cal

powersy
stems.

Abst
ract
:

Thisproj
ectresear chai mst odevelopanef f
icientmet hodforpredicti
vemai ntenanceofpower
transf
ormersusi ngintelli
gentalgori
thmsappl iedt opowert r
ansformermoni toringdat a.Power
transf
ormersar ecr i
ti
cal component sofelectri
cal i
nfr
astructur
e,andt hei
rreli
abl eoper ati
onis
essenti
alforuninterruptedpowersuppl y.Howev er,f
actorssuchasov erl
oading,oi ll
oss, aging,
thermalstr
ess, andweat herconditi
onscandecr easetheirl
if
espan, l
eadingtopot ential f
ail
ures
andsignif
icanteconomi closses.Tradit
ionalmai ntenancemet hods,suchasr outinephy si
cal
tests,
areineffi
cientandcost l
y.Hence,thi
sr esearchpr oposestheut i
l
izati
onofadv anced
moni t
ori
ng,diagnosis, andpr ognosti
cmet hodst oenhancet ransformerreli
abili
tyand
perfor
mance.

Themet hodologyi nvolvest womai nalgor

ithms:Unsuperv isedMachineLearning( UML)f or
cl
assi fi
cationandcl usteringoft r
ansformerdata,andSuper vi
sedMachineLear ning( SML)using
Arti
ficialNeuralNet works( ANN)t oassesst heexploi
tat
ionageoft r
ansformersbasedon
hi
stor i
cal monitori
ngdat a.TheUMLal gor
it
hm classif
iestransformersintogroupswi thsimil
ar
proper ti
esandpr obabiliti
esoff ai
lur
e, whi
letheSMLal gor i
thm predi
ctstheremai ningusef ull
if
e
oftransf or
mer s,enablingi nfor
meddeci si
on-makingregar dingmaintenanceactions.
Additionall
y ,
theintegrationofonl i
nemoni tori
ngsystemsandadapt ivecontr
olst rategiesfur
ther
enhancest hereli
abili
tyandef f
ici
encyoftransfor
mermai ntenance.

Backgr
ound:

Electrici
typlaysapi votalroleinpower ingvar i
ousor gani
zati
onsanddomest i
cpurposes,making
powert ransformersessent i
al f
orthef unctioningofelectri
calpowersy stems.Howev er,t
he
reli
abilit
yoft hesetransformer sisoftencompr omisedduet ov ari
ousf actor
sleadingto
decr easedl i
fespanandpot enti
al f
ail
ures.Theconsequencesoft r
ansfor merfai
luresi
nclude
powerdi srupti
ons,result
ingi nmonet arylossesf orutil
it
yanddi st
ri
butioncompani es,aswellas
negat iveimpact sondomest icelectr
icityconsumer s.Toaddresst hesechallenges,pr
edicti
ve
mai ntenancet echniqueshav ebecomei ncreasingl
yimpor t
antinmi t
igatingtheri
skof
unexpect edfail
uresandopt i
mizingmai nt
enanceschedul es.
Thel it
erat
urer evi
ewhi ghli
ghtst hecur r
entstateofpr edi ctiv
emai nt
enanceandcont rol
techniquesf orpowertransf or
mer s.Var i
oust echniquessuchasCondi t
ionMonitoring( CM),
DissolvedGasAnal ysi
s( DGA),Par t
ialDischargeMoni toring,andonli
nemoni t
oringsy stems
hav ebeenempl oyedforpr edi
ctivemai ntenanceoft ransf ormers.Addit
ional
ly
,predi ct
ivecont
rol
strategi
es,includi
ngPr ognosticHeal thManagement( PHM)andAdapt i
v eContr
ol ,
ai mto
opt i
mizetransf or
merper formanceandl ongev i
ty.Sev eral st
udieshavedemonst ratedthe
effecti
venessofmachi nelearningal gori
thms, part
icularlyANN, i
npredi
ctingtr
ansf ormerhealt
h
andr emainingusef ull
ifebasedonmoni tori
ngdat a.

Drawingupont heinsi
ghtsfrom t helit
eratur
erevi
ew, thi
sprojectr
esearchproposesa
compr ehensivemet hodol
ogyf orpr edict
ivemaintenanceofpowert r
ansformers.Bylev er
aging
i
ntel
li
gental gorit
hmsandadv ancedmoni tori
ngtechniques,t
hegoal i
stoshiftfr
om traditi
onal
ti
me-basedmai ntenancepracticest oconditi
on-
basedandpr edi
cti
veapproaches,t
her eby
enhancingther eli
abil
it
yandef ficiencyofelect
ri
cal powersystems.

Obj
ect
ives:

Dev
elopacost
-ef
fect
ivepr
edi
cti
vemai
ntenanceandcont
rol
syst
em f
orpowert
ransf
ormer
s.

I
mplementi
ntel
l
igental
gor
it
hmsf
orpr
edi
cti
veanal
ysi
soft
ransf
ormerheal
thandr
emai
ning
usef
ull
i
fe.

Int
egrat
esensor
sformoni
tor
ingkeypar
amet
erssuchast
emper
atur
e,oi
ll
evel
,vol
tage,
and
curr
ent.

Enabl
eonli
necommunicati
onusingaGSM modul
eforr
eal
-t
imedat
atr
ansmi
ssi
onandof
fl
ine
readi
ngoft
ransf
ormeroperat
ional
par
ameter
s.

Provi
deti
mel
ynot
if
icat
ionst
omai
ntenanceper
sonnel
incaseofabnor
mal
i
tiesori
mpendi
ng
faul
ts.

