Engineering Research Express,
Год журнала:
2024,
Номер
7(1), С. 015401 - 015401
Опубликована: Дек. 20, 2024
Abstract
This
study
proposes
a
machine
learning-based
fault
diagnosis
method
for
hydraulic
systems,
focusing
on
using
K-means
clustering
algorithm
data
preprocessing,
use
Support
Vector
Machine
(SVM)
and
Convolutional
Neural
Network
Gated
Recurrent
Unit
(CNN-GRU)
models
respectively
classification.
By
performing
cluster
analysis
sensor
data,
the
dimensions
can
be
effectively
reduced
efficiency
of
improved.
The
results
show
that
accuracy
SVM
in
cooler
status
valve
classification
tasks
reached
99.77%
100.00%
respectively.
After
introduction
algorithm,
its
training
time
was
significantly
reduced,
showing
extremely
high
real-time
capabilities.
CNN-GRU
model
performs
particularly
well
handling
complex
tasks,
especially
accumulator
task,
with
an
rate
as
96.60%,
which
is
better
than
model.
Although
longer,
advantages
pattern
recognition
give
it
obvious
high-accuracy
application
scenarios.
Multi-fault
conducted,
achieves
best
performance
without
employing
clustering,
emphasizing
importance
preserving
integrity
original
Centrifugal
pumps
are
essential
in
various
industrial
applications,
and
their
stable
efficient
operation
has
a
direct
impact
on
the
overall
performance
of
system.
This
study
simulates
different
lengths
fractures
at
LE
(leading
edge)
single
blade
to
conduct
an
in-depth
analysis
effects
internal
flow
transient
characteristics.
The
reveals
that
most
significant
pump
occur
near
rate
0.8Qd,
where
head
efficiency
can
decrease
by
up
6.19%
3.77%,
respectively,
compared
original
blades.
Blade
lead
deterioration
pressure
suction
sides,
creating
vortices
inducing
leakage
flow,
while
entropy
production
significantly
increases
this
area.
A
230.1%
increase
distribution
angle
26.6%
maximum
radial
force
reflect
changes
distribution.
Also,
make
wall
pulsations
stronger
SPF
(shaft
passing
frequency),
they
amplitude
surface
much
bigger
both
3SPF
frequencies.
Finally,
change
way
vibrations
behave
measurement
points
x
y
directions
big
way.
acceleration
amplitudes
frequencies
go
125.8%,
193.1%,
62.5%,
184.6%,
respectively.
These
findings
provide
important
theoretical
basis
for
early
warning
diagnosis
fracture
failures.
Applied System Innovation,
Год журнала:
2024,
Номер
7(4), С. 61 - 61
Опубликована: Июль 19, 2024
Essential
service
water
pumps
are
necessary
safety
devices
responsible
for
discharging
waste
heat
from
containments
through
seawater;
their
condition
monitoring
is
critical
the
safe
and
stable
operation
of
seaside
nuclear
power
plants.
However,
it
difficult
to
directly
apply
existing
intelligent
methods
these
pumps.
Therefore,
an
framework
designed,
including
parallel
implementation
unsupervised
anomaly
detection
fault
diagnosis.
A
model
preselection
algorithm
based
on
highest
validation
accuracy
proposed
diagnosis
selection
among
models.
novel
information
integration
fuse
output
According
experimental
results
modules,
a
kernel
principal
component
analysis
using
mean
fusion
processing
multi-channel
data
(AKPCA
(fusion))
selected,
support
vector
machine
(SVM
selected.
The
overall
test
false
negative
rate
AKPCA
(fusion)
0.83
0.144,
respectively,
f1-score
SVM
0.966
1,
respectively.
(fusion),
show
that
successfully
avoids
lack
abnormal
status
misdiagnosis.
meaningful
attempt
achieve
complex
equipment.
Measurement Science and Technology,
Год журнала:
2024,
Номер
36(1), С. 016023 - 016023
Опубликована: Окт. 30, 2024
Abstract
Deep
learning
has
been
extensively
applied
in
the
rolling
bearing
fault
diagnosis
domain
due
to
its
superior
data
analysis
and
feature
extraction
capabilities.
However,
practical
applications,
normal
operating
state
occupies
most
of
service
life
equipment,
occurrence
probability
each
kind
is
different,
leading
imbalanced
that
significantly
degrades
performance
neural
network.
In
order
solve
this
problem,
a
dual-feature
enhanced
hybrid
convolutional
network
(DEHCNet)
proposed.
Firstly,
an
impulse
segment
enhancement
module
constructed
enhance
features
raw
data,
helping
learn
more
accurately.
Then,
designed
fully
mine
discriminant
minority
classes
from
data.
addition,
feature-enhanced
combinational
pooling
devised
guide
focus
on
critical
maximize
retention
key
dimensionality
reduction
operations,
thereby
reducing
influence
imbalance
classifier.
Finally,
three
distinct
datasets
are
used
verify
DEHCNet.
Experimental
results
show
better
diagnostic
accuracy
robustness
under
conditions
imbalance.
Machines,
Год журнала:
2024,
Номер
12(12), С. 869 - 869
Опубликована: Ноя. 29, 2024
To
fully
exploit
the
rich
state
and
fault
information
embedded
in
acoustic
signals
of
a
hydraulic
plunger
pump,
this
paper
proposes
an
intelligent
diagnostic
method
based
on
sound
signal
analysis.
First,
were
collected
under
normal
various
conditions.
Then,
four
distinct
features—Mel
Frequency
Cepstral
Coefficients
(MFCCs),
Inverse
Mel
(IMFCCs),
Gammatone
(GFCCs),
Linear
Prediction
(LPCCs)—were
extracted
integrated
into
novel
hybrid
cepstral
feature
called
MIGLCCs.
This
fusion
enhances
model’s
ability
to
distinguish
both
high-
low-frequency
characteristics,
resist
noise
interference,
capture
resonance
peaks,
achieving
complementary
advantage.
Finally,
MIGLCC
set
was
input
double
layer
long
short-term
memory
(DLSTM)
network
enable
recognition
pump’s
operational
states.
The
results
indicate
that
MIGLCC-DLSTM
achieved
accuracy
99.41%
test
Validation
CWRU
bearing
dataset
data
from
high-pressure
servo
motor
turbine
system
yielded
overall
accuracies
99.64%
98.07%,
respectively,
demonstrating
robustness
broad
application
potential
method.
