Proceedings of the Institution of Mechanical Engineers Part C Journal of Mechanical Engineering Science,
Journal Year:
2025,
Volume and Issue:
unknown
Published: April 24, 2025
Existing
intelligent
fault
diagnosis
methods
commonly
rely
on
extensive
and
balanced
dataset.
However,
the
collection
of
samples
significantly
lags
behind
that
normal
in
practical
industrial
scenarios,
resulting
data
imbalance
problem
severely
impacts
diagnostic
accuracy.
To
address
this
challenge,
paper
presents
a
novel
approach,
termed
Hybrid
Distance
Generative
Adversarial
Network
with
gradient
penalty
(HDGAN-GP).
Initially,
stacked
autoencoders
(SAE)
is
incorporated
into
original
generator
to
form
an
auxiliary
generator,
facilitating
production
high-quality
samples.
Subsequently,
loss
function
devised
based
hybrid
distance
metric
comprising
cosine
similarity
maximum
mean
difference,
supplemented
by
term
ensure
stable
model
training.
Finally,
experimental
validation
conducted
using
gear
Comparative
analysis
existing
generative
adversarial
network
models
demonstrates
proposed
method
generates
superior
quality
samples,
effectively
addressing
challenge
posed
diagnosis.