Enhancet
her
eli
abi
l
ityandef
fi
ciencyoft
ransf
ormermai
ntenancecompar
edt
otr
adi
ti
onal
methods.

Compr
ehensi
veMet
hodol
ogy
:

Har
dwar
eSel
ect
ionandI
ntegr
ati
on:
ChooseAr
dui
noUnoast
hemi
crocont
rol
l
erboar
dfori
tsaf
for
dabi
l
ityandv
ersat
il
it
y.

Sel
ectappr
opr
iat
esensor
sfort
emper
atur
e,oi
ll
evel
,vol
tage,
andcur
rentmeasur
ement
.

I
ntegr
atesensor
swi
thAr
dui
noUnof
ordat
aacqui
sit
ion.

I
ncor
por
ateaGSM modul
eforonl
i
necommuni
cat
ionandnot
if
icat
ions.

Ensur
ecompat
ibi
l
ityandr
eli
abi
l
ityofhar
dwar
ecomponent
s.

Dat
aAcqui
sit
ionandPr
epr
ocessi
ng:

Dev
elopf
ir
mwar
eforAr
dui
noUnot
oacqui
redat
afr
om sensor
s.

Pr
epr
ocessacqui
reddat
ator
emov
enoi
seandout
li
ers.

Nor
mal
i
zedat
aforuni
for
mit
yandcompat
ibi
l
itywi
thmachi
nel
ear
ningal
gor
it
hms.

Machi
neLear
ningAl
gor
it
hms:

I
mplementUnsuper
visedMachi
neLear
ning(
UML)al
gor
it
hm f
orcl
assi
fi
cat
ionandcl
ust
eri
ngof
t
ransf
ormerdat
a.

UseK-
meanscl
ust
eri
ngt
ogr
oupt
ransf
ormer
swi
thsi
mil
arpr
oper
ti
esandpr
obabi
l
iti
esoff
ail
ure.

DevelopSuper
visedMachi
neLear
ning(
SML)al
gor
it
hm usi
ngAr
ti
fi
cial
Neur
alNet
wor
ks(
ANN)
topredi
ctremai
ningusef
ull
if
e.

Tr
ainANNmodel
usi
nghi
stor
ical
moni
tor
ingdat
atoassessexpl
oit
ati
onageoft
ransf
ormer
s.

Onl
i
neandOf
fl
ineMoni
tor
ing:

Enabl
ereal
-t
imedat
atr
ansmi
ssi
onusi
ngGSM modul
eforonl
i
nemoni
tor
ing.

Dev
elopauseri
nter
facef
oronl
i
neaccesst
otr
ansf
ormeroper
ati
onal
par
amet
ers.

I
mpl
ementof
fl
iner
eadi
ngoft
ransf
ormerpar
amet
ersf
orl
ocal
moni
tor
ingandanal
ysi
s.

Faul
tDet
ect
ionandAbnor
mal
i
tyNot
if
icat
ion:

Dev
elopal
gor
it
hmst
odet
ecti
mpendi
ngf
aul
tsbasedonmoni
tor
edpar
amet
ers.
Sett
hreshol
dsf
orabnor
mal
condi
ti
onsandt
ri
ggernot
if
icat
ionst
omai
ntenanceper
sonnel
.

I
ntegr
ateGSM modul
eforsendi
ngSMSal
ert
sincaseofabnor
mal
i
tiesori
mpendi
ngf
aul
ts.

Test
ingandVal
i
dat
ion:

Conductex
tensi
vet
est
ingoft
hepr
edi
cti
vemai
ntenancesy
stem undersi
mul
atedandr
eal
-wor
ld
condi
ti
ons.

Eval
uatet
heaccuracyandrel
i
abi
l
ityofpr
edi
cti
veal
gor
it
hmsi
npr
edi
cti
ngt
ransf
ormerheal
th
andremai
ninguseful
li
fe.

Val
i
dat
etheef
fect
ivenessoff
aul
tdet
ect
ionandabnor
mal
i
tynot
if
icat
ionmechani
sms.

Per
for
manceEv
aluat
ion:

Measur
etheper
for
manceoft
hesy
stem i
nter
msofr
eli
abi
l
ity
,ef
fi
ciency
,andcost
-ef
fect
iveness.

Compar et
hepredi
cti
vemaint
enancesy
stem wi
tht
radi
ti
onal
met
hodsi
nter
msofdownt
ime
reduct
ionandmaint
enancecost
s.

Anal
yzet
heeconomi
cimpactoft
hesy
stem onut
il
it
yanddi
str
ibut
ioncompani
es.

Opt
imi
zat
ionandDepl
oyment
:

I
ter
ateont
hedesi
gnbasedont
estr
esul
tsandf
eedbackt
oopt
imi
zesy
stem per
for
mance.

Depl
oyt
hef
inal
i
zedpr
edi
cti
vemai
ntenancesy
stem i
nreal
-wor
ldt
ransf
ormerenv
ironment
s.

Pr
ovi
det
rai
ningandsuppor
tformai
ntenanceper
sonnel
foref
fect
iveut
il
izat
ionoft
hesy
stem.

Byf ollowingthiscompr ehensi

v emethodology,t
hedev el
opedpredi
ctivemaint
enanceand
contr olsystem forpowert ransformerswi l
lbeabl
et oaccur
atel
ypredictr
emainingusefull
if
e,
detectabnor malit
ies,andnot i
fymaintenancepersonnelinati
melymanner ,t
herebyenhanci
ng
ther eli
abili
tyandef fi
ciencyoft r
ansfor
mermai ntenancewhil
eensuringaffor
dabil
it
yand
eff
ect i
venesscompar edt oexisti
ngmet hods.