Candidate:
Farook
Mustafa
Title:
Hydrophobicity
Classification
of
RTV
Silicone
Rubber-Coated
Insulators
Using
Deep
Learning
Algorithms
Date:
August
3,
2022
Time:
13:00
Place:
online
Supervisor(s):
El-Hag,
Ayman
Abstract:
Silicone
rubber-based
outdoor
polymeric
insulators
are
widely
employed
in
electric
power
transmission
and
distribution
networks
to
replace
the
conventional
ceramic
insulators,
owing
to
their
superior
performance
in
contaminated
and
wet
environments.
Silicone
rubber
(SIR)
insulators
offer
several
advantages
like
high
hydrophobicity,
low
cost,
vandalism
resistance
and
light
weight.
However,
when
exposed
to
electrical
(dry
band
arcing
and
partial
discharge)
and
environmental
stresses
(humidity,
ultraviolet
radiation,
acid
rain
and
pollution)
they
suffer
from
different
forms
of
aging.
The
first
form
of
aging
is
the
temporal
loss
of
hydrophobicity.
However,
SIR
insulators
can
recover
the
hydrophobicity
property
due
to
the
diffusion
of
the
low
molecular
weight
(LMW)
from
the
bulk
of
the
insulating
material
to
the
insulators’
surface.
Hence,
it
is
important
to
classify
the
hydrophobicity
status
of
SIR
insulators
as
an
indication
of
the
aging
degree.
Different
methods
have
been
implemented
to
classify
the
hydrophobicity
of
the
insulator
surface
including
static
contact
angle
measurement,
dynamic
contact
angle
measurement
and
hydrophobicity
class
(HC).
The
later
technique
is
the
most
practical
method
that
can
be
used
in
the
field
and
can
assess
wide
surface
area.
The
surface
wetting
tendency
is
manually
classified
using
one
of
six
classes,
i.e.
HC1-HC6,
where
HC1
refers
to
a
completely
hydrophobic
surface
and
HC6
is
a
completely
hydrophilic
surface.
The
main
objective
of
this
thesis
is
to
automatically
assess
the
hydrophobicity
classes
of
non-ceramic
insulators
under
a
variety
of
conditions
using
deep
learning
techniques.
A
dataset
of
hydrophobicity
classes
(HC1-HC6)
was
created
and
prepared
including
4197
images
each
having
2242×24
pixels
size
to
train
the
proposed
model.
Several
deep
learning
techniques,
including
Convolutional
Neural
Networks
(CNN),
Transfer
Learning
(TL),
and
Object
Detection
(OD),
were
used
in
this
thesis
to
categorize
and
assess
the
hydrophobicity
classes
of
ceramic
insulator
coated
with
room
temperature
vulcanized
silicone
rubber
(RTV-SIR).
MobileNet
model
was
found
to
have
the
highest
accuracy
and
less
training
time
after
comparing
with
other
CNN
pretrained
models.
This
model
was
then
trained
and
tested
under
several
conditions,
including
indoor,
bright,
and
dark
lighting
conditions,
and
achieved
accuracy
of
97.77%,
89.44%,
and
95%,
respectively.
Moreover,
the
proposed
model
achieved
a
recognition
rate
of
96.11%
when
tested
on
a
full-scale
silicone
rubber
insulator.
The
developed
model
was
then
deployed
as
a
web
application
for
convenience
in
the
assessment
of
hydrophobicity
classes.
The
proposed
model
could
be
utilized
to
evaluate
SIR
insulators
surface
conditions
in
an
effective
and
automatic
way
under
different
conditions.
Wednesday, August 3, 2022 1:00 pm
-
1:00 pm
EDT (GMT -04:00)