datasets,
successfully
realizing
classification
task.
Journal of Vibration and Control,
Год журнала:
2025,
Номер
unknown
Опубликована: Янв. 21, 2025
In
engineering
practice,
the
diagnosis
of
rotating
machinery
faults
often
faces
numerous
challenges,
including
noise
interference
and
changes
in
operating
conditions,
which
pose
new
difficulties
for
deep
learning
methods
lacking
prior
knowledge.
response
to
this
issue,
paper
proposes
a
fault
method
(OCML)
based
on
octave,
convolutional
neural
networks,
MOGRIFIER
LSTM.
This
can
simply
effectively
achieve
reduction,
feature
extraction,
classification.
Firstly,
through
octave
analysis,
redundant
information
be
conveniently
filtered
out,
enhancing
signal
representation.
Secondly,
designed
CNN-MOGRIFIER
LSTM
model
capture
local
features
temporal
dependencies
data
has
good
interaction
capabilities.
Experiments
CWRU
dataset
permanent
magnet
synchronous
motor
demonstrate
that
proposed
exhibits
diagnostic
performance
across
different
scenarios
conditions.
Furthermore,
compared
other
methods,
OCML
performs
better
terms
accuracy
stability.
These
results
collectively
confirm
generalization
method.
