Structural Health Monitoring,
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 10, 2024
With
the
rapid
development
of
railroads
and
yearly
increase
in
scale
operation,
safe
operation
maintenance
rail
trains
have
become
particularly
important.
Among
them,
deep
learning-based
bearing
fault
diagnosis
methods
attracted
more
attention
train
maintenance.
However,
usually
operate
normally.
Collecting
complete
data
for
learning
model
training
is
often
difficult.
Such
scenarios
with
a
large
difference
between
number
normal
affect
performance
models.
Here,
an
interactive
generative
feature
space
oversampling-based
autoencoder
(IGFSO-AE)
proposed
to
realize
sample
generation
under
imbalanced
data.
First,
original
vibration
signal
converted
into
semantically
stable
amplitude–frequency
by
fast
Fourier
transform
input
autoencoder;
second,
order
hidden
layer
features
randomly
exchanged,
strategy
then,
interpolation
oversampling
used
interpolate
samples
where
Euclidean
distance
large,
decoder,
generated
are
mixed
form
new
set,
which
intelligent
output
results.
Finally,
method
evaluated
using
publicly
available
dataset
bogie-bearing
simulation
bench
our
lab.
The
experimental
results
show
that
IGFSO-AE
can
generate
diverse
incremental
information
exhibits
robustness
superiority
different
proportions
Artificial Intelligence Review,
Journal Year:
2024,
Volume and Issue:
57(8)
Published: July 23, 2024
Abstract
Prognostics
and
health
management
(PHM)
is
critical
for
enhancing
equipment
reliability
reducing
maintenance
costs,
research
on
intelligent
PHM
has
made
significant
progress
driven
by
big
data
deep
learning
techniques
in
recent
years.
However,
complex
working
conditions
high-cost
collection
inherent
real-world
scenarios
pose
small-data
challenges
the
application
of
these
methods.
Given
urgent
need
data-efficient
academia
industry,
this
paper
aims
to
explore
fundamental
concepts,
ongoing
research,
future
trajectories
small
domain.
This
survey
first
elucidates
definition,
causes,
impacts
tasks,
then
analyzes
current
mainstream
approaches
solving
problems,
including
augmentation,
transfer
learning,
few-shot
techniques,
each
which
its
advantages
disadvantages.
In
addition,
summarizes
benchmark
datasets
experimental
paradigms
facilitate
fair
evaluations
diverse
methodologies
under
conditions.
Finally,
some
promising
directions
are
pointed
out
inspire
research.
Measurement Science and Technology,
Journal Year:
2024,
Volume and Issue:
35(4), P. 042003 - 042003
Published: Jan. 12, 2024
Abstract
Deep
learning
(DL)
has
attained
remarkable
achievements
in
diagnosing
faults
for
rotary
machineries.
Capitalizing
on
the
formidable
capacity
of
DL,
it
potential
to
automate
human
labor
and
augment
efficiency
fault
diagnosis
machinery.
These
advantages
have
engendered
escalating
interest
over
past
decade.
Although
recent
reviews
literature
encapsulated
utilization
DL
rotating
machinery,
they
no
longer
encompass
introduction
novel
methodologies
emerging
directions
as
continually
evolve.
Moreover,
practical
application,
issues
trajectories
perpetually
manifest,
demanding
a
comprehensive
exegesis.
To
rectify
this
lacuna,
article
amalgamates
current
research
trends
avant-garde
while
systematizing
anterior
techniques.
The
evolution
extant
status
machinery
were
delineated,
with
intent
providing
orientation
prospective
research.
Over
bygone
decade,
archetypal
theory
empowered
by
directly
establishing
nexus
between
mechanical
data
conditions.
In
years,
meta
methods
aimed
at
solving
small
sample
scenarios
large
model
transformers
mining
big
features
both
received
widespread
attention
development
field
equipment.
excellent
results
been
achieved
these
two
directions,
there
is
review
summary
yet,
so
necessary
update
Lastly,
predicated
survey
developmental
landscape,
challenges
orientations
are
presented.
